Home

Nutrient Moored Sensor Program Year 1 Progress Update

1. Apply and then Complete A 16 Era Servicing trips Retrieve sonde and download data If you are servicing any bridge site the first thing to do when arriving is to contact the U S Coast Guard and California Highway Patrol U S Coast Guard 415 399 3451 California Highway Patrol 510 286 6920 Inform them of your affiliation your purpose for being on site the kind of craft you are in and how long you anticipate being on site Retrieve sondes For Alviso Slough all servicing occurs on the boat o The instrument package is pulled from the bottom using an electric winch first pull up the weights then the package For bridge sites servicing occurs from the bridge platform o Unload all needed equipment from the boat using a haul line if tide is low o Using the deployment line not the communication cable pull the instruments up out of the water to within comfortable reach removing the black clips as you go and secure with the extra line on the davet o If there has been a significant amount of fouling the instrument carriage may be heavy and it would be helpful to scrape off the fouling as it pulled up o Remove the bolt at the top of the carriage remove the redundant line and communications cable and extract the EXO2 from the carriage Download data Start KOR EXO Connect the EXO2 to the computer using the USB adaptor KOR EXO can be picky about starting the software before connecting Na
2. 10760 San Pablo Ave El Cerrito CA 94530 510 525 3508 A 30 5 3 Ea Safety information We typically launch from San Leandro Marina Here are the nearest hospitals to that location Figure A 3 Hospitals near San Leandro Marina Kaiser Permanente Post Acute 3 6 k k 9 Google reviews San Leandro Hospital www alamedahealthsystem com 4 Google reviews Kindred Hospital San Francisco Bay www kindredhospitalsfoa com 2 Google reviews Google page John George Psychiatric Pavilion www johngeorgeahs org 1 Google review 1440 168th Ave San Leandro CA 510 481 8575 13855 E 14th St San Leandro CA 510 357 6500 2800 Benedict Dr San Leandro CA 510 357 8300 2060 Fairmont Dr San Lorenzo CA 510 346 1300 All Saints Subacute amp Rehabilitation C www allsaintscare com 1 Google review Google page Concentra Urgent Care San Leandro www concentra com 1 Google review Bancroft Convalescent Hospital www bancroftconvalescenthospital com Google page A 31 1652 Mono Ave San Leandro CA 510 481 3200 2587 Merced St San Leandro CA 510 351 3553 1475 Bancroft Ave San Leandro CA 510 483 1680 Ei References Wagner R J Boulger R W Jr Oblinger C J and Smith B A 2006 Guidelines and standard procedures for continuous water quality monitors Station operation record computation and data reporting U S Geological Survey Techniques and Methods 1 D3 51 p
3. a All in April 2014 removed from 10000 regression 15 20 0 10000 20000 30000 40000 YSI 6920 Temp C YSI 6920 SpC uS cm 4500 y 1 22x 7 81 r 0 89 150 y 1 08x 1 69 r 0 94 So Ss 1000 EXO2 Turb FNU EXO2 DO saturation a e a So 100 0 50 75 YSI 6920 DO saturation 800 0 200 400 600 YSI 6920 Turb FNU Figure 3 1 Comparison of SFEI YSI EXO2 and USGS SacSed YSI 6920 data for common parameters during co deployment in Alviso Slough Sept 2013 May 2014 when EXO2 T SpC probe began a prolonged malfunction 1 1 line is shown in red USGS data was accessed through http waterdata usgs gov nwis and should be considered provisional 3 1 2 Comparison Bay wide During Monthly Cruises We also conducted side by side comparisons of the EXO2 with a Turner 10 AU chl a fluorometer and nephlometer measuring optical backscatter on 13 monitoring cruises aboard the USGS Menlo Park research vessel R V Polaris from September 2013 through June 2014 The instruments were plumbed into the ship s flow through system that pumps surface Bay water sample depth approximately 1 m into the laboratory and measurements were made continuously as the R V Polaris conducted its 150 km survey along the Bay s axis The Turner 10 AU chl a fluorometer and nephlometer have been used aboard the R V Polaris for the past 9 years The raw instrument output between both chl a fluorescence and the turbidity probes we
4. 1 of air DO inputs and barometric pressure to calculate DO saturation concentration in mg L 5 above 200 Turbidity The turbidity probe detects light scattering by suspended 0 4000 NTU the greater of 0 3 FNU or particles at 90 of incident light beam The turbidity probe 2 of reading defaults to formazin nephelometric units FNU but can 5 of reading above also report raw signal nephelometric turbidity units NTU 1000 FNU or total suspended solids if the correct correlation factors are provided range Fluorescent fDOM is measured by a fluorescent probe that excites at 0 300 QSU not specified by dissolved 365 5 nm and measures emission at 480 40 nm We manufacturer organic report in relative fluorescent units RFU but the probe can matter also report raw sensor output or concentration quinine The pH probe measures the differential across a glass 0 14 pH 0 1 units when within surface the inside of which has a stable pH solution and units 10 C of calibration the outside of which is in contact with the environment temperature 0 02 units across entire fDOM sulfate units QSU 1ppb quinine sulfate if the correct correlation factors are provided Chlorophyll Both of these algal pigments are detected by a single dual 0 100 RFU not specified by a chl a channel fluorescent probe Chl a excites at 470415 nm manufacturer and and BGA PC excites at 590 15 nm Emission of both is phycocyanin measured
5. 3 Battery V 10 Chl RFU 17 ODO sat 4 Depth meters 11 Chl ug L 18 BGA PC RFU 5 Temp C 12 Chlorophyll Raw 19 BGA PC ug L 6 SpCond us cm 13 fDOM RFU 20 BGA PC Raw 7 pH 14 DOM QSU e Under the Advanced tab set the Logging Mode to normal the Averaging Duration to 10 seconds and the Samples Per Wipe to 1 Make sure Adaptive Logging is not enabled e Begin deployment by clicking on this icon 3 A 21 e Wait until the next 15 minute interval to confirm the program is running watch for the wiper to move and fluorescent sensor to illuminate Redpeloying the sonde e Put the sonde back in carriage attach the communications cable and redundant line reinstall the bolt making sure it goes through the bail and lower the carriage gently back into the water reattching the clips as you go e If you have access to internet check the IP address for Dumbarton http 166 140 153 235 to confirm telemetry is working timestamps are in PST A 22 Discrete sample collection In order to accurately infer chl a concentrations from probe fluorescence measurements it is important to take a discrete chl sample during each sampling trip that will be used to built a robust chl fl relationship When sampling at Alviso Slough without a power inverter filtering cannot occur on site You should put the water sample on wet ice until you have a chance to filter either at a bridge site or back at the dock ideally withi
6. 500 0 l aE _ E I I I I l I 0 5 10 15 20 25 depth Figure 2 Histogram of depths with extracted chlorophyll measurements Jan 2005 Jul 2013 3 Measurement uncertainty Uncertainty of the analytical method for chlorophyll a is the simplest to characterize because of the large number of replicated measurements made over the years For the time period of interest here 2005 to 2013 a total of 3564 replicated surface and bottom samples were measured Most were duplicates with a few triplicate and quadruplicate measurements as well The coefficient of variation CV for chlorophyll a measurements ranged from 0 to 0 63 but the median was only 0 018 mean 0 024 More than 90 of the CV values were less than 0 05 which is the recommended guideline for the method Figure 3 The standard error SE of the measurements ranged from 0 to 3 4 pgl but again the median was only 0 050 pgl mean 0 097 Most importantly these SE values were almost always a small fraction of the corresponding means even for the smallest measurements Figure 4 These measurement uncertainties are relatively unimportant compared to those arising from other sources of uncertainty see Section 5 2 5NEMI method number 445 0 at https www nemi gov home 1 00 0 75 empirical CDF mn oO I 0 25 0 00 0 001 0 010 0 100 chlorophyll a CV Figure 3 Empirical cumulative distribution function for chlorophyll a coef
7. before and after cleaning in two identical buckets of water Detailed procedures are as follows e Fill the 2 5 gallon buckets with identical water It may take several grabs to get enough water Mix water from all grabs together and then subset into the two buckets e While it is best to leave the biofouling as intact as possible it may be a good idea to rinse surface dirt sediment off the EXO2 to prevent it from mixing into the bucket and changing the water composition e Put the EXO2 into one bucket Try to keep the two buckets in similar environments Navigate to the Dashboard menu Take a complete set of probe readings and record on the field sheet Be sure to note time to that any variability in before and after checks due to changes in the water in the buckets i e reaeration can be backcalculated after the fact e Wipe the probes one time using the button on the Dashboard screen and take another set of readings in the w1 column If the values did not change significantly begin cleaning instrument If they did change significantly continue wiping until the stabilize and record readings in w2 w3 etc columns The goal is to assess how much variability between 15 minute readings could be due to the effect of the wiper possibly pushing something in front of the probes e Clean the instrument thoroughly including between probes and port plugs Q tips are good for port plugs If necessary add Krytox grease Be
8. raw to provisional data at each NSMP site 30 1 Introduction San Francisco Bay has long been recognized as a nutrient enriched estuary Cloern and Jassby 2012 but one that has exhibited resistance to some of the classic symptoms of nutrient overenrichment such as high phytoplankton biomass and low dissolved oxygen Recent observations however indicate that the Bay s resistance to high nutrient loads is weakening leading regulators and stakeholders to collaboratively develop the San Francisco Bay Nutrient Strategy SFBRWQCB 2012 The Nutrient Strategy calls for a research and monitoring program to address priority science questions and fill key data gaps to inform nutrient management decisions in San Francisco Bay Among its recommendations the Nutrient Strategy calls for developing models to quantitatively characterize the Bay s response to nutrient loads explore ecosystem response under future environmental conditions and test the effectiveness of load reduction scenarios and other scenarios that mitigate or prevent impairment As an early step in the Nutrient Strategy implementation a team of regional and national experts identified major science questions and specific data needs related to nutrients SFEI 2014 731 and among the high priorities was the collection of high frequency water quality data through moored sensors at key locations throughout the Bay In 2013 the San Francisco Bay Regional Monitoring Program initiate
9. Dumbarton Bridge through USGS SacSeds existing equipment and we are currently considering expansion to other NMSP sites San Mateo Bridge other expansion sites in Year 2 and beyond There are at least two major benefits of having real time data access First access to real time data would permit immediate notification that sensors have failed or that fouling is beginning to occur Assuming that these observations would then guide field maintenance schedules real time data access would help minimize instrument downtime and lost data In this case the cost of lost data needs to be compared with the cost of implementing real time data access Real time data could also identify low biofouling periods and conceivably allow for some maintenance trips to be postponed leading to program cost savings A second benefit of real time data access is that events such as a major phytoplankton blooms or low DO periods detected by sensors could initiate event based sampling and further study of conditions through discrete sample collection and analysis Moored sensors only detect conditions at fixed locations and for a limited set of parameters while complementary boat ship based sampling triggered by mooring observations would allow for information to be collected over a larger spatial area and a broader set of parameters e g phytoplankton community composition and toxin samples A coupled approach like this could contribute substantially to improved u
10. Suisun measuring chl a DO or nutrients there are major gaps in data collection for nutrient related parameters The NMSP aims to build capacity to deploy and maintain sensors as well as manage and interpret data such that the NMSP may augment these existing efforts where needed or to the extent possible play the coordinating role across these existing programs to maximize the utility of the data collected by other programs to help achieve NMSP s goals Figure 2 1 Locations of existing moored sensors in SFB by program Analytes monitored differ by program as described in Table 2 1 Both USGS programs are located at the California Water Research Center in Sacramento Table 2 1 Characteristics of current moored sensor monitoring in SFB by program Most moored sensor monitoring for nutrient related parameters has been limited to areas of north of the Bay Bridge which is why SFEI focused on South Bay and Lower South Bay in Phase 1 of the NMSP Period of Locations Analytes monitored record USGS Sacramento 1989 present 10 active SFB Sediment group USGS stations in SFB SacSed 6 active stations in the Delta USGS Sacramento 2011 present 6 active stations North SFB Delta in the Delta biogeochemistry group DWR EMP 1978 present 9 active stations in North SFB and the Delta SFSU RTC 2002 present 2 active stations SFB Environmental Assessment in North SFB and Monitoring Station NERR 2006 present 4 active stations
11. at 685420 nm We report in relative fluorescent blue green units RFUs but the probe can also report raw probe algae BGA output or concentration ug L if the correct correlation PC factors are provided A 5 Pe Site descriptions In this first year SFEI deployed sondes at 2 sites for the duration of the pilot program year the Dumbarton Bridge and Alviso Slough and briefly deployed a sonde at the San Mateo Bridge as well These sites were chosen because USGS Sacramento also has instrumentation at these sites and we could partner with them on servicing and maintenance trips since a boat is necessary to access all three sites and SFEI neither maintains a boat nor has a licensed operator on the project at this time These may not represent most meaningful locations for monitoring nutrient related impairment but the data gathered in this first year has still been valuable in our understanding of the system Future site placement will need to balance the optimal sensor placement for monitoring with feasibility of site access particularly if SFEI needs to quickly access a site either for maintenance or event based sampling The instrument at Alviso Slough is deployed on a metal cage that is approximately 1 4 meters underwater at all times depending on tide stage The instrument at Dumbarton Bridge are deployed via a suspension cable mounted to a davit on the footing of the bridge These instruments are also approximately 1 4 meters
12. community composition Otherwise it would be preferable to use the actual causal factor For example if the subregions differ because of SPM then there should be a single Bay wide calibration with SPM as a secondary predictor rather than subregions Choosing subregions and a secondary predictor is asking too much of the data 6 Dealing with low chlorophyll The improvement in prediction error from adding a secondary predictor is usually less than 1 pg lt chlorophyll a Figure 8 and therefore unlikely to change the qualitative conclusions from past data analyses except perhaps those involving low winter values of chlorophyll a In some situations there is an alternative way to reduce prediction error at these low val ues When the constant variance assumption of ordinary least squares regression is violated weighted least squares should be used instead In this situation large deviations present at high fluorescence values affect the regression line more than the smaller deviations of low fluorescence values causing a higher relative error for the low values Weighted regression counteracts this tendency see Almeida et al 2002 regarding calibration curves and Helsel and Hirsch 2002 for a general presentation How do we choose the weighting factors Population abundance in general and phyto plankton abundance in particular often appears to be lognormally distributed Halley and Inchausti 2002 Cloern and Jassby 2010 Direct proportion
13. equation for chlorophyll a Original usual estimate bootstrap estimate based on optimism bootstrap 5 3 Choosing when to add a second predictor Is there some practical method for deciding whether or not to add a second predictor using only the statistics usually available from a regression fit We did the same analysis on those calibrations for which the addition of a second factor resulted in statistically significant p lt 05 coefficients for both in vivo fluorescence and SPM n 65 compared to the original set of n 162 It was gratifying to find that the bootstrap results supported the use of SPM in all but four of these cases Table 1 We could therefore simply require in addition to the other assumptions of least squares regression that both predictors regression coefficients are significant regardless of the apparent i e biased size of the R and RMSE changes A problem with this heuristic criterion is the large number of false negatives n 15 when at least one of the predictor coefficients is not statistically significant In the language of public health statistics the test shows a high specificity 82 86 or 95 but a less stellar sensitivity 61 76 or 80 We wouldn t actually make predictions worse except in a few cases but we would miss many of the cases when adding a second predictor helps Table 1 Contingency table describing the extent to which improved R is related to statis tical signif
14. in North SFB specific conductivity temperature depth turbidity some sites dissolved oxygen some sites specific conductivity temperature depth pH dissolved oxygen turbidity fluorescent dissolved organic matter chl a and phycocyanin fluorescence nitrate phosphate specific conductivity temperature depth pH dissolved oxygen turbidity chl a fluorescence salinity temperature depth pH dissolved oxygen turbidity chl a fluorescence specific conductivity salinity temperature depth pH dissolved oxygen turbidity 2 2 Goals of the Nutrient Moored Sensor Program A recent report summarizing high priority data needs for informing nutrient management decisions in San Francisco Bay identified that the collection of nutrient related data at higher temporal resolution was essential both for better assessing condition in San Francisco Bay and for calibrating water quality models that will be used to explore nutrient cycling and ecosystem response under current and future environmental conditions SFEI 2014 731 Moored sensors offer a number of advantages over traditional ship based sampling and lab measurement by allowing logistically feasible and cost effective high frequency in situ measurements and autonomous operation However moored sensor use also comes with a number of challenges associated with deployment maintenance and data interpretation that need to be considered during program design The effort and cos
15. is a sample too small The poor performance when SPM was included for the 2006 08 02 cruise n 10 suggests that overfitting is a potential problem one that in principle increases as sample size gets smaller Is there any relationship between the change in prediction error and the number of samples This indeed seems to be the case Figure 9 There is a tendency for RM SE to rise i e predictions to get worse as n drops below 15 This is consistent with studies mentioned in Section 5 1 that at least 10 measurements per predictor are required to guard against overfitting Above 15 there is no obvious relation to n So the addition of a second factor for data sets smaller than n 15 and certainly smaller than n 10 brings much greater risk of getting a biased calibration equation A prudent approach to calibration would therefore eschew a secondary predictor for transects in which n lt 20 i e within subdomains of the Bay In a way choosing subregions is like choosing a second predictor with the main difference being that the predictor takes on discrete values namely subregion A B etc rather than continuous values This makes sense when the stations are grouped because of some 14 RMSE change when SPM added I I 10 15 20 25 number of measurements per cruise Figure 9 Change in prediction error RM SE versus sample number when SPM is added as a second predictor for chlorophyll a calibration 15 unmeasured variable like
16. located with a USGS SacSed site but that data is not discussed in this report The NMSP sensors at Dumbarton Bridge are fixed 12 m above the bottom and their depth below the water surfaces varies between 1 4m depending on tidal phase Figure 2 2 USGS SacSed has sensors deployed at two other depths at the same location 7m above bottom and 1 m above bottom with several parameters measured at each depth temperature specific conductivity depth turbidity and the mid depth location temperature specific conductivity depth turbidity DO at the deep location The USGS SacSed and NMSP sensors are deployed in a similar manner via a suspension cable attached to a davit on the bridge platform Access to the sensors is from one of the Dumbarton Bridge structural support platforms allowing for ample work space In addition the Dumbarton sensors are deployed between the concrete bridge support and a set of rubber bumpers that surround the bridge support This configuration prevents the sensor packages from inadvertent contact with boats and limits the risk of vandalism and theft The site is also equipped with hard power battery back up and telemetry that allows for real time data upload USGS SacSed also deploys an acoustic Doppler velocimeter ADV at Dumbarton Bridge which continually measures water velocity and allows for analysis of the effects of tidal currents on water quality davet gt Deployment a line Bridge Ln p
17. may or may not represent the optimal locations for sensor placement The current goal is to continue these deployments through at least the end of Year 2 The approach for determining future sensor locations is discussed in Section 6 2 Figure 2 3 Steel frame deployed at Alviso Slough pulled to surface for servicing left and a view of the site right Photos courtesy of K Weidich USGS 2 4 Initial Instrument Selection Prior to selecting instruments several options were researched to identify appropriate sensors and sensor packages Along with a number of programmatic considerations that helped focus the search for sensor packages Table 2 3 the goal was to begin with an off the shelf field ready instrument or set of instruments as opposed to developing customized sets of instruments with the rationale that this approach would allow us to more quickly reach the field testing stage and focus effort on the program development related to logistics maintenance and data management Year 1 efforts were also intended to focus on a limited set of analytes including several basic water quality parameters temperature specific conductivity turbidity dissolved oxygen chl a and nitrate Initial research into available sensor packages narrowed the options down to two realistic choices 12 the Satlantic LOBO system a package consisting of WQM multiparameter instrument built by WETLabs plus the Satlantic SUNA v2 nitrate sensor or the Y
18. phytoplankton variabil ity in estuarine coastal ecosystems Estuaries and Coasts 33 230 241 doi 10 1007 12237 009 9195 3 Cosgrove J and M A Borowitzka 2010 Chlorophyll fluorescence terminology An in troduction in Chlorophyll a Fluorescence in Aquatic Sciences Methods and Applications edited by D J Suggett O Pr il and M A Borowitzka chap 1 pp 1 17 Springer Netherlands doi 10 1007 978 90 481 9268 7 Draper N R and H Smith 1998 Applied Regression Analysis 736 pp Wiley doi 10 1002 0471722235 Efron B and R J Tibshirani 1993 An Introduction to the Bootstrap 436 pp Chapman amp Hall CRC doi 10 1111 1467 9639 00050 Falkowski P G and Z Kolber 1995 Variations in chlorophyll fluorescence yields in phy toplankton in the world oceans Australian Journal of Plant Physiology 22 341 355 Good P I and J W Hardin 2006 Common Errors in Statistics And How to Avoid Them 254 pp Wiley Interscience Halley J and P Inchausti 2002 Lognormality in ecological time series Oikos 99 518 530 Harrell F E 2013 rms Regression modeling strategies R package version 4 0 0 Harrell F E K L Lee and D B Mark 1996 Multivariable prognostic models Issues in developing models evaluating assumptions and adequacy and measuring and reducing errors Statistics in Medicine 15 361 387 Helsel and Hirsch 2002 Statistical methods in water resources in USGS Techniq
19. statistical models beforehand the less likely we will be misled by spurious and biased correlations So let s first consider the appropriateness of each variable and exclude it whenever possible e Pheophytin a Its emission spectrum has a broad overlap with that of chlorophyll a and pheophytin a can be important because of its role in photosynthesis and as a degra dation product of chlorophyll Pheophytin a however is measured only in conjunction with direct chlorophyll a measurements and cannot be used to refine chlorophyll a esti mates from in vivo fluorescence Nevertheless its variability may already be accounted for at least partly by chlorophyll a and SPM e Dissolved oxygen Not known to have an important effect on in vivo fluorescence yield at least not for the range typical of estuaries It will be ignored in the analyses here e SPM Overlaps phytoplankton and therefore both chlorophyll a and pheophytin a but also contains additional fluorescence sources in the form of resuspended micro phytobenthos and phytoplankton derived detrital particles Irigoien and Castel 1997 Also may interfere with light transmission during fluorescence measurement e Temperature Fluorescence is affected by temperature Early studies based on the limited data available Lorenzen 1966 reported a temperature coefficient of only 1 4 C implying an 8 change for the interquartile range in our data set But we now understand that th
20. stopper and attach the flask to the pump via a second piece of tubing Make sure valves on filter manifold are turned to off Remove filter funnels and wipe with kimwipe With forceps place 25 mm filter on frit black disc concave side up Replace filter funnel Shake sample vigorously and do a 2x rinse of measuring vessel Start with the 43 7 mL vessel Only use smaller one if water is extremely turbid and filters clog Shake sample again and fill measuring bottle creating a meniscus Pour into filter funnel Repeat the previous two steps for the other filter tunnel Turn on vacuum pump being careful not to exceed 5 psi and open valves As soon as all water has disappeared turn of vacuum pump A 23 Unscrew filter funnel and use forceps to fold filter in half inward so filter contents are protected and remove excess water with blotting filter Place filter on foil and fold on all sides fully enclosing the filter Repeat all steps for a triplicate sample Attach a label on all samples that includes date time in PST volume used and unique sample ID Place sample in an opaque container with dessicant and store on dry ice As soon as possible transfer sample to USGS Menlo Park for analysis Store on dry ice or on 80C freezer in the meantime A 24 Ea After servicing 324 Post servicing procedures Upon returning the office it is important to unload equipment as soon as possible to prevent mold or saltwater damage e Unp
21. underwater at all times Because of the shallow photic depth in San Francisco Bay SFEI is currently exploring options of deploying the instruments via floatation to measure the top 1 2m of the water column davit Deployment line r Bridge Sm platform 14m dependingon Sensor tide carriage Suspension cable weight Figure A 2 Deployment configuration at Dumbarton Bridge 2 3 Software overview The EXO2 communicates with the computer via the KOR EXO software This software is icon based so here is a brief glossary for what each icon means and is used for described once here rather than repeated throughout this manual z D 5 OF EVO Dashboard Gives real time sensor readings Used for pre post cleaning checks Calibrate Used for calibration checks and recalibration Deploy Used to set programming settings and stop deployment when you arrive on site Sites Ultimately will be used for GPS features but this is still in development by YSI Data Used to manage files stored on EXO2 visualize data and download to computer Settings The most common feature used here is to set the timezone for the sonde and select which data to report out i e raw RFU or ug L for chl a Connections Used to connect software to EXO2 Other icons wid Status of EXO2 connection Green check tells you KOR EXO is currently connected to the EX02 Status of programming Green check would indicate that EXO2 is currentl
22. water depths depending on tidal stage In Year 2 we are considering alternate deployment configurations because at both sites several parameters exhibit strong tidal variability and the fixed elevation variable depth configuration makes it difficult to distinguish whether differences in measured water quality result from vertical gradients detected as waters rise and fall or horizontal gradients detected as tidal action moves different water masses past the sensor Vertical profile data collected during R V Polaris cruises shows periodic development of thin 1 2m surface layers that have substantially different composition than slightly deeper waters e g factor of 4 difference in chl a between 1 m and 4 m Figure 4 1 To detect this variability and distinguish between vertical gradients vs horizontal gradients 2 or more sensors deployed at constant depths e g at 1m and 3m would be a better configuration In Year 23 1 we tested a floating buoy sensor combined with general davit configuration Figure 2 2 This configuration initially proved problematic apparently because forces from surface currents acting on the buoy shifted and damaged the steel mooring cables The floating designs can also be problematic for real time transmission when data or power cables are needed because slack in the cables at high tide can cause tangling and hang sensors out of the water With USGS SacSed s help we are currently testing other designs to ov
23. 0 data loss power failure DO 95 probe malfunction 85 data loss power failure fDOM 75 fouling 90 data loss power failure 90 85 f DO mg L probe malfunction data loss power failure and 20 estimated and 35 estimated 2 http gescience com Gescience 202 0 Templates probeguard html 30 Depth m SpCond uS cm Turb FNU 50 100 150 200 250 300 fDOM RFU 50000 5000 1 45000 40000 0 15 20 10 Temp C 15 25 1 1 oO ot 2 3 3 10 1 Chl a fl RFU O Q 7 my T T T zj iT 2 3 i f f E O Q T T T Jul2013 Oct2013 Jan2014 April2014 July2014 Jul2013 Oct2013 Jan2014 April2014 July2014 Figure 4 4 Provisional Year 1 data for Dumbarton Bridge Outliers have been removed and servicing dates are indicated by vertical dashed lines Data was omitted when lost due to telemetry failure t extreme fouling f or probe malfunction m When T probe was down T corrections for chl turbidity DO and DOM were estimated by the method described in Section 4 1 3 T and or SpC probe malfunction also interfered with accurate DO mg L measurements and was estimated by the method described in Section 4 1 3 and shown in green because of their potential uncertainty pH and phycocyanin fluorescence were not analyzed in detail in Year 1 Depth is depth of the instrument below water surface not total water depth to channel bott
24. 57 0022240973224274 Mayer L M L L Schick and T C Loder 1999 Dissolved protein fluorescence in two Maine estuaries Marine Chemistry 64 3 171 179 Twardowski M S and P L Donaghay 2001 Separating in situ and terrigenous sources of absorption by dissolved materials in coastal waters Journal of Geophysical Research 106 2545 2560 doi 10 1029 1999JC000039 Xia J Y Li and D Zou 2004 Effects of salinity stress on PSII in Ulva lactuca as probed by chlorophyll fluorescence measurements Aquatic Botany 80 129 137 doi 10 1016 j aquabot 2004 07 006 21
25. 8 attachments accessed April 10 2006 at http oubs water usgs gov tm1d3 Michael B Parham T Trice M Smith B Domotor D Cole B 2012 Quality Assurance Project Plan for the Maryland Department of Natural Resources Chesapeake Bay Shallow Water Quality Monitoring Program for the period July 1 2012 June 30 2013 Maryland Department of Natural Resources Annapolis MD A 32 Appendix B Improving estimates of chlorophyll from fluorescence in San Francisco Bay By Alan Jassby Improving estimates of chlorophyll from fluorescence in San Francisco Bay Alan Jassby February 7 2014 Contents 1 Introduction 2 2 The discrete monitoring data 2 3 Measurement uncertainty 4 4 Assessing calibration models 6 4 1 What form does the model take 2 2204 6 A2 Estimating prediction ertor lt 4 4 44 2 44 4 24 2284424 e 2 eR ES 7 5 Secondary predictors 7 51 Which factors to conside lt eses d aaro e854 Oe hes hwo RS 7 5 2 Are secondary predictors useful o oo oo a e a 11 5 3 Choosing when to add a second predictor a aos ao o a a 12 5 4 Salinity and temperature as predictors o o sooo e a ee eee 13 no When 16a Sample too small l s sos dorer S ario Ye we gi a wee a 14 6 Dealing with low chlorophyll 14 7 Discussion and conclusions 18 Many thanks to Jim Cloern and Tara Schraga for providing the data used in this report and for their advice on interpretation and ramifications of the
26. DO but uses an older generation of designs for its probes Numerous continuous sensor programs nationwide use the YSI 6920 and the USGS SacSed has been using this instrument configuration in San Francisco Bay for approximately 10 years however the EXO2 and its set of probes been less widely used Therefore this comparison is useful for assessing the EXO2 probes behavior and precision compared to widely accepted instrumentation for several parameters that are relatively straightforward to measure Figure 3 1 compares values for T SpC turb and DO saturation between the two sensors over 10 months of deployment n 20 000 for T turb and DO saturation fewer datapoints for SpC as discussed in Section 4 The comparisons indicate a high degree of precision r 0 89 for all parameters and strong correspondence among estimated values closeness of actual slope to the 1 1 line The turbidity probe measurements exhibited the most scatter and were furthest from the 1 1 line the slope over the full turbidity range was 1 2 but the relationship was closer to the 1 1 line at lower turbidity values that are more representative of conditions on the open Bay lt 100 FNU While the DO saturation results were strongly correlated r 0 95 the EXO2 registered slightly higher values 8 than the 6920 values 17 40000 y 0 98x 1360 r 0 93 y 1 00x 0 03 r 0 99 N 30000 EXO2 SpC uS cm 8 Q o o EXO2 Temp C
27. FB ou eeseestesseesseestesseeseeeseeeeestesneeseeeseesaeentesseesteeseenteeneenaees 8 Figure 2 2 Dumbarton Bridge moored Sensor site photo 0 eseessesseesseesesseesteesteestesseesteeseestesneeseesteseentesseeaneeaes 11 Figure 2 3 Alviso Slough moored Sensor Site photo oes cessesessesseessecstesseesteeseeseensesseeseeeseentesseeseeseeseenteaseeseeaes 12 Figure 2 4 Photos of NSMP equipment YSI EXO2 and SUNA V2 eestessesssesstesstesteeseesteestesseeseesteseentesseeseeass 13 Figure 2 5 Timeline of Year 1 NMSP activities e eeseessesssesssesseessesssesseestesseeseeestesseestesseeseeeseesteeneeseesteeaeeateaseeaseeass 16 Figure 3 1 Comparison of EXO2 and YSI 6920 during co deployment in Alviso Slough ees 18 Figure 3 2 Comparison of EXO2 and Turner 10 AU during co deployment on R V Polaris 4 19 Figure 3 3 Comparison of EXO2 and lab analyzed DO and suspended sediment samples during R V Polaris CVUIS S sec sccsczcssce sine snsitscisccxtcanee act nats avSaaneticielinit cine anftzersaitineausisdtinasueCanes aaia 20 Figure 3 4 Comparison of EXO2 and lab analyzed chl a samples during R V Polaris cruises 22 Figure 3 5 Comparison of Turner 10 AU chl fl values and discrete lab analyzed samples taken aboard the R V Polaris from 2005 2013 o ceeeesseessssssecseesseesseseessecseesseeseeeseestesseeseeeseesseensesseesneeaeenteeneeseeaneeaeentesseeaeeass 23 Figure 4 1 Vertical profile data at Station 32 near Dumbarton Bridge on
28. July 1 2013 24 Figure 4 2 Common EXO2 data challenges encountered at Dumbarton Bridge in Year 1 27 Figure 4 3 Typical EXO2 fouling from biological growth during warm summer month e 28 Figure 4 4 Provisional Year 1 EXO2 data for Dumbarton Bridge eesesssesseseesesseeseeestesseeseeestesteentesneesteeass 31 Figure 4 5 Provisional Year 1 EXO2 data for Alviso SIOUQH seesseesstesesseessesstecsteseeseeseestesneeseesteeeenteaneeseeaes 32 Figure 5 1 Comparison of EXO2 in site estimated chl concentrations from Dumbarton Bridge with discrete lab analyzed chl samples near Dumbarton Bridge over all Of Year 1 owe eseesseesesesesstecteeeeeeeees 35 Figure 5 2 Comparison of EXO2 in site DO mg L values from Dumbarton Bridge with discrete lab analyzed DO samples near Dumbarton Bridge over all Of Year 1 o eestessesssesstecstesteeseeeteestesneesteestesteeatesneesteeaes 36 Figure 5 3 EXO2 data at Dumbarton Bridge in June 2014 ees eeseesesseesseesteestesseeseeeseestesneeseeatesaeenteaneeseeaes 37 Figure 5 4 EXO2 data at Alviso Slough during June 2014 ee ceceeseeseesteseesteeneesseesteeseesteeneeseeateeseentenneesseeaes 38 Tables Table 2 1 Characteristics of existing moored sensor monitoring in SF Bay e ee eeseesesseesseeceeestesteeeeettenteene 8 Table 2 2 Nutrient Moored Sensor Program Goals uu cesssessesessssssesseesseeseeseeseesteeseeseeeseesseentesneeseeeatenteenteaneesseeass 10 Table 4 1 Percent of data retained from
29. O2 Note that the probes are not always installed in the exact order shown here The EXO2 is powered either internally from 4D batteries typically 90 day lifespan or 9 16V DC external power Data from all probes are temperature corrected to account for T effects on probe output When the T C probe is in place they get the T values from this probe with the exception of turbidity which always uses an internal thermistor However if the T C probe is not installed or is reporting NAs they all use an internal thermistor that is less accurate accurate to 1 1 5 C Below is a brief description of the methods operating range and accuracy of each probe Table A 1 EXO2 probe specifications ma pemen e pe Temperatur T probe uses a thermistor and reports C T 0 01 C e and 0 05 C when gt 35 C Conductivity C probe uses four internal pure nickel electrodes to T C measure solution conductance in uSiemen cm Can use C the greater of 0 001 conductivity data to caluclate specific conductivity SpC mS cm or 0 5 salinity sal and total dissolved solids TDS 1 above 100 ms cm Depth is calculated using pressure measured by a vented Upto100m 0 13 ft strain gage pressure transducer and water density there are calculated using T C data other deeper models Dissolved The optical DO probe uses a luminescent membrane to the greater of 1 of oxygen estimate percent saturation DO and then uses T C reading or
30. SAN FRANCISCO ESTUARY INSTITUTE e CLEAN WATER PROGRAM Nutrient Moored Sensor Program Year 1 Progress Update Prepared by Emily Novick David Senn Ph D CONTRIBUTION NO 723 Aaaa 4911 Central Avenue Richmond CA 948 p 510 746 7334 SFEI e f 510 746 7300 www sfei org San Francisco Bay Nutrient Moored Sensor Program Year 1 Progress Update July 2013 July 2014 December 2014 Acknowledgements This work was conducted as part of implementing the San Francisco Bay Nutrient Strategy and with funding from the Regional Monitoring Program for San Francisco Bay RMP and Bay Area Clean Water Agencies BACWA During Year 1 the San Francisco Bay Nutrient Moored Sensor Program s development benefited from discussions with and field laboratory assistance from a number of collaborators including D Schoellhamer G Shellenbarger M Downing Kunz P Buchanan K Weidich A Powell and P Castagna of the USGS Sacramento Sediment group B Pellerin B Bergamaschi B Downing and JF Saraceno of the USGS Sacramento Biogeochemistry group C Silva and T Von Dessoneck formerly of the USGS Sacramento Sediment group J Cloern T Shraga C Martin and E Kress of the USGS Menlo Park SF Bay Water Quality Research and Monitoring Program C Raleigh of SFSU RTC M Dempsey of DWR IEP A Malkassian of SFEI UCSC and A Jassby of UC Davis Table of Contents Tableof Contents cisscscciscctiasssinscasiceantectciascesictasensutescecateetnttainc duct a
31. SI EXO2 multiparameter instrument combined with the Satlantic SUNA v2 nitrate sensor Both options had pros and con and a summary of the decision criteria as well as the suitability of each option relative to these criteria are presented in Table 2 2 In upcoming years the question of the most suitable instrument s package will be revisited After considering the pros and cons of each option we selected the EXO2 as the basic water quality sensor package Figure 2 4a and the SUNA Figure 2 4b was determined to be the best option for nitrate While the convenience and biofouling prevention features of the LOBO WQM were compelling its much greater cost would have capped deployments in Year 1 to only 1 site limiting program development in terms of comparisons among sites and wrestling with the varied logistics associated with managing multiple stations with differing requirements In addition the LOBO system is designed to be a complete buoy package and not all its features were needed during Year 1 Lastly the EXO2s have been used successfully by the USGS Sac biogeochemistry group for the past 2 years and the USGS SacSed group may transition to EXO2s within the next couple years they currently use an older YSI sensor package The EXO2 sensor has 6 probes that measure depth temperature T specific conductivity SpC pH turbidity turb dissolved oxygen DO fluorescent dissolved organic matter f DOM chl a and phycocyanin fluorescence
32. San Francisco Estuary Institute Contribution No 724 SFEI 2014 Scientific Foundation for San Francisco Bay Nutrient Strategy Richmond CA San Francisco Estuary Institute Contribution No 731 SFEI in progress Lower South Bay Nutrient Synthesis Richmond CA San Fracisco Estuary Institute Contribution No 732 Wagner R J R W Boulger et al 2006 Guidelines and standard procedures for continuous water quality monitors Station operation record computation and data reporting U S Geological Survey Techniques and Methods 1 D3 Reston VA U S Geological Survery 44 Appendix A EXO2 Maintenance and Operation Manual December 2014 A 1 Fa Introduction This document is meant as a reference for the operation and maintenance of SFEI s moored sensor equipment We will begin with an equipment description and overview of specifications This is intentionally brief because more detailed information can be found in the EXO2 user s manual We will then describe procedures for before during and after instrument servicing that can hopefully serve as a step by step guide for fieldwork We will briefly discuss data management procedures but more information on data validation QA can be found in the main body of this report Lastly we include resources for supplies technical support and field safety Table of Contents 1 Introduction 2 Project Description 2 1 Equipment 2 2 Site descriptions 2 3 Software ove
33. Uncertainty esessssesessessssssesseessesssesseeseessesseeseesseesteeneeseesteeaeentesneesneeatenseens 19 3 2 1 Turbidity and Dissolved Oxygen estessesssesssessesseeseessessteeseesseeseecseeneesneesneeseecaeentesneeseeseesseentesneeseeatenseens 20 3 24 Chlorophy le ainceiciticasii at tenet dd Pee iaai daha iantitaain iene 21 4 Operation Maintenance and Data Manageme nt eeseessesessesseessecseesseesteeseesteeseesseesseeseestesneeseeateeneenteaneeaseease 23 4 1 Sensor Operation and Maintenance eseessssssesssesseeseeseesseesteeneesseeseesseeseeeneesseeseesaeeateaseeseeatesaeeneesneeseeatenseens 23 41 1 Sensor Deploym Ttm irura Hace ty ee lt teste dete eee a es eee ee Set diesels 23 4 1 2 Maintenance Schedule and Procedures sssssssssessesseesecnesnsessessseesseeeseesnessnsesnseenseeneessterseerseeesneeses 24 4 1 3 Sensor reliability under field COMGitIONS ee eeceestessessesstecstesseeseeeseesteeneesseeseeeseentesneeseeateseentenseesseeaes 25 4 2 Data Managementin iia aia a aeia 26 4 21 Data ACUI SWLON sissisodan iiaa stiveaieatesuesessidactedescatagadccensuatatuucnsuvsscavtast atest iadaaa niaaa rakasa daanan 26 4 2 2 Data Post Processing and Quality ASSUTaNCe ssss sssssssessesessssrsnrnnnnnnsnsnsnsnsnsssnnnnnnnnnnnnnnnnnnsnnnnsssnnnnnnnn nunn 26 4 2 3 Managing data for outliers sensor drift and FOULING eee eeestesneeseeseesseestesseeseesteseentesneesseeaes 27 4 2 A Provisional Year UGAataset ecissicsccissscessccaneccsccsa
34. ack chl filtering equipment to allow it to dry If needed rinse vacuum pump off saltwater could damage metal pieces Hang wet weather gear and lifevests out to dry Cooler containing chl a samples should be full of dry ice and placed in the freezer until transfer to USGS Menlo Park or a 80C freezer Summarize the day s activities in the Field Notes spreadsheet Enter the biofouling and sensor drift values into the Field Calibration Data spreadsheet A 25 Long term equipment storage If there is a sonde that is not going to be in use for some time it is important to store it properly e To store sonde for lt 1 month o pour about of water in the bottom of the calibration cup and store upright o Keep the port plug on to avoid damage to the pins e To store sonde for several months o It is best to remove all probes and replace with probe port plugs as well as remove batteries Protect probe connections with plastic caps Store turbidity DOM and chl a BGA probes with plastic cap on probe end Store DO probe submerged in a container of water m f DO probe is dry for gt 8 hours rehydrate by soaking in tap water for 24 hours o Store pH probe in pH4 buffer solution m If the pH probe has been allowed to dry soak overnight in 2M KCl solution 74 6 g KCI per 500 mL water Use pH 4 buffer if KCI is not available m pH probes occasionally need to be cleaned more intensely Soak for 10 15 minutes in a solution of dishwashing liquid th
35. ality between standard devia tion and mean is a characteristic of lognormal distributions We can see such a relationship in the Bay Delta by plotting the standard deviation of extracted chlorophyll a versus flu orescence first binning the data by increments of 0 1 fluorescence units Figure 10 The appropriate weighting for this situation regression of y on x where y is lognormally dis tributed is 1 x or the inverse square of fluorescence in our case Figure 11 compares a weighted and unweighted least squares regression for the calibration data of 2007 04 03 The weighted version essentially rotates the regression line so that it passes closer to the lower values and further away from the higher ones The end result is that the relative standard error is more equable at all values and in particular much improved at the lowest values Table 4 shows these relative standard errors for the 10 lowest values of chlorophyll a Weighting has decreased every one of these errors with especially good improvement for the lowest values The decision to use weighting like the decision to use a secondary predictor must be made for each calibration data set The major criterion is violation of the constant variance assumption which can be decided on the basis of statistical tests or a careful inspection of residuals from an unweighted regression Note that R will probably decrease and RMSE increase because the goal has shifted to minimizing the sum
36. anagement including developing automated QA QC scripts to clean data of obvious outliers Data interpretation Several other activities were carried out to test deployment approaches or to gather additional data 15 USGS SacSed has infrastructure and sensors installed at the San Mateo Bridge SFEI staff worked with USGS SacSed to install an additional mooring at the San Mateo Bridge to accommodate an EXO2 floating near the surface We field tested an EXO2 at the San Mateo Bridge for 1 month in September October 2013 However due to limited SFE field staff we elected to wait until Year 2 for prioritizing the San Mateo Bridge site and focused effort instead on the Dumbarton and Alviso deployments An EXO2 was deployed for 1 month along with other USGS SacSed equipment for a short term data collection effort on an intertidal mudflat 600 m southwest of the Dumbarton Bridge This informed NMSP goals by gathering data in yet another distinct and important subsystem intertidal mudflats comprise approximately 75 the area of LSB and 10 of the area in all of SFB and to explore how conditions there differed from deep subtidal Dumbarton and slough Alviso conditions Beginning in September 2013 an EXO2 was plumbed into the USGS Menlo Park research vessel R V Polaris surface water flow system that continuously pumps water from 1m below the surface while the ship is underway during sampling cruises The EXO2 was deployed on 13 full Bay o
37. astic brush e Q tips e rags e syringe e putty knife 6 Sonde supplies order replacements from YSI at www exowater com e probe wrench battery wrench copper tape USB adapter port plugs spare O rings Krytox grease Probe port plug magnet E Discrete Sampling Probe wrench and 1 Vacuum pump battery wrench 2 Filtering kit e manifold filters 25mm Whatman GF F blotting filters or clean paper towels Kimwipes Foil Labels Glassware tubing 3 Amber bottles for collecting sample 4 Dry ice cooler A 9 5 Wet ice cooler if Alviso is being serviced 6 Opaque amber container with desiccant for filters General field supplies 1 Zip ties 2 Tools e Crescent wrench e Screwdriver s e Snips Spare hardware Lifejackets Foul weather gear Field sheets see next page for example OF OL A 10 Field sheet pg 1 Site Staff SS Serial FileName Datafilechecked OT OSpC ODepth OpH Clwb ODO OCh OBGA CfpOM Noteson dataissues Visualize data from each probe EXO Battery Volts replaced Replace if lt 5 5V Fouling tevel _ Note fouling level and take pictures a ee ee eae ee Turbidity Value Field sheet pg 2 Calibration checks Calibrationchecks_____ __ SS Sta value error ae vd a One additiondl value 5k uSic ee E E ee gt standard standard ORFU Discrete sample collection time in PST i nk Oo 7 Z po a E ae ae A 12 Prepare turbidity standar
38. ata quality for multiple parameters The fouling rate was worst at Dumbarton Bridge in late spring summer and early fall Fouling appears to be less problematic at Alviso Slough and also less pronounced at other sites throughout the Bay based on USGS SacSed experience over the past 20 years Avoiding lost data at Dumbarton Bridge during periods with high biofouling rates would require maintenance trips at a frequency of lt 2 weeks During Year 1 we iteratively implemented several basic fixes to decrease the biofouling rate and some improvement was observed However to ensure high quality data minimize lost data and minimize required maintenance frequency we will continue exploring other ways to minimize biofouling in Year 2 Telemetry for real time data access has two major advantages knowing when sensors have failed or fouled and being able to schedule maintenance trips to minimize lost data and triggering event based sampling in response to a detected event Of these two minimizing 39 lost data is the most important in the near term and alone provides strong justification for installing telemetry where possible Even with only two active stations in Year 1 large amounts of data are being generated by the 15 minute sampling interval for multiple parameters As we add more NMSP sites and other potential analytes e g nitrate phosphate etc the influx of data will increase considerably In addition managing real time data requ
39. ated by vertical dashed lines Figure 5 3 shows 3 weeks of moored sensor data from Dumbarton Bridge from June 2014 capturing the development and breakdown ofa 10 15 ug L chl a phytoplankton bloom Chl a begins to increase following a period of 2 3 days of lower turbidity and continues to increase within a 5 day window coincident with a 5 C increase in water temperature that would favor higher growth rates The upward inflection in chl a also corresponds with neap tide suggesting that phytoplankton may have additionally benefitted from higher light levels due to less vertical mixing during a period of lower mixing energy Following the upturn in chl a DO begins to decrease and departures below 5 mg L are evident at lowest tide While the timing of low DO could be related to the respiration of newly produced biomass within the open bay areas of Lower South Bay it is also possible that the DO decrease is related to the spring tide which would draw more water out of margin habitats where DO concentrations are commonly lower 36 A i 23 w j Fo I Temp C 21 19 500 17 V II G La HY A AM Turbidity FNU 300 0 100 DO mg L 6 Chl a FI RFU May 30 Jun 04 Jun 09 Jun 14 Figure 5 3 The development and break down of a phytoplankton bloom captured at the Dumbarton Bridge in June 2014 Figure 5 4 shows time series at Alviso Slough over the same 3 week period in June 2014 Tem
40. bset of data and begins exploring the following questions e What do moored sensors capture that may have been missed by monthly or bi monthly sampling e What do we learn about system dynamics based on the shorter time scale observations from moored sensors Since this report was intended to focus primarily on program development and not synthesis interpretations the initial interpretations presented below only scratch the surface Subsequent reports will delve further into data interpretation The data and interpretations below should be considered provisional as more work on sensor calibration and data quality assurance are needed Figure 5 1a presents moored sensor time series data from Dumbarton Bridge for chl a concentrations ug L estimated from in situ fluorescence using the regression in Figure 3 4 overlaid with discrete samples taken during R V Polaris cruises at three nearby stations in Lower South Bay Considering the wide range of conditions and the potential uncertainties there is excellent agreement between USGS discrete samples and NMSP in situ chl a concentration estimates over the course of 1 year The tidally driven variability in concentration at Dumbarton evident as high frequency max and min and thick shaded areas correspond well with the measured concentrations in discrete samples collected at stations near Dumbarton The continuous data also captures the seasonal variability in chl a lower baseline chl a in the fall an
41. by temporary sensor obscuring or other sources of short lived error The procedures for this were adopted those used by the MD Department of Natural Resources B Smith pers comm Each Level 0 data point is compared to to the average of the data points from 1 hour not including the point in question If the value is lt 3x this rolling average the point is assumed real and is unchanged If the value is gt 3x this rolling average it is assumed an outlier and replaced with a linearly interpolated value All data at this point is being Level 1 processed Level 2 data involves correcting time series for the effects of biofouling and sensor drift While SFEI is currently collecting the necessary information in the field to make these corrections this level of data processing is currently not occurring See Section 4 2 3 in main body of the report for details on the difficulty instituting these procedures We are also still grappling with how to infer accurate chl a concentrations from fluorescence signals given the abundance of known interferences see Section 3 2 in the main body of the report We are currently working to collect a sufficient number of discrete samples to robustly explore this question and will update this manual as results become available A 28 EA Resources 54 Useful phone numbers Table A 2 Helpful phone numbers EXO2 manufacturer 937 767 7241 Satlantic SUNA manufacturer 902 492 4780 Cambell Scientific Datal
42. c increases in chl a which correspond to concentrations of 30 50 ug L cannot be explained by in situ production within the short time periods when the concentration increases occurred Instead co occurrence of chl a maxima and water elevation minima suggest that areas upstream of this site either within the slough or in margin habitats act as tidally driven sources of high phytoplankton biomass or that tidal action resuspended benthic algae whose chl a was then measured in the water column by the EXO2 Turbidity FNU 30 100 0 100 DO sat 40 60 80 15 Chl a fl RFU 10 T T T T May 30 Jun 04 Jun 09 Jun 14 Figure 5 4 EXO2 data from Alviso Slough during June 2014 Our EXO2 T sensor was down during this time so T data is not shown and we chose to show DO saturation b rather than estimated DO mg L 38 Depth m 6 0 Main Observations and Priorities for On going Work 6 1 Summary and Main observations from Year 1 During Year 1 of the Nutrient Moored Sensor Program NMSP instruments were deployed at 2 sites Dumbarton Bridge Alviso Slough beginning in Summer 2013 with probes for chl a dissolved oxygen temperature conductivity turbidity and several other parameters Activities during Year 1 included o Identification of appropriate sensors laboratory testing of sensors and selection of Year 1 deployment sites o 25 field days related to mooring and instrument installation and maintenance trips Si
43. composition of water draining from multiple sloughs and tidal wetlands at a location up estuary from where it mixes extensively with open Bay water Work will also continue at the newly established San Mateo Bridge site and at the Dumbarton and Alviso sites It is anticipated that the nutrient related monitoring San Francisco Bay will rely on both ship based and moored sensor monitoring SFEI 2014 724 A high priority related to that program s development is to determine the optimal combination of moored sensors and ship based sampling a balance between information gained accuracy reliability of data and cost for monitoring ecosystem condition and informing nutrient management decisions To date most monitoring in the Bay has been conducted in the main channel However it is well known that the Bay s broad shoals are areas of high productivity and that shoal conditions can differ substantially from those in the main channel e g Thompson et al 2008 In Years 2 3 as part of nutrient monitoring program development a refined plan for moored sensor distribution in the Bay will be developed through analyzing historic USGS monitoring data and NMSP data to determine what lateral and longitudinal spacing of fixed stations is needed to capture the greatest variability in the system 40 Comparisons of Year 1 data at Dumbarton Bridge and Alviso Slough clearly indicate that slough sites have extremely different conditions than main channel
44. cted by USGS USGS collects 10 20 discrete samples per cruise for chl a suspended sediment and dissolved oxygen These samples are collected throughout SFB and therefore across a wide range of conditions This approach allowed us to obtain a large number of paired measurements which is needed to develop meaningful in situ calibration curves and quantify the uncertainty associated with estimated values e Analyzing an 8 year record of paired chl a fluorescence and lab analyzed concentration data collected aboard R V Polaris cruises to assess prediction error in chl a concentrations While USGS Menlo Park uses different instrumentation aboard the R V Polaris and we will ultimately want to confirm these results for the EXO2 the existing USGS dataset is much larger than the NMSP Year 1 dataset of the NSMP and will allow us to develop an a priori understanding of in situ fluorometer precision over a range of conditions 3 2 1 Turbidity and Dissolved Oxygen During Year 1 155 115 and 125 paired measurements of chl a turbidity suspended sediment and dissolved oxygen respectively had been collected throughout the Bay The relationship between suspended particulate SPM matter concentrations and the EXO2 turbidity sensor response were linear and strongly correlated Figure 3 3a r2 0 93 although more data are needed in the medium to high SPM range gt 50 mg L DO estimated by the EXO2 which is calibrated based on 02 concentration in air agreed we
45. ctor coeff cients and an otherwise well behaved calibration model appears to be a conservative criterion for using a second predictor i e reliable but overlooking some cases when a second predictor would in fact help e Secondary predictors should be avoided in smaller samples n lt 20 based on both published simulation studies and results in this report e Weighted least squares regression should be used when the constant variance assump tion of ordinary regression does not hold Use of weighting will also reduce prediction error where it matters most namely for low values of chlorophyll a e Given the current understanding of factors influencing fluorescence locating homo geneous subregions and using weighted regression are preferable to using secondary predictors References Almeida A M M M Castel Branco and A C Falc o 2002 Linear regression for cal ibration lines revisited weighting schemes for bioanalytical methods Journal of Chro matography B 114 2 215 22 Babin M A Morel and B Gentili 1996 Remote sensing of sea surface sun induced chlorophyll fluorescence consequences of natural variations in the optical characteristics of phytoplankton and the quantum yield of chlorophyll a fluorescence International Journal of Remote Sensing 17 2417 2448 19 Chernick M R 1999 Bootstrap Methods A Practitioner s Guide Wiley Cloern J E and A D Jassby 2010 Patterns and scales of
46. ctsastGiascasbessneedttiancesaciaienssuteanedatuet NEA net agiaastdiaecaittas 3 Fig Tre Sgan kana n darn Aaistastacesiereeaeyd se Bade nandaletatveapctsd Hae asaaxtsdeetr sounder acters axe ean se 5 TADES a nea a A A a Ta a A A 6 T I trod ttg ensesinde aA aaia aiiai 7 2 Programi OVErViIEW satirici iaioa a a a a 7 2 1 Moored Sensor Programs in San Francisco Bay sesssessssessssssesssesseessesssecseeseesseeseeestesaeeatesneesteeatesteeneeaneeaeeass 7 2 2 Goals of the Nutrient Moored Sensor ProgramM ssssssssssssssessssssnnunnnnnnnsnensnsnssssnnnnnnnnnnnntnnnnonnnsnnssnnnnnnnnnnnnt 9 Zid Year I Site SelecHoNsssscssiranaani a aaa a i 11 2 4 Initial IMStFUMENE Selec Nains aiaia Daai 12 Zid OVETVIEW OF Year T ACU VItleS accccsscecanccccstiaceaecsacisacieaseisnscosvecctednctsscesicisieedaseeaneciviesuttdhecaestaciaastineeashesaniedndieneedis 15 3 Sensor In situ Calibration and Uncertainty uu cseescesssesssessesssessesseesseesseeseesseesecstesseesseeseeateeneeseestecaeestesneesteeatenseens 16 3 1 Comparison of EXO2 with co deployed SCNSOTS essesssesssecstesseeseessteseeneesseeseesseesteeneeseesteeaeentesneesteeatenseens 17 3 1 1 Stationary In situ COMPATis OM wssccrsescarssisscesceiescesscrsvstsce escrdsevareussacencsvatcudsessacsvesteseivenstedavcedseraventsesennvaisatis 17 3 1 2 Comparison Bay wide During Monthly Cruises eeseessesseessescsesseesseeseeeseeseestesseeseeeatesseentesneesteeatenseens 18 3 2 In situ Sensor Calibration and
47. d We want to aim for 100 NTU turbidity standard since this is the typical upper range of conditions at the Dumbarton Bridge To prepare this perform a 40x dilution of 4000 NTU formazin standard 25 mL standard 975 mL milliQ water for 1L of solution Check the value of the solution using the turbidimeter in SFEI s lab take the average of 3 readings pour the dilution into an opaque 1L bottle and label with the resulting NTU A 13 Calibrate the spare EXO2 It is important to have a fully calibrated spare EXO2 in the field in the event that a deployed sonde malfunctions All calibration criteria with the exception of fluorometric probes are consistent with USGS recommendations Wagner et al 2006 Fluorometric criteria are adapted from guidelines used in the Maryland Department of Natural Resources Shallow Water Quality Monitoring Program for Chesapeake Bay Michael et al 2012 these same procedures should be followed for field calibration Step 1 Start KOR EXO2 and plug the EXO2 to the computer using the USB adapter it can be picky about this order If the copper sonde guard is on the EXO2 replace it with the plastic sonde guard the copper sonde guard can leach into the standard and interfere with calibration Step 2 Connect to the EXO2 e Navigate to the Connections menu and select Rescan connections e Select EXO USB Adapter xxxxxx from the list and hit Connect Step 3 Perform a 1 point calibration on flu
48. d winter with few blooms and increasing baseline concentrations and higher peaks throughout the spring However while the ship based sampling program identified many of the blooms the continuous data captured much more structure related to the formation and termination of blooms and identified several blooms missed by discrete sampling which will translate into more accurate estimates of overall production and allow for better model calibration In fact the discrete sampling captures only a fraction of the variability in chl a Figure 5 1b presents a zoomed view of December 2013 and January 2014 and offers mechanistic insights into bloom size and origin For the December bloom the baseline chl a signal remains elevated over multiple tidal cycles indicating that the bloom extends both north and south of the Dumbarton Bridge and is sufficiently large that the highest and lowest tides do not bring low chl a water past the sensor at Dumbarton The early December bloom s chl a fingerprint differs substantially from that of the late December early January bloom during which chl a peaks at high tides but returns to baseline levels at low tide This fingerprint suggests that the bloom originated in LSB increases on the outgoing tide and does not extend north of the Dumbarton baseline chl a on flood tide and that biomass was tidally pumped out of LSB low chl a on flood tide These tidally driven variations in measured chl a at Dumbarton are likely
49. d funding for the Nutrient Moored Sensor Program NMSP The first NMSP sensors were installed in July 2013 and additional sensors were installed in Sept 2013 and July 2014 This report is a Year 1 July 2013 July 2014 progress update of the NMSP The report begins with a summary of the NMSP and an overview of progress to date Section 2 Year 1 observations and results are then discussed in terms of insights on sensor accuracy and calibration Section 3 protocols for sensor maintenance and operation Section 4 and initial interpretations of data Section 5 Lastly Section 6 discusses the value of moored sensor data based on the main observations from Sections 3 4 and 5 and presents recommendations for Year 2 and beyond 2 Program Overview 2 1 Moored Sensor Programs in San Francisco Bay Several programs currently operate moored sensors in SFB including multiple USGS groups San Francisco State University Romberg Tiburon Center SFSU RTC National Estuarine Research Reserve System NERR and CA Department of Water Resources Environmental Monitoring Program DWR EMP Figure 2 1 Each of those programs has its own set of goals which shape their geographic focus and the parameters measured Table 2 1 some of which overlap with the NMSP s goals and needs However because of the current spatial distribution of stations i e largely concentrated in Suisun and the Delta or the parameters being measured e g few stations southwest of
50. data The work was supported by U S Geological Survey award G12PX01343 to the author 1 Introduction Phytoplankton biomass is a fundamental characteristic of estuaries and in vivo fluorescence of phytoplankton chlorophyll is a well established biomass index But estimating chlorophyll from fluorescence is affected by several kinds of uncertainty This report attempts to char acterize two of these uncertainties measurement uncertainty and prediction error using long term observations from the USGS s Water Quality of San Francisco Bay monitoring program Measurement uncertainty refers to the variability of chlorophyll a measurements on sub samples extracted from the same water sample i e analytical uncertainty It is straightfor ward to characterize with replicate subsamples The monitoring program routinely replicates chlorophyll a analyses and those data will be summarized here Prediction error refers to the uncertainty arising from a model relating in vivo fluo rescence to chlorophyll a In the case of discrete monitoring programs these models are calibrated using extracted chlorophyll a measurements from a subset of the sampled loca tions the models are then used to predict chlorophyll a at the remaining locations Estuaries arguably offer the biggest challenge in terms of prediction error compared to inland waters and the ocean Suspended particulate matter SPM is high and often mostly mineral plus detrital particles not p
51. de by side deployment of EXO2 sensors with other in situ sensors to assess comparability among sensor types o Sample collection and analysis for in situ calibration o Data analysis Side by side measurements of EXO2 probes chl a fluorescence turbidity and dissolved oxygen alongside other sensors used in SFB monitoring found good correspondence among the sensors building confidence in the of EXO2 and indicating that it should be feasible to compare NMSP estimates with those from other stations in the Bay that employ different sensors Data comparability among sites is a prerequisite for developing a regional moored sensor network among multiple otherwise independent programs i e Figure 2 1 The primary EXO2 probes are capable of estimating parameters with fairly high accuracy Initial data analysis suggests that prediction errors 95 confidence interval for DO chl a and turbidity are 1 mg L 3 ug L 20 FNU respectively Continued effort directed toward calibration is needed to achieve these results across sites Collaboration with USGS SacSed on sensor deployment maintenance and data acquisition allowed considerable progress to be made within the NMSP in Year 1 It was also highly cost effective keeping maintenance costs at less than half of what would have been incurred had SFE staff carried out this work independently Biofouling is a major issue that varies in intensity by both location and time of year and degrades d
52. e also using other YSI products 2 5 Overview of Year 1 Activities A timeline of Year 1 activities is summarized in Figure 2 5 We had originally planned to focus Year 1 NMSP development effort on deploying maintaining and operating at only 1 site the Dumbarton Bridge In September 2013 though the opportunity arose to also deploy an EXO2 at the USGS SacSed site in Alviso Slough where they measure several relevant parameters turbidity DO SpC T but not chl a Through collaborating with USGS SacSed on field work and maintenance and with their technical assistance on sensor deployment the additional effort and cost to deploy at Alviso was modest Therefore with the idea that data collection at Alviso during Year 1 would further the Phase 1 goal of identifying locations for NMSP expansion we deployed an EXO2 at Alviso in September 2013 Major Year 1 activities related to Dumbarton and Alviso deployments are described below many of which were carried out in collaboration with staff from USGS SacSed Design and construction of floats and housings to attach instruments to the mooring cable Sensor testing and calibration including validation through side by side deployments with other sensors 18 maintenance trips to Dumbarton Bridge and 10 to Alviso Slough Testing approaches for minimizing biofouling Data logger programming sampling frequency data logging and data telemetry semi automated downloading of real time data and data m
53. e effects of temperature are actually much more complex Krause and Weis 1991 for example e Salinity In the case of the upstream Bay and Delta freshwater inputs may be accom panied by dissolved organic matter DOM originating in watershed soils a possible additional source of fluorescence Salinity decrease can serve as a surrogate for this DOM Twardowski and Donaghay 2001 while not having any apparent effect on its fluorescence Mayer et al 1999 Note however that strong salinity stresses can affect in vivo fluorescence of plants Xia et al 2004 0 20 40 7 9 10 20 0 600 Q 0 16 0 19 0 13 0 25 H N o oO wt Oo w wo oO 0 85 o60 P 0 044 N w N o oO N 2 f om oO oO 02 oO oO O o oO oO 0 5 2 0 0 5 15 0 200 0 15 30 0 4 8 Figure 7 Pairs plot of candidate variables for explaining fluorescence fluor chlorophyll a chl pheophytin a phe dissolved oxygen dox SPM spm temperature temp salinity sal surface irradiance sun and vertical attenuation or extinction coefficient ext 10 e Surface irradiance Eo An important source of potential variability in fluorescence yield through nonphotochemical quenching and other processes especially in transpar ent waters such as the open ocean Falkowski and Kolber 1995 e Vertical attenuation coefficient k Influences fluorescence yield at depth through its effect on irradiance The median k i
54. en soak for 30 60 minutes in white vinegar then soak for 1 hour in 1 1 bleach water solution If this is not done in the field every so often 3 4 months it is good to check O rings at all connections including probe ports and battery compartment as well as reapply Krytox grease as needed A 26 Er Data management and validation SFEI is still refining its procedures for ongoing data management and validation and this manual will be updated accordingly Ea Data storage SFEI is currently in the process of developing an automated real time data retrieval and storage system which will ultimately supercede the procedures described here However at this point data is still stored as bin or csv files e Archive the bin file with an appropriate filename site date ranges sonde serial e lf this was not done in the field transform the file from bin to xlsx using the KOR EXO software Navigate to the Data menu select View Export point KOR EXO to the correct tile and select e Copy and paste the newest data into existing spreadsheet with the complete record and save as a cSv A 27 42 Data quality The raw data that is taken off the EXO2 with no processing is internally referred to as Level 0 We store this data as bin and xlxs files as well as keeping a running spreadsheet of all Level 0 records in csv format for each site Level 1 data involves the removing and smoothing of obvious outliers caused
55. er monitored by moored sensors quality models Are there particular locations and or time periods where additional calibration data are needed beyond that collected at established moored sensor sites 8 Use moored sensor data to What indicators of nutrient related impairment can most accurately be assess condition in SFB monitored by moored sensors If when nutrient related impairment occurs along one or more pathways what extent duration can be detected by moored sensors 10 2 3 Year 1 Site Selection During Year 1 two major considerations guided initial site selection 1 Locations where there is currently limited continuous data collection for nutrient related parameters so new information would be gathered during Year 1 i e south of the Bay Bridge 2 Locations where other moored sensor groups are working to allow the NMSP to build upon existing infrastructure to collaborate and cost share on maintenance trips and to minimize logistical challenges and allow for increased attention to be focused on sensor operation and data management analysis Based on these considerations sensors were installed at two sites in 2013 both of them co located with instrumentation that was already installed by the USGS SacSed group Dumbarton Bridge near the deep channel and Alviso Slough 4 km upstream of confluence with Coyote Creek see Figure 2 1 NMSP sensors were installed at a third site San Mateo Bridge in July 2014 also co
56. ercome these issues Depth m ad T ji 23 2 236 240 62 64 66 68 70 10 15 Temp C DO mg L Chl ug L T T Figure 4 1 Vertical profile data for temperature a DO b and chl a concentration c at Station 32 taken during a USGS R V Polaris cruise on July 1 2013 As this plot shows surface layers can develop with considerably different water quality than the rest of the water column and moored sensor deployment locations should be designed with this in mind 4 1 2 Maintenance Schedule and Procedures During Year 1 sensor installations and maintenance were carried out in collaboration with USGS SacSed who already had sensors deployed at the current NMSP sensor locations and had the means to access sites current NMSP stations are only accessible by small boat Eighteen maintenance events were carried out for the 2 main NMSP sites with each event requiring 1 2 field days for NMSP sensor work and summing to 25 total field days Initially 2 SFEI staffers were needed for the NMSP maintenance however for routine servicing trips 1 SFEI staffer accompanied by USGS working on their sensors can now generally accomplish all the necessary work for the NMSP sensors with their current configurations Typical maintenance activities are described in more detail in a separate document maintenance standard operating procedure Appendix A and are described only briefly here Activities include i Downloadi
57. facturer replaced faulty probes at no cost In addition to losing T and SpC data for the malfunctioning periods that sensors malfunctions introduced uncertainty into estimates for other parameters T corrections are included when converting those probes raw output into estimated values When the T sensor is working correctly other probes use this temperature data for this correction If the T probe is no longer reporting data at all each probe defaults to an internal thermistor that is less accurate than the T probe For periods when T probes failed we worked with YSI engineers to fill data gaps and apply any corrections A 1 C uncertainty in T data would result in 1 5 uncertainty in DO sat values and less than 0 1 uncertainty in chl a RFU values so we feel fairly confident in temperature corrected estimates made using the internal thermistors Unlike T the SpC data was 25 not recoverable because there are no back up SpC sensors installed on the EXO2 Estimating DO concentration in mg L from DO saturation requires both T and SpC data since the saturated DO concentration varies as a function of both T and SpC To estimate DO mg L from saturation during periods when T SpC probes malfunctioned we used the internal thermistor values and SpC data from nearby USGS sensors The DO concentration estimates are moderately sensitive to potential uncertainties in SpC and T 0 2 mg L assuming SpC uncertainty of 5 000 uS cm and 0 15 m
58. ffort focused on goals related to program development and structure and building technical capacity to sustainably manage the program through a team including SFEI staff and collaborators Table 2 2 Nutrient Moored Sensor Program Goals Colors indicate the relative priority of each goal during Phase I and II dark medium and light blue indicate high medium or low priority or emphasis respectively Key Questions Phase 1 Phase 2 priority priority 1 Identify the best sensors or What locations are feasible for sensor deployment Sections sensor packages considering How frequently do the sensors need to be serviced How does this vary KATZ program goals and develop capacity to deploy and maintain moored sensors seasonally What biofouling prevention tools are most effective 2 Create adapt procedures for What are standard procedures exist for data processing i e removing Z4 WIO automated data acquisition outliers correcting for fouling drift 4 2 4 3 data ara AATA ae real What data processing can be automated ume us 7 requency data What are the data visualization needs of the NMSP and what is the visualization best way to address those needs 3 Develop understanding of What parameters can be accurately measured by moored sensors Sections sensor accuracy and potential How well does the EXO2 agree with other moored sensors being used ESPA interferences in SFB 6 3 What discrete sampling is necessary to verify sensor o
59. ficient of variation from replicates Vertical dashed line recommended guideline of 0 05 _ 1 00 chlorophyll a SE ug y 05 1 0 10 0 100 0 chlorophyll a mean ug 1 Figure 4 Standard error versus mean for replicated chlorophyll a measurements Dashed line SE 0 05 x mean 4 Assessing calibration models 4 1 What form does the model take Our goal is to come up with a calibration equation that gives the best chlorophyll a pre dictions for in vivo fluorescence measurements that are not part of the calibration process The exact form of the equation is constrained by the number of data available for fitting equation parameters which in this case is the number of extracted chlorophyll a samples per transect Transects generally extend throughout the Bay into the Delta but these are supplemented by shorter cruises in South Bay during periods of high biological activity The number of extracted chlorophyll a measurements per cruise day thus varies between about 10 15 and 25 30 depending on the transect length Figure 5 30 ee oe e ea ee e ee o esse D 0D 000000 6000 00000000 6 e 7p ee o o e o o eso oo e 09 a e e ee ro e o e e e o gt 20 e z Q O pas O o o eee eoo Fa Sees o ee ee O a o ZI e eo o ow aso 000o oco 6 e oo o ee e 10 e e e e e ee eee o e e e e 8 eo I l l I 2006 2008 2010 2012 date Figure 5 Numbers of extracted chlorophyll a samp
60. g Increasing maintenance trips is costly 1000 day Some new equipment developed for the EXO2 can be used to reduce biofouling In addition other installations designs are also possible e g placing the instrument out of the water and pumping water through biofouling resistant tubing to the surface for measurement or using a winch that raises and lowers the instrument package and parks it out of the water between samples Finally other instrument packages may resist some types of biofouling better than the EXO2 e g the WQM may not have the same problems with SpC fouling because of a different sensor design All of these options have associated costs and tradeoffs and will be explored in Year 2 Table 4 1 Amount of data retained from raw data in provisional clean data at Dumbarton Bridge and Alviso Slough as well as common reasons for data omission Probe malfunction and fouling accounted for the greatest amount of data loss with much of the T SpC loss due to an manufacturing error in early models of the T SpC probe data retained in Main reason for data retained Main reason for provisional 2o in provisional a Dumbarton dataset omitting dati Alviso dataset gating dat Depth eae Wa h 90 data loss power failure T 85 probe malfunction 70 probe malfunction SpC 55 probe malfunction fouling 50 probe malfunction Chl a fl 90 fouling 90 data loss power failure Turb 70 fouling 9
61. g L assuming T uncertainty of 1 C These uncertainties in DO concentration introduced by the T and SpC data gaps is low compared to the 2 3 mg L average daily fluctuations in DO observed at each site so this approach estimation would not overly impact our ability to detect changes in the system At Alviso Slough there was one period when the entire sensor experienced a power failure and was down for gt 1 month Sept Oct 2013 The USGS SacSed Alviso sensors also experienced power failures around this time and they speculated that the cause was possibly due to near by electrofishing activities 4 2 Data Management 4 2 1 Data Acquisition At both sites the EXO2 is programmed to collect data at 15 minute intervals Although more frequent e g continuous or less frequent measurements are possible Year 1 experience indicates that hourly resolution data yields relevant information that would be lost at lower resolution Moreover the 15 minute data is a compromise resolution that allows for occasional outliers to be removed without sacrificing a full hour of information and permits operation for more than 6 weeks on battery power At the beginning of each set of measurements the EXO2 triggers a plastic bristled brush that cleans the probes prior to taking measurements Data is stored internally by the EXO2 and downloaded during maintenance trips At Dumbarton data was also telemetered allowing for SFEI staff to monitor sensor performance In t
62. had low chl a concentrations approximately 70 of samples lt 5ug L increasing the relative importance of interferences Because some of the potential fluorometer interferences can also be measured continuously alongside chl a fluorescence measurements e g turbidity fluorescent dissolved organic matter it may be possible to develop relationships that adjust for interferences and more accurately estimate chl a concentration As a preliminary test of the potential to correct for some interferents we included turbidity in a multivariate regression and found that r improved modestly from 0 67 to 0 72 Over time we may find that developing site specific or segment specific fluor chl a relationships decrease the prediction error associated with chl a estimates With only 150 samples Bay wide and fewer than 20 samples some bay segments there is currently insufficient data to test the improvement in prediction error from site or segment specific calibrations A recent analysis of a much larger dataset from the USGS SF Bay Water Quality Monitoring Program aboard the R V Polaris Jassby 2014 explores several approaches for improving the precision of chl a concentration estimates That analysis indicates that developing segment specific calibrations is more effective at reducing uncertainty than including additional predictors see Appendix B for report However that finding may result in part from the fact that analyses focused on single cruises wi
63. he future real time data can allow for more rapid response to biofouling or sensor failure and minimize lost data Eventually real time data could also be used to trigger event driven sampling but this was not pursued in Year 1 Installing real time capability at San Mateo is straightforward and we anticipate doing this in Fall 2014 Adding the necessary equipment for real time data at Alviso while entirely feasible would require more effort and cost than San Mateo and equipment may be more prone to vandalism or theft at this site 4 2 2 Data Post Processing and Quality Assurance In Year 1 we began developing procedures for semi automated data post processing Raw data from in situ sensors require substantial post processing and evaluation for quality assurance in order to identify and if possible correct for interferences drift in sensitivity and noise The NMSP instruments are measuring parameters on a near continuous basis 24 hours per day resulting in large amounts of data Semi automated protocols are therefore needed to efficiently manage certain post processing tasks like basic data cleaning to remove outliers For other needs 26 sensor drift correction identifying and removing compromised data due to fouling and identifying failed sensors additional manual post processing is needed Figure 4 2 presents several examples of commonly encountered data post processing needs and the possible approaches are described
64. hytoplankton Freshwater inflows also carry fluorescent dissolved organic matter DOM that changes with season and position within the estuary Both SPM and f DOM modulate in vivo fluorescence measurements but whether their influence can be incorporated into routine chlorophyll a estimates is another matter The basic ques tion we ask is How and when if ever should additional factors like SPM be used to estimate chlorophyll a The analysis is limited to those factors measured routinely in the USGS monitoring program Other important factors phytoplankton species composition and consequent optical properties for example Babin et al 1996 cannot be analyzed here Monitoring programs must resolve where to place the effort in reducing uncertainty The specific answer may depend on whether the goal is to assess regulatory compliance or to increase our understanding of the mechanisms at work But by examining these uncertainties we can improve the basis of monitoring design in San Francisco Bay regardless of the goal 2 The discrete monitoring data The USGS monitoring program measures water quality characteristics including in vivo fluorescence at up to 37 stations along a fixed transect from South Bay through Suisun Bay A vertical profile of water quality is recorded at each station which may include discrete water samples from one or two depths for measuring extracted chlorophyll pheophytin and other variables Data were d
65. icance of predictor coefficients AR gt 0 AR lt 0 Sum Both P lt 05 61 4 65 At least one P gt 05 15 82 97 Sum 76 86 162 5 4 Salinity and temperature as predictors Although we didn t examine salinity and temperature at the same level of detail a summary of the results shows much the same behavior as with SPM Table 2 In fact salinity and temperature fare slightly better than SPM For example use of SPM improves RMSE decreases prediction error in 51 of cases whereas use of salinity and temperature improves RMSE in 58 to 59 of cases respectively Table 2 Change in bootstrapped RMSE when secondary predictors are added to the chlorophyll a calibration Predictor Min 1st Qu Median Mean 3rd Qu Max SPM 1 36 0 170 0 000 0 036 0 115 2 940 salinity 1 27 0 209 0 040 0 029 0 050 4 050 temperature 1 26 0 196 0 026 0 039 0 080 1 660 The within cruise cross correlations among these variables for the calibration dataset are much bigger than the correlations for the overall dataset Figure 7 For example the 13 correlations between SPM and salinity exceed 0 58 and between salinity and temperature 0 90 in a quarter of the cruises Table 3 It s likely that some of these variables are simply markers of water masses that differ in multiple ways i e that they are stand ins for one or more mechanisms actually affecting the calibration be it a different one of these variables or some o
66. ied linearly to the entire time series Fouling can affect readings in less systematic and less linear ways and the underlying true signal may be difficult to determine Individual probes varied in their rate of and susceptibility to fouling and the cause of fouling The fouling rate also varies seasonally and by site Figure 4 2 illustrates how individual probes respond to the onset of fouling The SpC probe fouled due to the accumulation of sediments in the conductivity cell which occurred fairly gradually over the course of deployment Figure 4 2a Although earlier YSI SpC probe models also experienced some signal attenuation due to sediment at Dumbarton Bridge their design was apparently less susceptible to this problem USGS SacSed pers comm We have notified YSI engineers of the problem and continue to look for ways to minimize SpC probe fouling Fouled SpC data affects DO concentration mg L estimates in the same way as described in Section 4 1 3 however even if readings drifted by 10 000 uS cm due to fouling this would still only result in lt 0 5 mg L error in DO mg L The discernible effects of biological growth on probe response arise more abruptly influencing readings little until a critical amount of growth develops after which the probe response becomes increasingly erratic Figure 4 2b c The turbidity probe s response appears to react first to fouling although non ideal for turbidity measurements this may allow the
67. ign investigations to further constrain our understanding of sensor accuracy As described in Section 3 2 there are many potential factors that introduce uncertainty into the relationship between chl a concentrations and in situ fluorescence In year 1 we began both field measurements and data analysis work to explore this issue with the goal of over time developing a reliable chl a fluorescence relationship The data available to date have shown a fairly good chl a fluorescence relationship Bay wide Site specific calibrations or the addition of secondary 42 predictors turbidity DOM may help further reduce prediction error In Year 2 we will continue sample collection for in situ calibration due to interferences and potentially carry out one or more intensive studies to investigate factors that may influence the fluorescence per unit chlorophyll relationship particularly the importance of quenching 6 2 4 Strengthen collaboration across programs Several programs currently operate moored sensors in SFB Figure 2 1 In Year 2 SFEI will continue engaging with other programs to identify ways for increased cooperation and collaboration and for inter program data quality and calibration activities that will allow for reliable comparisons among datasets collected by different programs As an initial step toward engaging other moored sensor programs SFEI is developing a web based data visualization tool that is compatible for use across multi
68. important for assessing condition that information is also needed To explore the issue of the best vertical location s of sensors in Years 2 3 we will conduct pilot studies using 2 or more sensors deployed at multiple depths 6 2 1 3 Highest priority additional analytes In terms of additional analytes the highest priority in Year 2 is to install and develop calibrations for the SUNA nitrate sensor at Dumbarton Once the first SUNA is running reliably a second SUNA sensor may also be deployed at another site Additionally SFEI is part of a team led by UC Santa Cruz that was recently awarded two imaging flow cytobots IFC for real time high frequency measurement of phytoplankton abundance size and taxonomy These are expected to arrive at UCSC in early 2015 After laboratory studies one IFC will be deployed aboard the R V Polaris during cruises The second IFC is planned for in situ deployment at one of the NMSP sites for continuous measurements e g one sample every 1 2 hours and Dumbarton Bridge is a logical first choice given the excellent on site infrastructure and the strong gradient in chemical nutrients and biological phytoplankton biomass conditions Pilot deployments at Dumbarton Bridge would likely begin mid to late 2015 after sufficient experience is gained with real time sampling aboard the R V Polaris Depending on time and budget we will also consider piloting other analytes assuming instruments can be borrowed or
69. in Sections 4 2 3 50000 SpCond pS cm 40000 30000 b we ae wu ems cn ON oa o j JI Cc 5 o 297 amp Z 3 8l sa 5 E 4 V wX NY Ng a me NG Figure 4 2 Portions of Dumbarton Bridge SpC a chl a fl b and turbidity c data to illustrate common data challenges encountered in Year 1 1 indicates fouling due to fine sediment accumulation in SpC ports 2 indicates probe malfunction due to manufacturer defect 4 T SpC probes and 2 pH probes defective in Year 1 indicates potential outliers requiring further inspection 4 indicates fouling due to biological growth on and around probe heads see Figure 4 3 In year 1 we were only able to automate correction for outliers correction included in Figures 4 4 and 4 5 To the extent possible developing and automating procedures for fouling correction will be a high priority for Year 2 4 2 3 Managing data for outliers sensor drift and fouling Outliers are generally easy to identify because they do not persist for a substantial amount of time and the data recovers that is sensor readings return quickly to values similar to those before the disturbance occurred Adapting a procedure used by the Chesapeake Bay Monitoring Program B Smith pers comm any value that was more than 3x the mean of the surrounding 1 hour was considered an outlier and was replaced with a linearly interpolated
70. in decreasing predicton error by understand ing which mechanisms are at work at the within cruise scale Our uncertainty about the causal basis for adding a second predictor is it simply a stand in for some other variable suggests that at some point we will need to turn to laboratory studies of fluorescence 18 in estuaries Chlorophyll changes in new water masses can easily be confounded with other changes that alter measured fluorescence there are too many potential variables in estuaries The monitoring data provide a critical check on experimentally determined mechanisms but it s unlikely that accurate models based on monitoring data alone can be developed There appears to be a surprisingly small amount of unpublished laboratory work on fluorescence in estuaries which is a more complex phenomenon than in oceans and inland waters This is a research topic with real practical values and a good chance of success In summary e Measurement or analytical uncertainty for chlorophyll a is almost always a small frac tion of the measured value and relatively unimportant compared to other sources of uncertainty e A second predictor such as SPM can sometimes help reduce the error in chlorophyll a prediction from fluorescence about half the time for cruises since 2005 e Conventional statistics such as R and RMSE are overoptimistic about the usefulness of a second predictor But requiring statistical significance p lt 05 of predi
71. ing water and it may be difficult to confidently actual conditions In the specific case of turbidity the wiper s action may be creating the turbidity that is measured Lastly when the instrument is removed from the water the growth on and around the sensors may be disturbed to a degree that Efoul cannot be accurately estimated 29 Data directly affected by substantial drift or fouling have not been included in the provisional data shown here since procedures for cleaning that data are still being developed The full data record in Figures 4 3 and 4 4 at Dumbarton and Alviso consists of 500 000 measurements across the 8 parameters including bad data The percentages of data for each parameter that remained after removing either bad data or failed probes are summarized in Table 4 1 The datasets were most complete for chl a DO DO mg L and depth turbidity SpC and T had the most lost data In several cases the amount of lost data was non trivial Biofouling led to loss of a substantial amount of data for turbidity fDOM and chl a Some amount of data loss is to be expected however a major goal in Year 2 is to develop approaches for minimizing data loss due to fouling Broadly speaking there are two approaches for minimizing data loss due to fouling 1 conduct maintenance visits more frequently at problematic sites or during problematic seasons and 2 make changes to instrument configuration that help reduce the rate of biofoulin
72. ires periodic attention Data management developing and maintaining a database QA QC procedures and interpretation will require an on going investment in personnel SFE is also currently developing a web based data visualization tool to allow scientists stakeholders and regulators to explore water quality data in near real time across multiple sites Continuous data at Dumbarton and Alviso from Year 1 are already yielding valuable insights into ecosystem condition and dynamics that cannot be readily inferred from discrete sampling 6 2 Priorities for On going Work 6 2 1 Identify highest priority sites and analytes for future sensor placement 6 2 1 1 Geographic location The current plan is to add one more station to the network of NMSP stations in Year 2 The current plan is to carry out a pilot deployment in Coyote Creek near its confluence with Alviso Slough which is also close to where Coyote Creek opens up into Lower South Bay The reasons for considering this site are e USGS monthly ship based sampling does not extend this far south in the Bay and there is limited consistently collected water quality data there e Based on data that do exist this is an area where nutrient concentrations are substantially greater than the relatively well mixed open area of LSB due to proximity to the City San Jose s wastewater effluent There is little or no chl a data from this region e This location would allow for an integrated measure of
73. latform WZ 14m dependingon Sensor tide jcarriage Suspension a weight Figure 2 2 Photo of SFEI s Dumbarton Bridge moored sensor site pulled to the surface for servicing left and schematic for deployed configuration right 11 The Alviso Slough site is more basic because of the lack of nearby infrastructure i e no hard power or existing structures to attach instruments Figure 2 3 Instruments are attached to a steel frame that rests on the bottom in the middle of the slough The frame is tethered to a weight to keep it in place and prevent theft and the frame is retrieved via a steel cable that is tied to shore Sensor depth below water surface varies between 0 5 3 5m depending on tidal phase USGS SacSed deploys a multi sensor sonde temperature specific conductivity depth turbidity DO and an ADV at this site as well at the same depth as the SFEI sonde There is currently no telemetry at Alviso During Year 1 both sites have yielded interesting observations in regions that are otherwise underrepresented in terms of nutrient related parameters Co locating NMSP instruments alongside USGS Sac instruments also brought considerable benefit in terms of cost savings for maintenance trips technical capacity building and allowing NMSP deployments to use existing infrastructure and well tested deployment designs As the NMSP develops or as USGS SacSed station priorities shift these sites
74. latter overfitting occur To get a better understanding of the extent of this problem we examined each cruise date since 2005 testing SPM for its merits as a second factor in reducing prediction error The difference in behavior between the ordinary and bootstrap estimates of R and RMSE is illustrated in Figure 8 which shows the empirical CDF for the change in these quantities when SPM is added to the calibration equation The CDF based on the original i e uncorrected method of estimating R suggests that the change is always positive i e R always increases when SPM is included as a predictor along with in vivo fluorescence The CDF based on the resampling based bootstrap estimate however implies that R actually goes down slightly more than half the time Similarly the estimated change in RMSE using the original method is always negative i e the prediction error improves goes down when SPM is added Figure 8 But the bootstrap based CDF suggests that prediction error actually gets worse about half the time 1 00 0 75 0 50 0 25 gt l estimate bootstrap original empirical CDF gt I gt O o oO oO oO 0 75 0 50 0 25 0 00 I I I I I 1 0 1 2 change when SPM added as predictor Figure 8 Empirical cumulative distribution functions for the change in R top panel and RMSE bottom panel when SPM was included as an additional predictor in the calibration
75. leased short term Two leading candidates are phosphate and ammonium 41 6 2 2 Refine maintenance and data management procedures 6 2 2 1 Fouling prevention Year 1 observations demonstrated that biofouling is the greatest obstacle to achieving reliable high quality data from moored sensors especially at Dumbarton Bridge In the summer months at Dumbarton Bridge fouling began within just 7 10 days of deployment and can compromise a large portion of the data depending on the maintenance schedule We explored several strategies in Year 1 to reduce biofouling i e placing copper guards around and on probes and these were somewhat successful but high fouling rates still occurred More advanced antifouling devices are available from various manufacturers but at considerable expense and with propensity for mechanical or communication failure Field testing one of these devices is among the potential activities for Year 2 More frequent maintenance trips e g as soon as fouling becomes evident based on real time is another option for reducing the impact of biofouling However this option is also expensive 1000 2000 per servicing trip In addition it would require closer collaboration with USGS SacSed since the NMSP maintenance schedule is tied to USGS SacSed s maintenance schedule We are also considering other options such as different instruments with better biofouling prevention or other deployment configurations e g shifting to a
76. les per cruise day Studies show that a regression model is subject to possible overfitting when there are less than 10 20 observations per predictor Harrell et al 1996 Draper and Smith 1998 similarly suggest that the number of observations should be at least 10 times the number of terms Good and Hardin 2006 are even more stringent in their data requirements maintaining that n observations are required for m variables when n observations are required for a univariate model All of these works imply that in vivo fluorescence should be used as the only factor in the shorter transects and perhaps one additional predictor for the longer transects Accordingly we confine ourselves here to regression models with chlorophyll a as the response variable fluorescence as the predictor variable and at most one additional predictor We consider only linear regression models that are also linear in the predictors 4 2 Estimating prediction error The usual statistics describing a regression such as the coefficient of determination R or the standard error of residuals are not necessarily a good guide to the predictive accuracy for new data They are based on the data subset used for calibration which may not be very representative of the data as a whole and so they tend to be biased and in particular overoptimistic for out of sample data A better estimate of predictive ability can be gained from the use of resampling procedures such a
77. lity discussed in Section 4 2 3 e Even though the EX02 s wiper brush was successful at keeping individual sensor heads clean extensive growth occurred on the EXO2 housing on the titanium sensor stands and on the instrument carriage This growth appears to cause a microenvironment to develop around the sensors and during high fouling periods measured values likely do not accurately reflect water quality in the surrounding water during extensive fouling periods e During some periods e g at Dumbarton Bridge in July September growth occurred so rapidly that only 7 10 days of reliable data were obtained for some parameters This was not the case year round and for more than half the year monthly maintenance was sufficient e The seasonality of fouling rates points to the need for higher frequency maintenance trips during certain times of the year additional equipment to further minimize biofouling or both Possible further measures are discussed in Section 6 3 1 4 1 3 Sensor reliability under field conditions The EXO2 performed with a high degree of reliability during Year 1 in terms of power programmed operation measurements every 15 minutes and data logging during extended gt 4 weeks unattended deployments In addition most of the individual probes worked reliably during deployments However 2 pH probes and 4 T SpC probes malfunctioned during in Year 1 the result of bad probe batches YSI personal communication and the manu
78. ll with discrete samples measured by Winkler titration slope close to 1 Although the r2 0 61 for the DO relationship indicates a fair degree of scatter the prediction error 95 confidence interval is approximately 10 All of the DO measurements were made at fairly high DO concentrations and measurements between 2 7 mg L are needed to assess sensor response in this range a b 150 1 Discrete SPM mg L 100 Discrete DO mg L 7 0 50 100 150 i 4 S 7 8 2 19 EXO2 Turb FNU EXO2 DO mg L Figure 3 3 Comparison of EXO2 values and simultaneous discrete lab analyzed samples for turbidity suspended particulate matter a and dissolved oxygen b collected during transect cruises aboard the R V Polaris Red lines show the 95 confidence bands on the prediction error The r in a is 0 93 and the r in b is 0 61 One outlier was removed in a 20 3 2 1 Chlorophyll a In situ chlorophyll a probes measure fluorescence in the bulk water surrounding the probe and a corresponding chl a concentration is estimated based on a fluor chl a Chlorophyll a estimates obtained from in situ fluorometers are prone to a higher level of uncertainty than some other parameters e g SpC T DO because of interferents present in natural water and variability in the phytoplankton s physiological response that can complicate the fluor chl a relationship Suspended sediment dissolved organic matter or degraded phytoplank
79. more cost effective on a per sample basis than measuring discrete samples While sensory accurary is checked against standards of known value on every servicing trip see Appendix A and some parameters can be estimated with a high degree of confidence using in situ sensors e g T SpC and DO the estimates obtained for other parameters can be subject to substantial uncertainties e g chl a fluorescence turbidity due to potential interferences in natural water This section focuses on Year 1 efforts related to the in situ calibration of moored sensors for measuring chl a turbidity and dissolved oxygen We took two approaches to assessing the precision and uncertainty or prediction error of calibration samples under field conditions e Comparing concentrations measured in discrete samples collected alongside the sensor with EXO2 sensor readings at the time of sampling We carried this out in two ways First 19 we collected discrete samples during routine maintenance trips at the actual sites where sensors are deployed Building a dataset of paired discrete and in situ measurements by this approach has the benefit of providing site specific calibration data However the maintenance trip frequency of approximately once per month makes building the dataset a slow process We therefore also deployed an EXO2 aboard the R V Polaris plumbed to its flow through system and compared the probe readings to measured values of discrete samples colle
80. much more finely resolved than would ever be explicitly used in nutrient related regulations However the high frequency data and the tidal time scale variability it captures will allow for more mechanistically accurate water quality models to be developed and 34 increased confidence in the application of models to forecast response under future conditions which combined will aid in developing better informed water quality objectives 30 N gt StationNumber o EA oO 34 at 36 oO h oO Jul 2013 Jul 2014 Nov 18 Nov 25 Dec 02 Dec 09 Dec 16 Dec 23 Dec 30 Jan 06 Figure 5 1 A comparison of EXO2 estimated chl a concentrations from Dumbarton Bridge with discrete lab analyzed chl samples taken at the 3 nearest stations to Dumbarton Bridge over all of Year 1 a and during a bloom event b Concentration was estimated from EXO2 chl a fl values in RFU using the preliminary chl fl regression formula shown in Figure 3 4 but possible interferences from turbidity have not been rigorously considered Outliers have been removed and servicing dates are indicated by vertical dashed lines Similar to the observations for chl a the discrete DO and continuous DO concentration at Dumbarton Bridge agree well Figure 5 2 Both capture broad seasonal trends including a gradual increase in DO through winter as water temperature decreases due to increased saturation concentration at lower T and decreased in micr
81. n 2 hours When possible we should just subset from the USGS discrete sample but if our sondes are at different depths then it is necessary to take our own samples Samples should be taken on a logical 15 minute interval Taking a sample Use the Niskin sampling bottle to take samples Attach it to the bridge board with the retracting cable winch and set bridge board up on boat or on bridge platform About 2 minutes before you intend to take a sample engage the doors of the Niskin in the up position attach the hold wires to the pins in the middle of the bottle lower the Niskin so that the middle of the bottle is even with the platform or water surface at Alviso and zero the dial at this point Lower the Niskin bottle so that it is even with the EXO2 At Dumbarton this is 17 5 below the bridge platform At Alviso that depth depends on what the water depth is at the time the EXO2 is approximately x from the bottom of the slough Approximately 2 3 seconds before you intend to take a sample release the messenger assume it travels 10 ft sec to figure out timing Pull up the Niskin Do a 3x rinse of the opaque plastic sampling bottle and then fill at least half full Filter immediately if possible or put on wet ice to filter later within 2 hours Record the discrete sample time in PST on field sheet Filtering These procedures are adapted from those used by USGS Menlo Park Attach the filter manifold to the flask via the tubing
82. n the Bay is 1 3 m t which implies a very strong vertical gradient Cosgrove and Borowitzka 2010 describe three components of nonphotochemical quenching that can be distinguished by their dark relaxation kinetics with relaxation times ranging from seconds to hours But the dominant form accounting for up to 90 has a characteristic time of less than 1 min suggesting that the instantaneous sample depth irradiance E Eye is relevant for estimating quenching effects The existence of thresholds for these irradiance effects however may render them less important in the turbid waters of the Bay Marra 1997 for example suggested that the relationship between nonphotochemical quenching and irradiance was linear but only above some critical threshold On empirical grounds Holm Hansen et al 2000 suggested a critical PAR threshold of 40 pmolm s in Antarctic waters Hersh and Leo 2012 a threshold of 100 pmolm s in Massachusetts coastal and estuarine waters and Kinkade et al 1999 a threshold of 200 pmolm s in the Arabian Sea For our samples 100 pmol m s71 represents the 0 923 quantile i e more than 92 of the samples are below this threshold for nonphotochemical quenching We won t deal further with the complications of nonphotochemical quenching here although it should be considered at some point for near surface transect samples and moored sensors These considerations imply that SPM temperature and salinit
83. nderstanding of ecosystem response and condition assessment However although event based sampling sounds promising having a boat and field crew on stand by would be prove costly and logistically challenging other than for targeted studies SFEI staff are currently developing a web interface for visualizing continuous data from NMSP sensors and from collaborators sensors at a number of sites throughout the Bay The major emphasis of this effort is on developing a tool that allows easy access to and meaningful visualization of continuous datasets multiple sites multiple parameters multiple years that are managed by different entities across the Bay see Figure 2 1 for the range of potential sites Both past data multiple years and real time data will be viewable using this tool with real time data retrieved and appended to the records from real time sites Customized notifications could be built into the tool e g email notifications when a sensor fails or when a bloom begins Thus there will be a powerful tool available to manage and utilize real time data to improve program efficiency 33 5 0 Year 1 Data Interpretation While the primary goals in Phase 1 of the nutrient moored sensor program are related to building a solid program foundation the Year 1 data are already contributing to an improved understanding of ecosystem condition and will help us identify priorities for program development in Year 2 This section focuses on a su
84. ner 10 AU OBS volts a Neve al Turner 10 AU Chl a fi volts 2 11 2014 2 11 2014 0 90 b d D 2 24 2014 201021 2 24 2014 0 92 20140311 3 11 2014 60 3 11 2014 0 807 4 15 2014 5 4 15 2014 0 90 35 z Vane Rae Turbidity FNU 4 23 2014 Zao 412312014 0 92 A eS EbHCty MENU SA 40 5 13 2014 9 5 13 2014 0 89 g2 Re 20 XxX 20 6 6 2014 0 69 6 6 2014 1some data quality issue with pa 18 P Ta a ss 1 EXO turbidity probe 0 span removed r 0 2 approx 30 values of gt 200 10 Turner 10 AU oon volts 30 with these included 5 10 NTU removed r 0 29 with 2 portion of high turbidity Turner 10 AU Chl a fl volts shown in figure 3 2d is removed from regression r 0 44 with these included these included Figure 3 2 Comparison of turbidity and chl a fl data from the EXO2 and a Turner 10 AU fluorometer nephlometer during flow through deployment aboard R V Polaris cruises The Turner instrument reports optical backscatter OBS and fluorescence in raw voltage and the EXO2 reports turbidity in FNU and chl a fluorescence in RFU and for this reason only r results and not the full regression are given in the tables Scatterplots for a select number of individual cruises are shown a d Turbidity explain some of the scatter in the chl fl plots as indicated by color in the chl fl plots c d 3 2 In situ Sensor Calibration and Uncertainty Moored sensor data is valuable because it can be obtained at high frequency and can be much
85. ng and viewing data and replacing batteries as needed ii Performing measurements before and after cleaning probe heads in buckets of identical site water to assess the magnitude of 24 biofouling iii Removing remaining growth from sensor body and carriage iv Performing measurements in standard solutions of known quantity to assess probe drift and where necessary recalibrating and or replacing failed probes vi Reprograming and redeploying the instrument and vii Collecting and filtering a discrete chl a sample As the biofouling effect on data quality became more apparent we iteratively implemented additional measures to decrease the fouling impact The main observations related to maintenance based on Year 1 NMSP experience include e When only routine maintenance is required Year 1 experience suggests that 2 stations can be maintained per day assuming the stations are located in close proximity Maintenance frequency of approximately once per month during Year 1 appeared to be sufficient during colder low growth times of the year e g November March e During warmer months June October and perhaps starting as early as April May biofouling occurred quickly on and around the instrument package This was especially true at Dumbarton Bridge Figure 4 3 where growth of hydroids was the biggest problem e While in general low growth did not appear to impact data quality extensive biofouling resulted in highly compromised data qua
86. obial respiration rates The Dumbarton continuous data indicate that DO frequently drops by as much as 2 3 mg L on the outgoing tide during late summer fall and spring which is evident in the fine scale variability in the year long record and more clearly in the zoomed views from September 2013 and May June 2014 Figure 5 2b Because USGS cruises over the past 20 years have tended to sample in Lower South Bay at high tide this lower DO signal has been missed by most of that sampling Given the range of DO observed on a single tidal cycle especially considering that it dips near or below the current DO criteria for SFB 5 mg L high frequency data provides valuable insights that are missed by discrete sampling 35 StationNumber Ld si wo JW2013 O O23 Jan2014 apr20fa _ O Jul 2014 Li ea Sep 07 Sep 09 Sep 11 Sep 13 Sep 15 Sep 17 Sep 19 Figure 5 2 A comparison of EXO2 DO mg L values from Dumbarton Bridge with discrete lab analyzed DO samples taken at the 3 nearest stations to Dumbarton Bridge over all of Year 1 a The EXO2 shows that DO can dip by 2 3 mg L at low tide b which may be missed by the discrete samples taken by the R V Polaris which frequently samples at high tide EXO2 DO mg L data needed to be estimated during times during T and or SpC probe malfunction see Section 4 1 3 as shown in green These estimates are thought to be 1 mg L uncertain Outliers have been removed and servicing dates are indic
87. ogger modem 435 227 9100 provider U S Coast Guard Call when you arrive at 415 399 3451 bridge sites California Highway Patrol Call when you arrive at 510 286 6920 bridge sites Kurt Weidich USGS technician 916 698 7510 c 916 278 3065 w Paul Buchanan USGS technician 916 278 3121 w Amber Powell USGS technician 916 278 3060 w A 29 EI Supplies Table A 3 Satlantic Shape Products Whatman Thermo Scientific McMaster Carr Berkeley Plumbing Supply Walter Mork Ashby Plumbing Supply TAP Plastics Buoyant foam EXO2 manufacturer SUNA manufacturer for calibration standards 25mm glass fiber filters grade GF F Thermo Scientific Nalgene Polyolefin Pressure Sensitive Labels for chl a filters polyolefin 80 labels sheet L x W 0 5 x 1 75 in 1700 Brannum Ln Yellow Springs OH 45387 937 767 7241 Satlantic LP Richmond Terminal Pier 9 3481 North Marginal Road Halifax NS B3K 5X8 CANADA 902 492 4780 1127 57th Ave Oakland CA 94621 510 534 1186 www shapeproduct com ordered from Sigma Aldrich http www sigmaaldrich com unite d states html ordered from Fisher Scientific http www fishersci com ecomm s ervilet home specialty hardware http www mcmaster com PVC materials for sonde carriages Copper sheeting Copper pipe 2160 Dwight Way Berkeley CA 510 841 0883 2418 6th St Berkeley CA 510 845 0992 1000 Ashby Ave Berkeley CA 94710 510 843 6652
88. om 31 Depth m SpCond uS cm Turb FNU fDOM RFU kod g py N Ga O o 9 Q N D E 2 i 2 pg N 8 Q R g 5 p S fe amp 5 wo S Qo e 8 8 g g 7e HD 8 2 8 Q s m 8 N oO Q F O N o 2 8 O gt Q E o oO m Q kng N o o G T T T T T T T T T T i Sep Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Figure 4 5 Provisional Year 1 data for Alviso Slough Outliers have been removed and servicing dates are indicated by vertical dashed lines Data was omitted when lost due to power failure p or probe malfunction m When T probe was down T corrections for chl turbidity DO and fDOM were estimated by the method described in Section 4 1 3 T and or SpC probe malfunction also interfered with accurate DO mg L measurements and was estimated by the method described in Section 4 1 3 and shown in green because of their potential uncertainty pH and phycocyanin fluorescence were not analyzed in detail in Year 1 Depth is depth of the instrument below water surface not total water depth to channel bottom 32 4 3 Value of Real time Access to Data The technology for telemetering data at regular intervals e g hourly and enabling real time access to moored sensor data is readily available It is also fairly inexpensive 3000 for hardware per site plus data transmission fees Real time transmission is in place for the NMSP sensor at
89. or program SFE purchased 4 YSI EXO2 multisensor sondes one SUNA v2 nitrate sensor and telemetry equipment datalogger modem antennae Despite successful laboratory testing prior to deployment our SUNA v2 field deployment was complicated by what we believe to be power supply issues and the SUNA was pulled from the field one month after deployment Datalogger programming and telemetry set up was performed mainly by our USGS colleagues who have prior experience with these technologies Therefore we will focus this manual on the EXO2 with potential updates in the future as SFEI becomes proficient with these other instruments A 3 Equipment Several different multi sensor sondes were considered but we went with the EXO2 because it measures all analytes of interest and its lower cost gave us greater potential for field experiments involving multiple sensors during the pilot program EXO2 is a multi sensor sonde that can accommodate up to 6 probes plus a wiper and pressure transducer It has internal programming and datalogging up to 512 MB There is also an auxiliary port that can be used connect the EXO2 to other YSI instrumentation The EXO2 is always reporting in standard time Sensor body AE battery com partn ent sensor hardware Portplug USB adapter for com puter connection Auxiliary Cable connecter USA or field cable Turhidity Battery compartment Fig A 1 Important features of the EX
90. orometric probes fDOM chl a BGA using MilliQ water e For all fluorometric calibrations wrap calibration cup in something opaque towel black plastic bag during calibration Fill calibration cup to the bottom line with MilliQ water Navigate to the Calibrate menu and navigate to Chlorophyll RFU Select 1 point calibration with the standard value set to 0 00 and select Start Cal e Wait for the data to stabilize If the pre calibration value is lt 0 05 RFU there is no need to recalibrate and you should select Exit If the pre calibration value is gt 0 05 RFU hit Apply and then complete Step 4 Perform a 2 point calibration on turbidity probes using MilliQ water 0 FNU and the prepared turbidity standard e Wrap the plastic cup in something opaque during calibration e Fill calibration cup to the bottom line with MilliQ water or use the same water that is already in the cup from fluorometric calibrations Navigate to the Calibrate menu and navigate to Turbidity FNU Select 2 point calibration Enter the first point as 0 00 FNU and the second point as whatever the value is of the turbidity standard you created Select Start Cal Wait for the data to stabilize If the pre calibration value is lt 0 05 FNU there is no need to recalibrate and you should select Exit If the pre calibration value is gt 0 5 FNU hit Apply and then Proceed Empty the MilliQ water and rinse three times
91. ownloaded from the data query site on 2013 08 31 Total http sfbay wr usgs gov access wqdata http sfbay wr usgs gov access wqdata archive tabldescrip html 3http sfbay wr usgs gov access wqdata query solar radiation is available from the California Irrigation Management Information System recorded by a pyranometer at 2 m above ground level We used the hourly mean irradiance W m for Union City The same fluorometer Turner Designs Cyclops 7 Chlorophyll Sensor and settings have been used since 2005 except for 2013 02 26 at stations 30 through 34 Our analysis therefore begins with 2005 data and we ignored these few exceptional cases resulting in a total of 66941 water sample records from 2005 01 11 to 2013 07 23 Of these records 3579 include extracted chlorophyll a measurements Fifteen stations each with a minimum of 100 ex tracted chlorophyll a measurements since 2005 contribute most of the data in the analysis Figure 1 I i 122 122 122 122 122 longitude Figure 1 Stations along the USGS sampling transect with gt 100 extracted chlorophyll a measurements since 2005 During a single cruise calibration samples were typically taken in the surface layer at nttp wwwcimis water ca gov cimis data jsp 3 2 m and near bottom Figure 2 186 samples deeper than 25 m not shown These direct chlorophyll a measurements ranged from 0 5 to 67 3 pg1 with a mean of 5 9 1500 S 1000 O O
92. pH and phycocyanin results will not be discussed in this report For the remainder of this document we refer to the entire EXO2 as the sensor and the individual equipment for measuring specific parameter as the probe s i Be s CRER EA ANATA UR Figure 2 4 YSI EXO2 multiparameter sensor a and Satlantic SUNA v2 nitrate sensor b deployed during year 1 of the NMSP There were some challenges with deploying the SUNA resulting in limited field data and therefore it will not be discussed in this report 13 Table 2 3 Sensor selection criteria Consideration What parameters can be measured and to what level of accuracy Equipment needs to measure a range of nutrient related parameters and needs to be robust ina saline turbid environment What is the capacity for data storage transmission Large data storage allows for longer deployments between servicing Easy integration with telemetry allows for real time data How resistant is the instrument to biofouling High bio fouling resistence improves data quality and reduces the frequency of servicing trips Does the equipment and necessary cables adapters fall within the 80 000 budget What equipment would best allow for comparisons to and eventual integration with existing moored sensor programs in SF Bay 14 conductivity SpC turbidity turb dissolved oxygen DO chlorophyll a chl a fluorescence fluorescent dissolved organic matter f DOM and nit
93. perature data are not shown due to a faulty sensor DO is presented as sat because SFEI T SpC probes were malfunctioning and co located USGS SacSed T SpC data was unavailable and therefore DO mg L could not be reported Between high and low tide DO saturation varies over a range of 30 90 minima at low tide while turbidity and chl a fluorescence vary by a factor of 4 5 maxima around low tide Based on the estimated T and SpC at this site the DO concentration at 100 saturation corresponds to 8 mg L Using this approximation DO concentrations decrease to well below the Basin Plan standard of 5 mg L Although not shown here when evaluating DO data at Alviso over longer periods of time it becomes evident that the DO dips are most pronounced during neap tides in spring summer During those periods DO at Alviso drops to and remains at 2 3 mg L for 12 18 hours before returning to higher concentrations for several hours and this pattern repeats itself for several days The periodic occurrence of low DO around neap tides could be due to longer residence times of water within the slough i e not efficiently flushed out of because of weaker tides and or periodic stratification that may develop at this location around low tide SFEI 2014 732 The observed maxima in chl a sensor readings Figure 5 4d maxima 10 15 RFU co occurred with low tide and were 3 5 higher than those 37 observed at Dumbarton Bridge Figure 5 3d The sharp periodi
94. ple programs In its Year 1 pilot phase in addition to the NMSP sites we are incorporating data from 6 other sites operated by 2 distinct USGS programs The visualization tool allows the user to build interactive time series to explore relationships between analytes or between sites In Year 2 we hope to expand this tool to include more sites and additional programs incorporate real time data acquisition and include additional features based on desired functionality by researchers and managers in the region 43 7 0 References Cloern J E K A Hieb et al 2010 Biological communities in San Francisco Bay track large scale climate forcing over the North Pacific Geophysical Research Letters 37 Cloern J E and A D Jassby 2012 DRIVERS OF CHANGE IN ESTUARINE COASTAL ECOSYSTEMS DISCOVERIES FROM FOUR DECADES OF STUDY IN SAN FRANCISCO BAY Reviews of Geophysics 50 Cloern J E A D Jassby et al 2007 A cold phase of the East Pacific triggers new phytoplankton blooms in San Francisco Bay Proceedings of the National Academy of Sciences 104 47 18561 18565 Jassby A D 2014 Improving Estimates of chlorophyll from fluorescence in San Francisco Bay Prepared for the U S Geological Survery Marra J 1997 Analysis of diel variability in chlorophyll fluorescence Journal of Marine Research 55 4 767 784 SFEI 2014 Development Plan for the San Francisco Bay Nutrient Monitoring Program Richmond CA
95. profiling buoy configuration that raises and lowers the instrument through the water column and parks the instrument out of the water between profiels Although we may begin exploring these options in Year 2 any major shifts in equipment or configuration would likely wait until Year 3 or later 6 2 2 2 Data processing and management As SFEI expands its moored sensor network data processing and data management efforts will scale accordingly As much as possible the NMSP should develop and apply automated processes In Year 1 we found that correcting for outliers and modest sensor drift is fairly straightforward and the outlier removal step is already semi automated see Section 4 2 3 However addressing the effects of biofouling in time series is more complex and will likely be more manual and time consuming work In Year 2 we intend to continue refining these procedures while simultaneously working to curb fouling so the biofouling or drift issue becomes less pronounced SFEI has also begun developing best practices for data acquisition and storage SFEI has developed codes to autonomously pull real time data off our sensor and store in a database The goal is that as the NMSP network expands new data streams can be seamlessly plumbed into this existing database However not all our sites have the ability to be real time due to site constraints so this database also needs to be flexible enough to accept other means of input 6 2 3 Des
96. r South Bay cruises and collected data continuously to test EXO2 sensor response across a range of Bay conditions to compare EXO2 sensor response to USGS sensors also plumbed to the flow through system and to acquire simultaneous discrete samples across several parameters for sensor calibration e Bench top and initial field tests were carried out with the SUNA Despite successful benchtop testing of the SUNA prior to deployment we experienced power and communication issues during two trial field deployments at the Dumbarton Bridge Through discussions with the SUNA manufacturer and the USGS Sac biogeochemistry group who is successfully using the instrument at sites in the Delta overcoming these issues should be straightforward However given limited staff capacity in Year 1 we decided to focus instead on the YSI EXO2 deployments at Dumbarton and Alviso and shift work with the SUNA to Year 2 E Q E 3 O July Sept Nov Jan Mar May July 2013 2013 2013 2014 2014 2014 2014 Figure 2 5 Timeline of Year 1 NMSP activities across the three sites and aboard R V Polaris cruises Because of the limited field deployment at San Mateo Bridge and of the SUNA at Dumbarton Bridge these results will not be discussed in this report Dots show approximate timing of servicing trips Multiple dots indicate that a servicing trip spanned across several days 3 Sensor In situ Calibration and Uncertainty In Year 1 we evaluated sensor performance
97. rate NO3 in one integrated data stream Does not measure pH or phycocyanin fluorescence Suitable for estuarine environments More accurate for temperature T specific conductivity SpC and turbidity turb Good On board data storage and telemetry capabilities Couples telemetered data with a web visualization tool Good Uses copper and bleach injections in flow through tubes to discourage growth around sensors Fair LOBO system 77 000 WQM SUNA deployment cage datalogger modem cables Data storage visualization tool 5 500 Fair USGS Sac Biogeochemistry researchers are currently using the SUNA but not the entire LOBO package LOBO package EXO2 SUNA Good Measures temperature T specific Good Can measure all the same parameters as the LOBO plus pH and phycocyanin fluorescence PC More accurate for DO Suitable for estuarine environments Chl a and PC fluorescence accuracies not specified Fair On board data storage but not telemetry Requires external datalogger and model as well as the development of web data storage visualization Good A variety of copper accessories are available to discourage growth Integrated wiper keeps sensor faces clean Good EXO2 16 000 SUNA 26 000 Datalogger modem cables 5 000 Good USGS Sac researchers both groups are currently using YSI products some EXO2 and some older models Other monitoring programs in SF Bay i e DWR EMP ar
98. re highly correlated with most relationships having r 0 9 Figure 3 2 During the cruises when r2 was lower than 0 9 for either parameter the poor correspondence was typically due to a cluster of measurements and the other measurements were strongly correlated For example on 3 11 2014 the r2 for EXO2 and Turner 10 AU fluorometers was 0 44 and this was due primarily to a group of poorly correlated data that had high turbidity Figure 3 2d turbidity indicated by color which can interfere with fluorometer readings One possible explanation is that real fluorescence from chl a that was detected by the Turner 10 AU was underestimated by the EXO2 due to particles either scattering or 18 absorbing the fluoresced light It is also possible that something like a gas bubble or a particle interfered with one of the sensor s readings While the overall agreement is encouraging these occasional differences may require further investigation to better define the sources of uncertainty and conditions under which the sensors have substantially different responses a Cc 2013 09 26 2013 09 26 60 3 9 26 2013 9 26 2013 S gt w 2 asin 10 24 2013 0 90 Z 10 24 2013 Turbidity FNU 10 25 2013 240 10 25 2013 F po a 0 11 16 2013 8 11 16 2013 0 69 a 20 XxX 12 3 2013 20 12 3 2013 0 85 z 1 24 2014 1 24 2014 0 95 A 0 2 4 6 1 31 2014 10 Tu
99. respondence in the sensor responses over a wide range of conditions Figure 3 2 Figure 3 5a presents in situ discrete chl a concentration vs fluor for 1879 paired measurements that were collected over a wide range of conditions across all stations years and seasons from 2005 2013 This dataset is 10 fold larger than the Year 1 dataset collected for the EXO2 Figure 3 4 and shows a strong correlation r2 0 81 and chl a prediction error of approximately 5 ug L despite the widely varying conditions A multivariate analysis not shown suggests the relationship between chl a concentration and fluor differs significantly between some subembayments When stations in Lower South Bay and South Bay are considered individually or in groups of 2 3 adjacent stations the scatter in the chl a fluor relationship decreases considerably Figure 3 5b most notably for chl a values lt 10 ug L where the prediction error is reduced to 2 3 pg L Figure 3 5c Considering the wide range of conditions under which these samples were collected and the fact that no other predictors were included the fairly low prediction error 2 5 ug L is encouraging since this uncertainty is comparable to or considerably less than many other potential uncertainties inherent in assessing ecosystem condition or modeling ecosystem response A more detailed exploration of this dataset will be carried out as part of on going monitoring program development work to help identify future
100. rift or fouling it may be possible to systematically and reliably correct data from an individual probe We are continuing to investigate approaches and guidelines for robust correction of data from drift and fouling One basic procedure recommended by USGS Wagner et al 2006 for correcting a period of data is as follows i T To i Toul Voorn Vraw i Earift Te T Efoul Tem Trout where Veorr i the corrected probe value V aw i the raw probe value being corrected Earift the error due to drift difference between probe reading and standard value T the timestamp of the value being corrected T the first timestamp in the period being corrected Te the last timestamp in the period being corrected Erow the error due to fouling difference between clean and unclean readings Trou the timestamp of when fouling is thought to begin Figure 4 3 Typical fouling from biological growth on sensor carriage a and probe heads b during warm summer months In Year 1 SFEI tried several methods to reduce fouling and this will remain a high priority in Year 2 28 In Year 1 we experimented with applying such corrections to our data but it is not yet a systematic automated post processing step Probe drift occurs gradually over the course of a deployment and is fairly easy to correct for because its magnitude is relatively small it is easy to quantify by measuring clean probes against standards of known value and corrections can be appl
101. rview 3 Servicing and Maintenance 3 1 Pre servicing 3 1 1 Gather necessary field materials 3 1 2 Prepare turbidity standard 3 1 3 Calibrate the spare EXO2 3 2 Servicing trips 3 2 1 Retrieve sonde and download data 3 2 2 Assessing biofouling and sensor drift 3 2 3 Re deploying 3 2 4 Discrete sample collection 3 3 After servicing 3 3 1 Post servicing procedures 3 3 2 Long term equipment storage 4 Data management and validation 4 1 Data storage 4 2 Data validation 4 2 1 Fluorescence calibration 4 2 2 Sensor drift corrections 4 2 3 Biofouling corrections 5 Resources 5 1 Useful phone numbers 5 2 Supplies 5 3 Safety information 6 References A 2 E Project Description The moored sensor pilot program is intended as a multi year effort in which SFEI develops it s capacity to operate moored sensors but also simultaneously develops collaboration with existing moored sensor programs in San Francisco Bay and crystallizes the structure of the moored sensor sub program of the nutrient monitoring program see the main body of this report In this first year of the pilot program SFEI partnered with researchers from USGS Sacramento Sediment group USGS SacSed for equipment deployment and maintenance USGS already maintains several moored sensor sites throughout South SF Bay monitoring temperature conductivity turbidity and dissolved oxygen and were able to lend valuable expertise and field support In this first year of the moored sens
102. s in SFB While this latter observation is not surprising the amount of data for margin habitats is severely limited SFEI 2014b making it difficult to assess condition or understand processes there Given those data limitations identifying the best locations for moored sensor sites in margin habitats will need to proceed by incremental and iterative additions of stations based careful planning and a conceptual model of system behavior 6 1 1 2 Vertical spacing As noted in Section 4 1 1 sensors at Dumbarton Bridge and Alviso Slough reside at fixed elevations above the bottom and therefore are under variable depths of water depending on tidal stage How much of the observed variations in water quality parameters that occurs at tidal frequencies is due to water masses moving horizontally laterally or longitudinally with the tide and how much is due to the sensor pass through vertical gradients as its relative position in the water column changes What is the best depth to place sensors On the one hand it is desirable for sensors to be positioned in the water column to capture the important processes occurring in the photic zone which in SF Bay extends only to a depth of 1m On the other hand in order to allow for estimates of average conditions and for use in mass flux estimates e g as a function of tides data representative of average conditions throughout the water column are needed Lastly if low DO in bottom waters is
103. s of the relative standard errors and decreasing the bias at low values 16 N wo Oo oO I I chlorophyll standard deviation I fluorescence Figure 10 Standard deviation of extracted chlorophyll a versus fluorescence The data have been aggregated into bins of 0 1 fluorescence units Above a fluorescence value of 2 n lt 5 and the standard deviation values are not reliable 32 D I chlorophyll a ug i I I 0 2 0 4 0 8 1 6 fluorescence Figure 11 Two least squares regressions of chlorophyll a versus fluorescence for 2007 04 03 Solid line weighted dashed line unweighted 17 Table 4 Relative errors of prediction for 10 smallest chlorophyll values measured on 2007 04 03 with and without weighted regression chlorophyll unweighted weighted 2 60 0 36 0 14 2 70 0 31 0 10 2 80 0 40 0 21 2 90 0 35 0 17 3 20 0 59 0 46 4 60 0 49 0 43 4 90 0 52 0 48 5 40 0 16 0 10 5 80 0 08 0 03 6 40 0 10 0 06 7 Discussion and conclusions When is it important to reduce prediction error Accurate estimates of chlorophyll a are useful for assessing the state of an ecosystem and perhaps its compliance with regulatory thresholds In most cases the reduction in prediction error from using a secondary predictor is not that great 10 20 of median chlorophyll a not much different from uncertainties in individual measurements when all aspects of sampling and analyzing are taken into account And the absolu
104. s the ordinary optimism Efron Gong bootstrap Efron and Tibshirani 1993 Here we make bootstrap estimates of a calibration s coefficient of determination R and root mean square error RMSE a measure of prediction error n RMSE e pi n i 1 where y are the observed and y the predicted values and n is the number of samples RMSE is on the same scale as our predicted variable chlorophyll a pg17 and so can be thought of as the typical prediction error R and RMSE are intimately related RMSE describes how much of the variability was not accounted for by the regression R simply describes the remainder how much was accounted for but as a fraction of the total variability They give us the same picture but from different perspectives We used the validate function in the rms package for R to make the calculations Harrell 2013 The default boot method for validate is the optimism bootstrap mentioned above We calculated R and RMSE for each cruise date since 2005 To make sure all planned comparisons were balanced we used only those samples for which chlorophyll a SPM salin ity and temperature were simultaneously available Also only cruises with at least 10 samples for measured chlorophyll a were included smaller bootstrap samples can lead to unreliable results Chernick 1999 A total of 162 cruise dates met the criteria The results when fluorescence alone was used as a predictor are displa
105. sampling and ancillary measurements that may minimize chl a concentration prediction error 22 w All stations 2005 2013 b Station 32 2005 2013 J L J i J L 10 20 30 40 50 60 10 20 30 40 50 60 L Discrete chl a ug L Discrete chl a ug L i 0 1 0 f Chl a fl Volts Figure 3 5 Surface chl a fl values Turner 10 AU fluorometer in volts compared to simultaneous discrete lab analyzed samples taken aboard the R V Polaris from 2005 2013 at all stations a and just Station 32 nearest to the Dumbarton Bridge b c Red lines show the 95 confidence bands on the prediction error Focusing just on Station 32 reduces scatter particularly when considering samples where chl a lt 10 pg L prediction error band reduced from 5 ug L to 2 3 pg L 4 Operation Maintenance and Data Management Building capacity for operating and maintaining moored sensors in SFB and optimizing the effort required to sustainably support a program are high priorities for Phase 1 of the NMSP development and were major foci of Year 1 activities This section describes activities and observations from Year 1 related to moored sensor operation maintenance and data management 4 1 Sensor Operation and Maintenance 4 1 1 Sensor Deployment The Dumbarton and Alviso Slough sensors were deployed in the water column at fixed elevations above the bottom and were under variable
106. several ways e Compared the EXO2 output to the response of other commonly used sensors during side by side deployments Section 3 1 e Developed calibration curves for EXO2 probes using environmental samples and evaluated the goodness of fit Section 3 2 16 e Further analysis of the relationship between in situ chl a fluorescence and lab analyzed concentration in a 7 year 1800 sample dataset from SFB to assess the level of precision or prediction error that can ultimately be expected for NMSP chl a readings as more samples are collected Section 3 2 2 3 1 Comparison of EXO2 with co deployed sensors During Year 1 of the moored sensor program we had several opportunities for side by side comparisons of EXO2 with other continuous monitoring equipment already in use in SFB While this comparison is different than comparing an in situ measurement with a true lab analyzed value it provides a means of assessing the response and precision of EXO2 sensors relative to other instrumentation that has been widely used in the Bay over the past 20 years allowing us to assess the degree to which data collected in different areas in the Bay and over time can be compared in terms of relative response 3 1 1 Stationary In situ Comparison At the Alviso Slough site the EXO2 was deployed side by side with another YSI instrument 6920 model multi parameter sensor from USGS SacSed which measures several of the same parameters T SpC turb and
107. somewhat delicate around the tips of the probes particularly pH Use the small black plastic brush and the syringe to clean the ports on the T C probe Use the syringe to clean the depth port Wait to clean the carriage until after the post cleaning values are reported to as to keep the two buckets of water as identical as possible e Every so often about 3 4 months check the probe connections If needed replace o rings and reapply Krytox grease e After cleaning put the EXO2 in the second bucket and record post cleaning values using the Dashboard Calibration checks After cleaning the instrument calibrate each probe in the similar manner described in Section 3 1 3 The only addition to these procedures is that pre and post calibration values should be recorded on the field sheet If a probe appears to not be working properly A 19 1 Insepct and re clean the probe if necessarily particularly SpC port 2 Check probe connections 3 Swap the malfunctioning probe with another to determine if it s an issue with the probe the port or both If the probe is still malfunctioning swap it with a probe on the spare and note serial numbers of removed and installed probes A 20 Re deploying After the sensor is cleaned and calibrated it is ready to be returned to the water If battery volatge is lt 5 5 and next servicing won t be for 3 4 weeks replace batteries unscrew the battery cap with the plastic wrench replace batteries
108. te errors of around 1 pg are also not large compared to the chlorophyll a thresholds that might be of concern a minimum of 10 pg1 for zooplankton food limitation perhaps more for eutrophication The situation is different however for understanding how the estuary functions For example winter chlorophyll a minima are typically on the order of 1 pg1 and these have perhaps doubled in the past 15 years These changes are important because they imply a doubling of the photosynthetic energy and organic carbon input into the food web in winter a change of potentially great importance to the zooplankton benthos and higher trophic levels A prediction error of 1 pgl is a problem in trying to identify and understand these changes In these circumstances reducing prediction error is worthwhile but weighted least squares regression is probably more effective and reliable than a second predictor Given the drawbacks of a secondary predictor relatively low benefit for high chlorophyll a values less desirable than weighted regression for low chlorophyll a values difficulty of guaranteeing decreased prediction error lack of a clear cut causal basis often too small sample size perhaps it is best to avoid them altogether Currently the effort is better spent on identifying relatively homogeneous subregions with respect to chlorophyll a fluorescence ratios and using weighted regression when warranted We are likely to make the biggest strides
109. tesssssscgdacceseesiececasceateeduecsaceagstsaeseaittsuerednccentessttaacnsactastenivccieecss 29 4 3 Value of Real time Access to Datta essesessesssssssssssersesssessstesnsessseessessseerseersessnsessseensesneesaterseersnersneesneesnseenaes 33 5 0 Year 1 Data Interpretation ou cesseessesesssssesstecseesseeseesseeseeeneesseeseessecaeesseeseeeseessecstesseesneeseeateeneeseestesaeeateaneesteeaeeaseens 34 6 0 Main Observations and Priorities for On going Work eesseesessesssesseecstesseeseesseesteeneeseesteeseentesneesteeatenseens 39 6 1 Summary and Main observations from Year 1 eesessseessesssecstesseeseesseeseeneesseeseeeseestecneesseeseeeaeentesseesteeatenseens 39 6 2 Priorities for On going WOK ees esseessesssecseesneeseesstestecneesseesesseeseesseesseeseesseeseesneeseesseesteeneesseesteeaeenteaseesneeatenseens 40 6 2 1 Identify highest priority sites and analytes for future sensor placement 40 6 2 2 Refine maintenance and data management procedures eesseesessecseeceecstesseeseeeseeseesstesseesteeteateens 42 6 2 3 Design investigations to further constrain our understanding of sensor accuracy 42 6 2 4 Strengthen collaboration across PrOQramS eesseessescsesssecseeseeseecseesteeseeseeeseeseeatesseeseeeateseeatesseesteeaeenseens 43 Li OQ RELET ENCES cscs Seceaecdacstaaeheesct n sats suid asvensnatsanlt nde aseeteartashivuiner a cba aac 44 Appendix A Appendix B Figures Figure 2 1 Locations of existing moored sensors in S
110. th only 10 30 samples available for each cruise limiting both the number of predictors that could be added to the model and the ability to detect an effect 21 lt y 3 65x 1 08 r2 0 67 Discrete chl a ug L 8 EXO2 Chl a FI RFU Figure 3 4 Comparison of EXO2 values and simultaneous discrete lab analyzed samples for chl a Red lines show the 95 confidence bands on the prediction error Adding EXO2 turbidity measurements as a secondary predictor in a multivariate linear regression improved r to 0 72 While the size of the Year 1 in situ calibration dataset makes it insufficient for assessing the prediction error of chl a it is nonetheless desirable to begin developing some a priori sense of the degree of confidence that can eventually be placed in moored sensor chl a concentration estimates data to help in determining the justifiable level of effort and expense that should go toward establishing moored stations We began quantitatively exploring the issue of minimizing in situ chl a prediction error by using the R V Polaris discrete chl a concentration and shipboard fluorescence dataset from 2005 2013 Although this is a retrospective analysis without the benefit of the EXO2 to a first approximation we expect that the EXO2 sensor would respond similarly to the Polaris fluorometer In fact data from the 2013 2014 side by side comparisons between the Turner 10 AU fluorometer and EXO2 fluorometer confirm the strong cor
111. the top line Navigate to the Dashboard and note what the SpC reading is If error is gt 3 repeat Step 5 recalibrate and recheck Step 6 Perform a 2 point pH calibration using pH 7 and 10 buffer solutions Rinse the calibration cup 3x with pH 7 buffer Fill calibration cup to the bottom line with pH 7 buffer Navigate to the Calibrate menu and navigate to pH Select 2 point calibration Enter the pH 7 as the first point and pH 10 as the second Select Start Cal Wait for the data to stabilize If the pre calibration error is lt 0 2 pH units there is no need to recalibrate and you should select Exit If the pre calibration error is gt 0 2 pH units hit Apply and then Proceed Rinse the calibration cup 3x with pH 10 buffer Fill calibration cup to the bottom line with pH 10 buffer Wait for the data to stabilize Once stable hit Apply and then Complete Step 7 Perform a 1 point DO calibration A 15 Rise the calibration cup 3x with MilliQ or DI water Fill the calibration cup with approximately 1 2 of room temperature MilliQ or DI water Let equilibrate 5 10 minutes Navigate to the Calibrate menu and navigate to DO Select 1 point Air Saturated calibration Select Start Cal Wait for the data to stabilize If the pre calibration error is lt 3 there is no need to recalibrate and you should select Exit If the pre calibration error is gt 3 hit
112. ther one like community composition Table 3 Within cruise correlations of secondary predictors Predictors Min I1st Qu Median Mean 3rd Qu Max SPM salinity 0 00 0 18 0 31 0 38 0 58 0 99 SPM temperature 0 00 0 19 0 41 0 39 0 59 0 90 salinity temperature 0 01 0 42 0 69 0 64 0 90 1 00 The coefficient signs in chlorophyll a calibration equations give us some clues about whether the corresponding predictors are having a direct effect or are merely correlates In the case of SPM the signs are almost all negative 95 in cruises for which both coefficients for in vivo fluorescence and SPM are statistically significant A little rearranging of the calibration equation shows that SPM therefore has a positive effect on fluorescence almost all the time This is consistent with chlorophyll degradation products such as pheopigments being the causal mechanism associated with SPM Irigoien and Castel 1997 perhaps as detritus resuspended with other SPM from the benthic environment Similarly in the case of salinity the signs are almost all positive 88 when both coefficients are significant Salinity thus has a negative effect on fluorescence which is consistent with an effect due to fluorescent DOM originating in watershed soils and entering with inflow from the Delta Temperature on the other hand shows a split in coefficient signs with 65 negative and 35 positive which is less supportive of a direct effect by temperature 5 5 When
113. ton pheophytin in the water column can artificially amplify or quench the fluorescence signal of chl a leading to an over or under estimation of chlorophyll concentration The amount of fluorescence per unit chlorophyll a can change depending on the physiological state of the phytoplankton in response to temperature or light availability or the phytoplankton community composition In this analysis we focus mainly on the effects of intereferents in the water column particularly suspended sediment Characterizing the variability in fluorescence per unit chlorophyll will be a priority in Year 2 of the program particularly diel cycles in fluor chl a due to quenching Marra 1997 While the correlation between discrete chl a measurements and EXO2 chl a fluor signal from R V Polaris cruises is highly significant p lt lt 0 001 Figure 3 4 the relationship based on Year 1 data does exhibit more scatter than turbidity and DO Figure 3 3 at least on a percentage basis within this chl a concentration range The 150 samples used for the calibration relationship were obtained from multiple sites across the entire Bay and their collection was distributed over nearly one year The abundance of interferents such as suspended sediment dissolved organic matter and phaeophytin vary both spatially and seasonally in San Francisco Bay T light levels and phytoplankton community composition also vary seasonally and spatially In addition most of the samples
114. ts associated with moored sensors can be non trivial due to initial set up of moored sensor stations on going maintenance of sensor packages sensor calibration and data management In addition moored sensors are in general not a total replacement for ship based sampling for several reasons only a limited number of analytes can be reliably measured with existing sensors data is often needed at locations between moored sensor sites and regular calibration and corroboration with discrete samples through ship based sampling remains necessary Lastly for some sensors the relationship between values measured by in situ sensors and the true value can vary due to interferents and measurement uncertainty needs to be considered when determining if program goals can be adequately achieved through a moored sensor approach The overarching goal of the NMSP is to address data needs for ecosystem assessment and model calibration and to do so through a sustainable program design that maximizes efficacy and cost effectiveness Table 2 2 presents a set of more specific NMSP goals and key questions identified to guide NMSP development Table 2 2 The program can roughly be divided into two phases Phase 1 focused on program development and site selection and Phase 2 focused on program implementation and the application of continuous data within other Nutrient Strategy elements e g modeling to inform management decisions During Year 1 the majority of e
115. turbidity signal to serve as an indicator of biological fouling One potential explanation for the early detection of fouling by the turbidity probe is that the mechanical wiper actually disturbs fine particulates that have accumulated in any biological growth near the probes and the turbidity probe detects those particles Although the resuspension of particles had a large effect on the turbidity signal they often did not have a major impact on other probes It may be possible to correct for some noise from biological fouling during data post processing but at some point the fouling effect became too large or erratic that it could not be readily corrected 4 2 4 Provisional Year 1 dataset The cleaned provisional Year 1 datatset for the Dumbarton and Alviso sites are presented in Figures 4 4 and 4 5 This provisional dataset has undergone the following post processing 1 Outliers removed via automated process described in Section 4 2 3 2 Data from periods of heavy fouling removed correction procedures still in development 3 T SpC data removed during periods of malfunction use of other T data for temperature related corrections in probe response and substitution of estimated T and SpC data for calculating DO concentration mg L as described in Section 4 1 3 1 For example if the extent of growth is such that the sensors are in fact semi encapsulated within a microenvironment they are not actually measuring conditions in the surround
116. ues of Water Resources Investigations of the United States Geological Survey vol 4 Hydrological Analysis and Interpretation chap A3 pp 1 524 U S Geological Survey Hersh D and W Leo 2012 A New Calibration Method for in situ Fluorescence Report 2012 06 11 pp Massachusetts Water Resources Authority Boston Holm Hansen O A F Amos and C D Hewes 2000 Reliability of estimating chlorophyll a concentrations in Antarctic waters by measurement of in situ chlorophyll a fluorescence Marine Ecology Progress Series 196 103 110 doi 10 3354 meps196103 Irigoien X and J Castel 1997 Light limitation and distribution of chlorophyll pigments in a highly turbid estuary the Gironde SW France Estuarine Coastal and Shelf Science 44 4 507 517 20 Kinkade C J Marra T Dickey C Langdon D Sigurdson and R Weller 1999 Diel bio optical variability observed from moored sensors in the Arabian Sea Deep Sea Research Part IT Topical Studies in Oceanography 46 8 9 1813 1831 doi 10 1016 S0967 0645 99 00045 4 Krause G H and E Weis 1991 Chlorophyll fluorescence and photosynthesis the basics Annual Review of Plant Biology 42 313 349 Lorenzen C J 1966 A method for the continuous measurement of in vivo chlorophyll concentration Deep Sea Research 13 223 227 Marra J 1997 Analysis of diel variability in chlorophyll fluorescence Journal of Marine Research 55 4 767 784 doi 10 13
117. utput What potential interferences on fluorometer results are most likely in SFB and how to best infer accurate concentration from fluorometers What is an acceptable level of uncertainty or unexplained variance 4 Establish collaboration with What moored sensors currently exist how well are they distributed other moored sensor programs and what is the relative completeness of each site What common elements would be needed to integrate existing sensors from multiple programs into one network How does the cost of integrating existing programs together compare to developing an SFEI moored sensor program at similar sites 5 Identify NMSP structure What is the optimal combination of ship based and moored stations Sections that along with ship based What spatial distribution lateral longitudinal vertical is needed to 4 1 6 2 monitoring addresses the sufficiently capture major features of bloom dynamics monitoring and data collection What parameters are most important to measure in terms of their needs for the Nutrient Strategy quantitative influence on predictions or model interpretations 6 Use moored sensor data to What factors influence the onset and termination of a bloom Section address priority science What frequency magnitude and duration of a bloom is possible 5 questions and data gaps 7 Use moored sensor data to What time series required for model calibration can be accurately calibrate validate wat
118. value reflected in data shown in Figure 4 4 and 4 5 We believe 1hr window is narrow enough to identify and remove outliers but not unintentionally remove any meaningful real but sharp changes in a parameter which one would expect to be accompanied by one or more comparable magnitude measurements within a 1 27 hr window However as we become more familiar with the system we could revise our code to account for more or less variability as is typical of the system We have developed code that automates the outlier identification and removal process As the NMSP moves toward web based hosting of real time data for visualization and download that code can be integrated into the data flow path and run periodically e g on the newest four hours of data before that data is posted to the web queried database Probe output can also be affected by drift away from last calibration value and fouling Figure 4 3 During each site visit we take a series of measurements to quantify the effects of probe drift and fouling on sensor readings See Appendix A for detailed description of field servicing procedures The change due to fouling is determined by placing the instrument into buckets of identical site water immediately after removing from SFB and then again after cleaning The amount of change due to drift is determined by taking measurements in standards of known value and then recalibrating as necessary Depending on the magnitude of d
119. vigate to the Connections menu select Re scan and select EXO USB Adaptor xxxxx from the list and select Connect Navigate to the Deploy menu and select Stop Deployment Make sure the KOR EXO is pointing to the correct folder on the computer hard drive Navigate to the Data menu select Settings and edit the Default File Location to your preference Select Apply Select Transfer within the Data menu select the most recent file from the list and hit Selected It will download as a bin file After the data downloads visualize it by selecting View Export from within the Data menu and pointing KOR EXO to the file you just downloaded You can view each timepoint in a table a graph with KOR EXO or you can export to Excel will create a xlsx file Note any data irregularities on the field sheet e Delete the file from the EXO2 after confirming transfer to laptop by selecting transfer within the Data menu selecting the most recent file and selecting Delete A 18 Assessing biofouling and sensor drift Accuracy of sensor readings can be affected by biofouling and sensor drift since the last servicing visit We attempt to quantify these two sources of error on each servicing trip and retroactively apply corrections to the time series adapted from USGS standard methods in Wagner et al 2006 Assessing biofouling To assess the effects of biofouling we compare probe readings
120. with terminal facing away from the probes Confirm timezone is PST e Navigate to the Settings menu and select User e Make sure Local Time Zone is UTC 08 00 Pacific Time US amp Canada e Confirm the EXO2 is synced with the computer Navigate to the Settings menu and select Sonde and Update Time Make sure the Relative to PC option is enabled and select Apply Even if the computer reads PDT the sonde will always default to PST so the times may look an hour off in Mar Nov Set up programming Navigate to the the Deploy menu You can either Open an Template to load a saved program or Read Current Sonde Settings to edit the program currently on the Sonde e There are a few settings to adjust to confirm before deploying To do this click on this icon gt and check the following o Under the Basic tab confirm the sampling interval is 15 minutes the timezone is PST and the file prefix is appropriate for the site will help with file management later on If the time needs to be corrected see step 2 in Section 3 1 3 o For Dumbarton Bridge and other sites with real time under the SDI 12 tab confirm the parameters are in the following order important for proper telemetering of data Add parameters if needed For Dumbarton Bridge the SDI 12 address should be set to 2 1 Date mm dd yy 8 Turbidity FNU 15 fDOM Raw 2 Time hh mm ss 9 Turbidity RAW 16 ODO mg L
121. with the non zero turbidity standard pour a small amount in the cup put EXO2 in the cup and shake and then empty the cup by pouring over the sensors Pour the non zero turbidity standard to the bottom line Wait for the data to stabilize This may take a while for turbidity If data won t stabilize but values are fluctuated lt 2 NTU for approximately 5 minutes the proceed with calibration Once stable hit Apply and then Complete Step 5 Perform a 1 point calibration on SpCond probes using one of the following standards 50k 24 8k or 15k uS cm If you are at a more freshwater site use 15k or 24 8k The 50k is for the more saline sites You will check one other standard value after the 1 point calibration see below Rinse the calibration cup 3x with the desired standard value Fill calibration cup to the top line with chosen standard Navigate to the Calibrate menu and navigate to Specific Conductivity uS cm Select 1 point calibration Enter the standard value Select Start Cal Wait for the data to stabilize If the pre calibration error is lt 3 there is no need to recalibrate and you should select Exit If the pre calibration error is gt 3 hit Apply and then Complete KOR EXO only allows 1 point SpC calibrations but these probes have given us some difficulty during the first year So after completing the 1 point calibration rinse 3x with the other standard and then fill to
122. y running a program and logging data A 7 Servicing and Maintenance Standard servicing trips include cleaning instruments downloading files checking calibration and recalibrating probes as necessary taking discrete samples and re deploying the instruments If any additional activities are planned desired i e special investigations changing the deployment configuration it is best to discuss with USGS 2 3 weeks in advance of the next planned trip to determine if they can accommodate these activities and if so refine the field schedule as needed It is always recommended to have 2 SFEI staffers attend every servicing trip Even though USGS staff will be on site they are not expected to assist with SFEI activities When looking for staff to join on trips consider staff who are more comfortable in the field particularly on boats or staff who may need more billable hours A 8 EJ Pre servicing ka Gather necessary field materials EXO2 1 Spare EXO2 storage case 2 Laptop charger 3 Calibration supplies Bring 1L of each standard per site e pH standards 7 and 10 e Specific conductivity standards 15k 24 8k and 50k uS cm e Turbidity standard approximately 100 NTU e MilliQ organic free water for fluorometric probes and 0 NTU turbidity e Calibration cup Plastic sonde guard and e Plastic sonde guard calibration cup e Opaque plastic bag or towel 4 2 5 gallon plastic buckets 5 Cleaning supplies e large and small pl
123. y should be considered for possible inclusion as additional factors 5 2 Are secondary predictors useful To appreciate the importance of using bias corrected results consider the transects of 2005 03 18 and 2005 03 22 when the number of simultaneous measurements for chlorophyll a and SPM were n 10 and 22 respectively In the case of 2005 03 22 the linear regression of extracted chlorophyll a on in vivo fluorescence yields uncorrected estimates of R 0 74 and RMSE 0 94 Adding SPM as a second predictor variable increases R to 0 88 and decreases RMSE to 0 65 Apparently the inclusion of SPM as a predictor has merits Now examining the bootstrap estimates of these values we note that R again increases from a bias corrected 0 72 to a bias corrected 0 86 and RMSE decreases from a bias corrected 1 0 to a bias corrected 0 74 In other words the bias corrected estimates confirm the value of including SPM although the performance of both the 1 and 2 predictor calibration equations is less than suggested by the usual methods Turning now to the shorter transect of 2005 03 18 the usual regression method exhibits an increase in R from 0 88 to 0 90 and a decrease in RMSE from 0 91 to 0 81 when SPM is included again apparently supporting the inclusion 11 of SPM But this time the bias corrected R actually drops from 0 86 to 0 62 and the bias corrected RMSE increases from 1 1 to 1 6 pgl7 How often and under what circumstances does this
124. yed as empirical cumulative distribution functions CDFs for R and RMSE Figure 6 The mean R was 0 65 and the mean RMSE was 1 4 pglt Shttp stats stackexchange com questions 62576 R2 1 00 0 75 0 50 0 25 5 I i 0 00 0 25 0 50 0 75 1 00 RMSE empirical CDF z I 0 75 0 50 0 25 0 00 h 0 2 4 Figure 6 Empirical cumulative distribution functions of R top panel and RMSE bottom panel when fluorescence was the only predictor in the calibration equation for chlorophyll a Estimates based on the optimism bootstrap 5 Secondary predictors 5 1 Which factors to consider Which monitoring program variables affect in vivo fluorescence besides chlorophyll a The available variables available include pheophytin a dissolved oxygen suspended particulate matter SPM temperature salinity solar radiation and vertical light attenuation The pairs plot Figure 7 shows all pairwise plots in the lower triangle and corresponding abso lute correlations in the upper histograms are along the diagonal Text sizes for correlations are scaled by their fourth root to highlight the largest ones Five of the distributions fluor chl phe spm and ext are highly skewed Chlorophyll has the largest correlation with fluorescence as expected with pheophytin a distant second The correlations among SPM pheophytin and vertical attenuation are also notable The more variables we can exclude from

Nutrient Moored Sensor Program Year 1 Progress Update

Contents

Download Pdf Manuals

Related Search

Related Contents