Home

High resolution precipitation intensity

1. T E Weight g E E calculated T rainfall Intensity indicated rainfall e intensity 01 18 22 18 04 18 07 18 Figure 36 Comparison between calculated and indicated intensity during the different temperature runs With a better analyze of these jumps it appears that they occur when the heating mechanism of the climate chamber was set off Thus the noise was dramatically reduced because of the stop of the ventilation device See figure 37 This abrupt change of the environment probably disrupted the filter mechanism and leaded to wrong simulations So because these noise features are totally artificial and would not occur in the nature it is probably not a big problem that the rain gauge simulated wrong rainfall in this case Temperature weight dependence 4 9945 u 9940 B 2 9930 Noise 2 15 dl 1 9923 ___ weight g 05 9920 0 SES iia 9915 00 00 00 00 00 00 00 00 co O A N O 99 N oO oO oO Figure 37 Noise values during the temperature run Finally after these different experiments 1t seems that under all operative conditions the filter mechanism of the gauge are able to take into account this complex weight temperature variation and avoid thus the simulation of wrong rainfall 34 This temperature dependence seems thus to not represent a problem for questions relat
2. intern sensor 2 O Lo kl Generally this error should not be problematic in the sense that other experiments seems to indicate that the bias stays constant over the different temperature ranges On this way it should not lead to non optimal corrections made by the filter mechanism that accounts for temperature weight variations Indeed as demonstrated in the chapter 235 the temperature weight variation seems to stay constant among wide temperature ranges With this constant error the real threshold for engaging the heating ring amounts to 4 7 instead of 4 degrees Regarding at the energy consumption of the heating mechanism it could be worth to decrease this threshold In this experiment it was difficult to assess securely 1f the heating ring worked properly The ring stayed cold but the snow was not able to stay on this black part On the contrary on the white housing the snow accumulated a little bit 3 2 2 Temperature influence on the weight As already mentioned the temperature changes the resistance measured by the strain gauge and affect thus the weight measurements It is logical because the material of strain gauge presents a thermal expansion For more information concerning the effect of temperature on the strain measurement accuracy an interesting explanation can be found at the following address http zone ni com devzone cda tut p 1d 3432 toc0 The goal of this sect
3. 4 3 3 1 The short memory event Extending the first irregular intensity event 1t appears that the autocorrelation structure changes quite dramatically Indeed a memory effect s visible up to lag of 500 minutes even if almost all the values are below 0 5 This change is logical because now the rain is no more isolated and thus the memory effect of the whole rain process becomes more visible Logically these about 500 minutes of correlation effect correspond to the rain duration The smallest peaks in the decrease of the autocorrelation coefficient could attest of the correlation that occurred during the transit of one rain cluster of the storm 40 2 20 rainfalldepth mm intensity mm min Sample Autocorrelation 0 500 1000 1500 2000 26500 0 BE i 00 00 00 00 00 00 Minutes Figure 78 Left intensity and cumulated rainfall depth Right ACF 6l Looking at the variance among the different aggregation scales see figure 79 about the same behaviour as for the selected case s observed Indeed the slope seems to regularly slightly increase However for the highest aggregation scale a big decrease of the slope is available This increase could thus just be temporary Looking more carefully at the figure representing the variance pattern obtained by the different data set of Marani see figure 59 1t appears also that sometimes some temporary increase of the slope occurs This
4. 23 09 1089 715 2 0 171814 1089 864 1089 712 1089 712 23 7 1036 23 09 1089 715 2 0 1718t9 1089 806 1089 712 1089 712 23 7 0 066 10 36 23 09 1089 715 2 0 171824 1089 794 1089 712 1089 712 23 7 0 066 10 36 23 09 1089 715 gt 0 171829 1089 952 1089 712 1089 712 10 36 2309 1089715 1 0 171834 1306 681 1089 712 1089 712 23 7 0 066 1036 2309 1089715 1 0 171839 1306 762 1089 712 1089 712 23 7 0 066 10 36 2309 1089 715 1 0 171844 1306 75 1089 712 1089 712 23 7 0 066 1036 231 1089 715 1 o 171849 1306 751 1089 712 1089 712 23 7 0 066 1036 231 1089 715 1 0 171854 1306 835 1089 712 1089 712 23 7 0 066 1036 231 0897t8 1 o 171859 1306 779 1089 712 1089 712 23 7 0 066 10 36 o 6o 23 1 1089 718 1 o 171904 1306 823 1089 712 1089 712 23 7 0 066 1037 231 t0897t8 1 o 171909 1306 843 1089 712 1089 712 23 7 10 37 231 1089 718 1 0 171914 1806 843 1089 712 1089 712 23 7 0 066 1037 231 1089 718 1 o 171919 1306 831 1089 712 1089 712 23 7 0 066 10 37 231 1089 718 1 ol171924 1306 832 1089 712 1089 712 23 7 1037 2341 1089 718 1 o 171929 1306 802 1089 712 1089 712 23 7 0 066 10 37 231 1306732 1 0 171934 1806 838 1089 712 1089 712 23 7 0 058 1037
5. 231 1806782 1 o 171939 1306 803 1089 712 1089 712 23 7 0 058 1037 231 1806782 1 0 171944 1306 828 1089 712 1089 712 23 7 0 058 1087 231 1306 732 1 0 171949 1306 73 1089 712 1089 712 23 7 0 058 1037 231 1306 782 1 0 171954 1806 755 1089 712 1089 712 23 7 0 058 1037 231 13806782 1 o 171959 1306 868 1306 732 1306 732 23 7 0 124 1037 109 6ol 23 1 1306 732 1 o 172004 1306 821 1306 732 1306 732 0 124 The first column of this table represents the time of the related records In the text file the seconds are indicated Even if it s 5 seconds resolution data the rainfall intensity PR1M is indicated only each minute because of the time needed by the different algorithms to treat the raw data The rain duration RDIM in the third column indicates logically 60 seconds because no sensor for rainfall detection 1s installed with this gauge Otherwise it could detect when the rain began and indicates the duration of the rain during the minute The sensor temperature TW1 as well as the external temperature TA1 are also available each 5 seconds There are many measurements of the weight The raw weight WRAW presents logically the higher variations The Total Weight WABS is updated every 25 and 35 seconds As already mentioned it reacts with a delay of 1 minute The other indicated weights WBEGIN and WCOMP present a greater delay but stil
6. This success rate must be also analysed in relation with the demanded accuracy of the classification In this case it appears that the different wind speed classes are really close together Surely on the field such a precision is not needed It could thus maybe ameliorate in a further way these success rates Unfortunately this test period was too short and presented only low wind velocities It is a shame because these tests seemed to be able providing quite good results with more data and wider wind ranges The advantage of the chosen threshold selection mechanism is that it should be theoretically less sensitive to the possible changes in the wind speed noise relation For these low wind speeds it appeared that the noise increased quite regularly with the wind speed but only after a certain wind speed threshold Thus on this stage these tests provide results that will not be useful on the field because the observed wind velocities were too small for these 2 usable days of data Moreover this test are assumed valid for the empty gauge case The gauge will however never be totally empty because of the antifreeze liquid that will be present at least in winter These tests could be completed thanks to the next testing phase that will install the gauge near to existing meteoswiss stations These stations also provide wind measurements with 10 minute resolution However the anemometer is placed quite high in the air and probably the indi
7. at getparam apn gprs swisscom ch E Transmit CLEAR __SendFie CR CR LF Br RTS Macros Set Macros Mi M2 M3 M4 M5 MG M7 M8 M9 M10 M11 M12 at getparam gprs gt Send at last SIE Le Connected Rx 1085 Tx 554 Figure 16 Terminal settings for dialoguing with the data logger After the opening of the terminal the following window will open Check that this window presents the same settings than in the figure 16 Note that the DTR and the CTS buttons below right and above right have to be green The CTS button becomes right when the button connect is pressed And thus the connect button changes in disconnect It 1s not a problem if the CDM port is not the same Just keep the one that 1s automatically selected https www distrelec ch ishopWebFront catalog product do para node is DD 6787 and highlightNode 1s 67 1492 and id is 02 and series is 1 html 18 Trough different commands that have to be entered in the grey field at the bottom of the window and not in the white one 1t s possible to interact with the data logger In the operating mode some commands allow to dialogue with the data logger mainly to ask about its state The table 4 presents these commands Table 4 Some commands available in the operating mode Request Command to type Is 1t running at wopen if it is running wopen 1 otherwise wopen 0 Is it con
8. 3 Computation of different parameters computation of 5 minute resolution noise data for j 1 c noisemeanStreated j mean noisetreated jJ 1 5 1 5 J noisevarbtreated 3 var noisetreated jJ 1 5 1 5 3 end computation of direction and speed changes for k 2 c directionchange k WDtreated k WDtreated k 1 1f directionchange k gt 180 directionchange k 360 WDtreated k WDtreated k 1 end if directionchange k lt 180 directionchange k 360 WDtreated k 1 WDtreated xk end directionchange k abs directionchange k windchange k WStreated k WStreated k 1 end directionchange 1 0 windchange 1 0 OLS 6 computation of the different cross correlations crosscorWSnoisemean corrcoef WStreated noisemeanStreated crosscorWSnoisevar corrcoef WStreated noisevar5treated crosscorWDchangenoisemean corrcoef directionchange noisemeanStreated crosscorWDchangenoisevar corrcoef directionchange noisevar5Streated crosscorWSchangenoisemean corrcoef windchange noisemeanStreated crosscorWSchangenoisevar corrcoef windchange noisevar5treated 4 Best threshold selection procedure o definition of the different wind classes wind_range 0 5 1 1 5 69 control of the choosen wind classes if max wind_range gt max WStreated display attention the maximal wind range speed is greater than the maximal display
9. 2 4 E Weight E 0 50 9950 2 40 AR 9940 2 30 N 2 Temperature 20 9930 d 3 10 9920 gt F 0 9910 00 00 00 00 00 00 SS E gt Figure 31 Temperature weight dependence with two different initial weights 31 It appears that th s dependence presents sometimes quite strange behaviour In these two temperature runs the before noticed negative correlation was sometimes not any more observed This anomaly appears better looking at the figure 32 5 9940 9935 2 3 2 9930 D2 D 9925 0 9920 0 20 40 60 0 20 40 60 Temperature deg cels Temperature deg cels Figure 32 Temperature weight relation for the both initial weights Selecting only the data of this test presents a regular behaviour it appears that the initial weight plays an important role in regard to this temperature weight relation The bigger the initial weight the stronger the relation See figure 33 9935 yred 0 4031x 9941 2 5 0 6505 E yred 0 099x 6 062 R 0 792 3 9930 0 79 23 lt T 3 9925 o yblue 0 084x 4 0897 yblue 0 2993x 9931 5 R 0 9412 2 _ ee R 0 9761 0 20 40 60 0 20 40 60 Temperature deg cels Temperature deg cels Figure 33 Relation temperature weight for selected data of the both experiments This observation was con
10. also present The cable of the rain gauge arrives indeed to the multiconnector with 6 wires This 50m long cable is a prolongation of the original rain gauge cable Unfortunately the wires of this additional cable don t have the same colours as the original one The table 2 presents the different wires Figure 9 The multiconnector Left the different wires 6 coming from the gauge Table 2 The different wires coming from the rain gauge Negative heat power Vheat Negative data D Note that the additional cable contains a black wire that 1s empty and thus useless Other wires link different elements of the electronic systems The table 4 1 n the original user s guide v 3 1 describes these other wires 13 2 4 Data provided with a temporal resolution of 5 seconds This resolution 1s available when a computer is directly connected to the rain gauge see points 2 3 3 and 2 3 6 To collect and see in real time the data provided by the gauge with a temporal resolution of 5 seconds the program Universal Sensor Manager usm must be used It is located in the directory MPS tools usm ETH APS Note that the data is collected only when this program is running otherwise the data sent by the gauge will be neither collected nor stored With a double click on usm exe two different windows will be opened In the universal sensor manager one click on new This will open another window proposing a choice of
11. are not s gnificantly correlated with the noise mean or the noise variance Concerning the wind direction changes a weak negative correlation between the measurements s present An explanation could be that during low wind conditions the direction of wind is not very well defined leading thus to higher changes in direction These great direction changes are thus linked with small values of noise because of the very discrete feature of wind leading so to rather negative correlations On the contrary the wind speed presents a high positive correlation with the noise features Generally the noise mean presents a higher correlation with all the wind properties indicating thus that it 1s a better indicator n order to assess the wind influence on the gauge These results are very good regarding at the purpose of assessing the wind speed from the noise values The effect of the wind speed and wind direction changes will be interpreted indeed as negligible If a significant correlation would be present between the wind direction and speed changes it would be quite more difficult to assess the wind speed with the noise values Indeed t would introduce an uncertainty in the interpretation of the noise values because t would be impossible to separate the effect of these different factors only thanks to the indicated noise values However it doesn t necessarily signify that the in this case neglected factors don t have influence on the noise P
12. cable For more technical detail such as for example a complete description of the output the range of allowed input or the limit operating conditions all the characteristics of this element are presented here http www meanwelldirect co uk product DR 30 15 DR 30 15 default htm 2 3 2 MPS05 CPU Figure 6 The place where the sim card must be inserted It is the central processor unit It collects the measurements from the connected sensors and can send it for example via GPRS to a server It works thus as data logger and modem To send the data through GPRS a sim card must be inserted behind the data logger See figure 6 To separate the data logger from its metal bar support pull on the little black ring For further information concerning the sim card and the data logger see the part 2 6 1 or consult the file MPSOSCPU pdf that is located at TRWS manual MPSOSCPU logger 11 2 3 3 CONV RS232 422 It s an analogue to digital converter This device s necessary because the sensors produce analogue s gnals for example n the form of voltage These analogue s gnals must be first converted in digital signals before that they can be used by a computer or the data logger par To read the data with 5 seconds resolution the gauge must be directly connected to a computer through the grey cable going out from the converter It is a 9 pin junction cable see figure 7 Pay attention to have a such connection possibili
13. different noise values for the 1 minute resolution case According to the previous mentioned characteristic of noise 1t would thus introduce partly noise information that 1s not concerning the considered minute It could on this way provide less accurate values for the 1 minute resolution case Looking at the noise values provided with a resolution of 1 minute it appears very clearly that the noise comes one minute before the related intensity see figure 39 Considering the delay of 1 minute that affects the intensity values the noise values should thus be real time indications a Intensity Intensity e Noise 4 Noise g mm min g mm min 12 13 12 21 12 29 12 13 12 21 12 29 iN Figure 39 Right Intensity and noise measurements Left the noise is retarded of 1 minute 3 3 2 Wind noise relation In order to find out this relation 1t 1s first necessary to better understand which factors create noise and what s their relative contribution of the total recorded noise It 1s assumed that two factors the wind characteristics and the rainfall intensity are responsible for the major part of the recorded noise Moreover it is as well assumed that the total weight of water present in the bucket could also play a role on the noise response for example through a more pronounced attenuation of the gauge vibrations with increasing weig
14. distribution Table 14 Basic statistics of the ground noise for 2 different weights Initial weight g 3800 9800 Noise mean g 0 0468 0 0496 Standard deviation of the noise g 0 0136 0 0143 Minimum noise value g 0 013 0 013 Maximum noise value g 0 2 Box plot ofthe two sample data q 0 1 5 01 Gi gt 0 08 r 0 0 02 0 04 0 06 0 06 0 1 0 12 u 4 Quantiles 0 06 o 3 0 15 0 15 E E 0 04 E D 1 a 0 05 0 05 a 0 02 2 piti am J 0 05 0 05 Gi 0 5 5 0 5 3500 y 3000 q Standard Normal Quantiles Standard Normal Quantiles Figure 41 Left Box plot median lower and upper quartile values The black line extending each end of the box show the extent of the rest of data The red outliers are values that don t belong to this interval Up right QQplot of the two series It represents more a less a line signifying thus that the two series share the same distribution Down right QQplot of each series compared with a normal one 38 Looking at the distribution of the data h gh ground noise values that could be wrongly associated with extern perturbations occur quite rarely Moreover because the following tests in order to exploit the noise values are conducted with a resolution of 5 minutes high ground no1se values should be diluted reducing thus the associated risk of wrong interpretations 3 3 2 2 Wind induced noise The wind can involve noise in the measurements mainly because of dynamic pressure v
15. forget to enter the before and after the parameter name Table 6 The procedure to check set and save parameter values Check current value of parameter at getparam name Set new parameter value at setparam name new value Save new parameter value attsaveparam AAA After the save the logger will restart and the maintenance mode will be closed Before saving it s worth to check that the modifications were well done checking again the current value of the changed parameters 19 There are a lot of parameters that can be changed it is as well possible to set different alarms More information is available into the directory TRWS manual MPSOSCPU logger The files are however in Slovak Different parameters concerning the SIM card or the GPRS transmission are however presented n the table 7 To install the SIM card it is just necessary to indicate its apn For this test phase it was gprs swisscom ch No values were entered for the apnlogin and the apnpassword Table 7 The parameters concerning the sim card 1 SIM card with GPRS 0 no GPRS SIM card APN SIM card APN login SIM card APN password 2 6 2 Communication with the gauge The terminal allows as well speaking with the rain gauge directly from the computer Thus the green and yellow wires have to be connected to the PC place and the information will transit by the converter and not be sent to the data logger The 9 pin connection cable that go
16. highest intensities a greater spread is present Figure 54 Noise rainfall intensity relation for different intial ranges of weight Analyz ng separately each set of data 1t appears that the relation noise intensity yempty 11034x 0 0382 x 1050 1609 seems to depend from the initial weight un en Indeed the trend lines computed for each u ee weight range seem to present light different behaviours among the different weights as ae visible on the figure 55 The table 18 y o presents the different obtained equations It ee PRE 95009 was necessary to first correct the indicated intensity because of the warming phase of the 2 intensely an mim data logger more information n chapter 3 1 1 2 Figure 55 Trend lines for the empty and 9 kg cases Table 18 The different derived linear equations for each initial weight The 2500 3000 trend line is not represented because small intensities were not available for this data set weight trend line y ax b of samples y ax b not corrected values empty It appears thus that increasing the initial weight present in the bucket the noise induced by the rain particle tends to decrease It 1s in fact quite comprehensible because a higher water level in the rain gauge bucket can better absorb the impact caused by the rain drops On this way the weights measurements become less noisy It seems as well that after a certain amount of
17. light intensity to an amount proportional to its diameter They generally suffer from systematic non linear errors amounting up to 20 for high rainfall intensities These errors can be strongly reduced thanks to software or mechanical corrections The water level can be measured with any desired temporal resolution Without funnel there is noise in the weight measurement due to different perturbations This noise has to be filtered by software involving thus a delay of 1 to 10 minutes in the output With funnel there are wetting losses involving a delay for the residual water The solid precipitation has first to melt The thin nozzle requires a lot of attention for the field operation It is therefore more used for research purposes Little knowledge is available on the attainable measurement uncertainty of these devices As for the impact disdrometers little knowledge is available Moreover the calibration of this device is really difficult The TRwS s a we ghing rain gauge without funnel produced by the company MPS from Slovakia It 1s able to indicate as well the liquid as the solid precipitation with a resolution of 0 001 mm and an accuracy of 0 1 In addition sophisticated data algorithms eliminate the undesirable effects in the weight measurements due to the wind the evaporation or the temperature influence Moreover unreal step changes of we1ght are also eliminated 2 2 Description and installation of the mech
18. of the 5 considered 1 minute noise values see figure 43 noise 1min resolution een cli do A vd il ici aci aa ll raciocinio ecc rl cli i ra eto dada itas ali alli 500 1000 1500 2000 2500 3000 3500 noise variance and wind speed 5min resolution m s 9 0 AREMA N heta coto WI AA w 4 0 1 00 200 300 400 500 600 700 1 00 200 300 400 500 600 700 noise and wind direction change Smin resolution noise varlance and wind direction change 5min resolution Wi variance 150 noise 3100 023 za ML I tat Al ii M Bi Al h m AI Mi 0 LINA Al ANN ach an mi AK N HR MM A ih AO A A Ur N I Li Bo wW 20 300 A 500 600 700 100 200 300 400 500 600 700 noise and wind speed change Smin resolution noise variance and wind speed change Smin resolution noise variance O ID g a ALM HW Ah it a un HIN 0 0 100 200 an 500 500 700 100 200 300 400 500 600 700 noise versus wind speed noise variance versus wind speed Cross Noise Noise ee BR correlation variance O 05 1 15 2 Wind m s m s noise versus wind direction change noise variance versus wind direction change speed Wind EEE direction 0 1568 0 0988 0 100 200 change deg deg noise versus wind speed change noise variance versus wind speed change Wind speed change Figure 43 Above the two different data sets Below cross correlations 40 It appears that for the 5 minute resolution the w nd speed changes
19. of the data set on the threshold selection a calibration period presenting less clear dependence between noise and wind speed was selected see figure 51 The correlation coefficient amounts to 0 6866 As expected as well on the calibration as in the validation period lower success rates are obtained 64 21 for the calibration period and 69 82 for the validation period see figure 52 Wind velocity m s Corresponding noise threshold g nolse q WS m s Figure 50 noise wind speed relation for the calibration period first third of data 0 4 nolse q WS m s Figure 51 noise wind speed relation for another calibration period second third of data 0 057320 0 076615 0 107630 Specificity 09 os SS l Sensitivity 0 68 obser ed observed 0 0 5 1 predicted 15 2 1 1 3 2 predicted Figure 52 Above selected threshold Below left comparison for the calibration period Below right comparison for the validation period 44 However 1t appears that the differences between the success rates obtained with the different cal bration periods are not so high Moreover even f the cal bration s made on a data set presenting a smaller correlation between the noise and the wind speed the selected threshold values are however able to provide good results for the validation period So 1t attenuates a little bit the influence of the data set quality on the success rate It indicat
20. rainfall depth aggregated at different scales It showed theoretically that for aggregation intervals T tending to 0 seconds the variance pattern behaves as T Moreover for large aggregation scales the variance behaves as T with 1 lt B lt 2 The value of B depends on the correlation structure of the observed rainfall For the case with a finite memory B should equal 1 On the contrary if the rainfall presents an infinite memory B should be greater than 1 Because two different power law scaling are available for very small and large aggregation scales t implies the existence of a transition regime linking these two distinct parts This transition regime is thus inadequately represented by a power law because the exponent should decrease from 2 in the inner regime to 1 lt B lt 2 see figure 59 53 finite memory A ae A rn variance mm Wariancea inner Scaling ar regime 4 Y one r regime transition regime scaling regime Transition regime 1015 min 20 20 hours 10 io 10 10 10 10 10 104 10 10 10 40 10 10 oo 40 aggregation interval hours Aggregation interval 3 Figure 59 Left theoretical variance curve of the aggregated rainfall process as a function of the aggregation interval Right experimental variance Source 5 and 6 Marani conducted different analysis and observations and concluded that in the wide set of different locations climates and observa
21. reference 4 gauges selected on the base from the laboratory tests were chosen These references were placed in a central pit in the middle of the field while the other rain gauges were installed with equal distance as visible in the figure 25 Figure 25 The field installation of all the tested rain gauges Source 3 Unfortunately even if these tests are already achieved the final report and the results are not yet available With the experiments that were conducted during our first test phase it is not possible to roughly assess the performance of the gauge under different climatic conditions because no references were available 2 Just a little problem was observed concerning the catching feature of the gauge but t probably affect numerous other gauges that collect the water Logically at the end of a rainfall some drops are still present on the orifice They fall with a little delay in the bucket leading thus to wrong simulations of very small intensities after the end of the rain see figure 26 Burra Figure 26 Wrong intensity simulation of about 0 003 mm min after the end of a sprinkler test In the second phase of the tests the rain gauge will be placed at Payerne and Zermatt near to existing rain gauges from Meteoswiss Thus it will be possible to become a better idea about the performance of the gauge concerning the catching errors During the first week a snow fall was caught by the both devices i
22. response leading thus to problems for the interpretation of the results Looking at the relation noise intensity for sprinkler tests and natural conditions 1t seems that even 1f t 1s not exactly the same processes the induced noise are quite comparable and that the sprinkler tests are able to provide rain that present plausible features concerning the induced noise see figure 89 u sprinkler tests natural rain E sprinkler tests natural rain noise g Intensity mm min 0 2 0 4 Intensity mm min Figure 89 Left comparison of the noise induced by sprinkler and natural rain Right zoom on smaller intensities A2 Effect of rain particle falling on the housing Under real no wind conditions the rainfall particle should fall quite vertically in the gauge With this type of sprinkler test it 1s quite impossible to obtain drop particles falling vertically in the gauge so that a lot of particles are hitting the housing with non vertical trajectories It could thus induce an additional noise This effect was assessed covering the orifice of the gauge while sprinkler tests still occurred trying thus to separate the effect of these drops hitting the housing On a second phase the sprinkler rain was stopped and the orifice was again opened It appears that the noise induced by the hitting of
23. special feature represents thus not ne inevitably a contradiction of the Maran obsevations On the Aggregation seconds contrary Marani itself wrote that the shape of the variance curve in the transition regime is not always smooth as it depends on the shape of the autocorrelation characterizing rainfall 6 As for the selected case the inner regime presents a B value of 1 59 It represents this time a contradiction with the Marani observations 1000 10 100 1000000 0 1 sample variance mm42 Figure 79 Aggregation scales between 2 minutes and 32h30 4 3 3 2 The longer memory event With an extension of the before considered event to about 62h30 logically the patterns of the autocorrelation function change as well During about the first 500 minutes a regularly decrease up to 0 of the autocorrelation coefficients occurs For the first lag very high values of these coefficients are available see figure 80 Sample Autocorrelation Function ACF zinfalldepih ewe Ein IN ENE sample Autocorrelation a 1000 2000 3000 4000 Lag minutes Figure 80 Left intensity and cumulated rainfall depth Right ACF Again the analysis made over this extended set provided o similar results as before see figure 81 It shows however Se di ee ode 00 00 00 an accelerated decrease of the slope for the highest variance 3 Y scales where a power low fit prov
24. that the results are quite similar It 1s indeed quite logical because the noise 1s assumed to be random and should thus disappear with increasing aggregation scales Some little discrepancies between the two variance values could maybe provide from the filter algorithms that sometimes provide other intensities than the difference of weight Looking al scales smaller Aggregation between 10 seconds and 12h30 than the smallest one provided with 1 minute resolution data 2 minutes it 1 006402 appears that the var ance provided by the 5 seconds resolution is not directly 1 00E 00 usable below aggregation scales of E 100 fo 1000000 5 seconds about 1 minute Through its oscillating i ee behaviour the noise logically tends to E af increase the variance for small 2 1 00E 04 resolution ageregation scales Thus the use of the E 5 seconds resolution data don t provide er quite additional interesting results that time seconds justify the quite bigger amount of time needed to calculate the variance patterns Figure 62 Sample variance among aggregation scales from 10 seconds to 12h30 55 10 0 5 Moreover the use of data set coming from raw values involves additional problems Indeed on the contrary to the data provided with treated intensity values the raw weight data are affected by the evaporation leading thus to accumulated rainfall depths that are not any more usable fo
25. the service mode if any other commands than one expected by the terminal is entered the service mode will quit and return to the operating mode The most important commands and the related answers are well presented in the user s guide chapter 5 3 and will be thus not developed here Just a precision will be made concerning the way to change the rain gauge s address There are indeed two different possibilities indicated in the user guide but it is rather recommended to use the one of the service mode described on the point 5 3 8 Pay attention that the address should be a 2 hexadecimal chars but in fact the terminal accepts all the value that are entered During this test phase a wrong manipulation was made instead entering 2 hexadecimal chars the address 9 lt enter gt was introduced Because such an address can for example not be read by the usm program it 1s absolutely necessary to change it For this purpose it was first necessary to enter in the service mode thanks to the command OPEN9 013WS 013 and after change as normally the address 21 3 First tests This part presents further experiments that were conducted with the different sensors available on the rain gauge These tests will provide a better general understanding of the working style of the gauge and will also allow making some statements about the sensors accuracy Thus the intensity and temperature values provided by the gauge will be analyzed Moreover 1t will be also
26. try to better understand and exploit the provided noise values thanks to different analysis of the factors that engender noise 3 1 Tests about rainfall intensity With weighing rain gauges the intensity is directly derived from the continuous weight measurements Thus in order to better understand how works the gauge the indicated intensity and the one calculated from the weight variations will be compared for the both available temporal resolutions This comparison will as well allow assessing if some features of the data processing could lead to small rainfall deformations Finally this comparison also allowed demonstrating that the gauge was as indicated by the manufacturer able to eliminate unreal jump weights It is difficult to assess properly the intensity accuracy of the delivered rain gauge because reference values are missing It was thus impossible to assess precisely the performance of the gauge regarding at the counting and catching errors Therefore the results of experiments conducted by the World Meteorology Organisation WMO will be in a second part briefly presented They compared the performance of different rain gauges including the TRwS in laboratory and field conditions 3 1 1 Calculation of the intensity 3 1 1 1 Data with 5 seconds resolution Table 8 Calculation of the intensity The calculated intensity simply results from the difference of the absolute indicated ed weights taken with an interv
27. water present in the gauge the absorption of the rain drops shocks is not more dramatically increased Indeed the difference in the trend lines between 5300 g and 9100 g is not so big On the contrary for the first ranges of weights the differences are bigger 48 A very important feature of this relation is i A the spread Indeed as visible on the figure k 56 the noise induced by the rain amounts to l intensity some multiples of the one induced by the E 06 _ mm min wind Thus even if the different R E 0 4 i noise al coefficients are very good for the different aay derived linear relations it appears already that it would be quite impossible to obtain er m values about the wind velocities when it is S E amp raining Indeed the slightest spread in this linear relation changes the noise values in Figure 56 noise before and during a rainfall domains that correspond to really high wind velocities It was however optimistically or naively according to the point of view tried to isolate the wind effect from the noise response during rainfall c Assessment of the wind during rainfall from the noise values The chosen procedure to isolate the wind effect on the noise during rainfall is very easy A simplified model assuming that the noise depends only from the wind and the rainfall intensity is adopted Thanks to the known noise rainfall intensity relation the noise values are corre
28. when an intensity greater than 0 01 mm min occurs The first value indicated by the gauge is quite higher than the indicated one It is a logical consequence because all the precipitation that occurred before that the intensity threshold was reached are however taken n account Thus for this example the sum of the calculated intensity between 07 42 and 07 45 equals the indicated intensity of 07 45 Once this threshold reached the gauge can after simulate lower intensities with a resolution of 0 001 mm min The residuals smaller than 0 001 mm min are stored by the gauge and added to the next minute It explains thus the small discrepancies observed in the table 9 between the both intensities This characteristic leads to a small deformation of the rainfall feature mainly at the beginning of light rainfalls However looking at the involved small concerned intensities it 1s not a major problem Moreover the total amount of rainfall stays the same for the both intensity values In order to have a full comprehension of the different rainfall deformations that could be made by the gauge it 1s also necessary to look at the raw weight measurements A more detailed analysis of the data allowed remarking that apparently the filter mechanism implemented to treat the raw data doesn t lead to additional rainfall deformations Indeed for all the cases observed it appeared that the treated weight values WABS always were able to reproduce withou
29. 0 2 0 1 0 0 1 0 2 0 3 05 i i 2 O 2000 4000 6000 8000 2 00 00 18 00 00 Lag 5 seconds Modas Icon NEE are NEED 3 rainfalldepth mm intensity mm min sample Autocorrelation ia Ei 5 4 3 2 1 g 10000 Figure 58 Left intensity derived from raw weight measurements A quite symmetrical behaviour attests of the oscillations of the raw weight Right These oscillations are confirmed looking at the negative autocorrelation coefficient that was computed for the first lag of 5 seconds Moreover some other limitations reduce the chance to obtain wind indication from the noise values Indeed the wind can induce some catching errors from the gauge mainly for rains with a larger fraction of smaller drops and higher wind speeds 4 With such conditions it 1s probable that the gauge underestimates a little bit the real precipitation The wind seems thus to affect the noise also in an indirect way changing sometimes the amount of rainfall caught by the rain gauge Regarding at the big difference in the noise responses to rainfall and wind it could therefore occur that high wind periods correspond to moments where the noise was smaller because of the possible catching errors In closing the best indication about the wind conditions that occur during rainfall is probably provided by the noise characteristics obtained just before the rainfall beginning If wind corrections on the indicated intensity wan
30. 000 2 E E 5 D 0 01 0 01 time seconds time seconds Figure 70 Left comparison of the sample variance of the two sprinkler tests Right zoom on the second test from 2 to 10 minutes A more complete discussion concerning this strange effect will be made at the end of the different analysed set of data 58 4 3 2 Tests on short isolated rainfall events During the test phase the gauge captured different types of rainfall events For this test two different quite long events presenting short memory and long memory effects were selected 4 3 2 1 Short memory event This selected rainfall presented quite interesting characteristics Different peaks of vary ng intensity occurred during the day It provided about 35 mm of water with a record intensity peak of 1 2 mm min Logically this event doesn t present long memory behaviour the different peaks seem independent from each other They could correspond to the arrival of rain clusters see figure 71 rainfalldepth mm h m intensitw Irmewrninl Sample Autocorrelation da Le AA ar li be atl Sf Ih A ila eu Rag O 20 40 50 80 100 09 00 00 12 00 00 15 00 00 Minutes Figure 71 Left intensity and cumulated rainfall depth Right ACF The scaling based analysis provided the variance an patterns visible n the figure 72 At the f rst s ght an inner and trans tion regime s difficult to recognize It seems however that the same strange effects as befo
31. 1 wind larger windspeed_pred windspeed_pred2 lt threshold 3 0 wind smaller Computation of the success indicators of the prediction a true positive observed and predicted A wind_obs windspeed_pred a 3 sum A d true negative non observed and non predicted A 1 wind_obs 1 windspeed_pred d j sum A c false positive error type I observed and non predicted A wind_obs 1 windspeed_pred c j sum A b false negative error type II non observed and predicted A 1 wind_obs windspeed_pred b j sum A Security if the denimators are 0 A this security is the reason why the program provides wrong results if a wind range is greater than the maximal observed wind velocity Only 0 values would be present and thus for this range the sens and spec would not be new defined the program would thus display the same values than for the case i l 1E ayj CC Fe 0 ara 1 0 sensitivity proportion of positives cases correctly predicted sens j a j a j te J specificity proportion of negative cases correctly predicted spec 3 d j d 4 b 4 Ao ol A end selection of the best threshold maximize the sensitivity and the specificity sum_sens_spec j sens j spec j end 70 best_value i indice i max sum_sens_spec best_threshold i threshold indice 1 sensivity i sens indice i specificty i spec indice i
32. 1s necessary to introduce the transport rest under the strain gauge bridge as shown in the figure 4 Also after emptying _ operation or every time that the bucket is placed on the support plate it would be better Figure 4 The white transport rest is introduced to insert as well the transport rest under the strain gauge When the base is correctly installed on the pedestal just set the bucket on the support plate and install the white housing For the installation of the housing it is important to control that the heating wire is connected to power supply through the two little metallic stems At the end when all is set the three lateral screws outside must be gripped in order to stabilise the housing to the base 10 2 3 Description of the electronic part All the electronic part is installed in a little white box that isn t water proof For the further tests that will be made in Payerne and Zermatt the electronic components are installed in a water proof box It was successfully checked that the gauge was still able to send data trough GPRS inside this box The different electronic devices are briefly presented beginning from the right of the figure 5 Ds m Figure 5 Picture of the box and its different parts 2 3 1 DR 30 15 This component is a power supply able generating a precise output of 30 Watt and 16V from different input ranges It 1s connected to the electrical network through the big white
33. 2 00 00 18 00 00 12 00 00 18 00 00 Figure 60 Accumulated rainfall depth and intensity Left 1 minute resolution Right 5 seconds resolution 54 It appears directly that the noise 1s very high in the case of the 5 second resolution Indeed the thus computed intensities don t make sense They have a quite oscillating behaviour over a wide range of intensities and present as well negative values Logically these features play an important role looking at the derived sample autocorrelation function See figure 61 sample Autocorrelation Function AF sample Autocorrelation Function ACF O A O E E o 5 E l es T E ES as ner da an EEE Do aw Soe vee ae bees eee H E z 5 T a Z 0 a m m i O 200 400 BO ano O 2000 4000 e000 8000 10000 Lag minutes Lag 5 seconds Figure 61 Autocorrelation function Left 1 minute resolution Right 5 seconds resolution In the 5 seconds case the quite negative autocorrelation obtained for the 5 seconds lag attests of the big oscillations of the weight Comparing with the 1 minute autocorrelation it appears as well that a great part of the correlation structure was removed For the first lags autocorrelation coefficients above 0 2 are not visible and the negative correlation coefficients obtained for the case with 1 minute resolution almost disappeared Comparing the both variance patterns obtained over different aggregation scales see figure 62 it appears
34. 6sremove the data that are smaller than the i th wind speed range in order to evitate less good selection of threshold WStreatednew WStreatednew noisemeandStreatednew gt best_threshold 1 noisemeandStreatednew noisemeandStreatednew noisemean5treatednew gt best_threshold 1 ren ROC ee sensitivity against specificity figure 22 subplot 1 length wind_range 1 plot spec sens xlabel specifity ylabel sensitivity hold an straight line y 1 x indicates the limit with the random model fplot l x 0 1 TES holodt ori oe oo 5 Classification of the wind thanks to the best threshold value for m 1 length noisemean5treated if noisemean5treated m lt best_threshold 1 wind_calculated m 0 5 end if noisemean5treated m gt best_threshold 1 amp amp noisemean5treated m lt best_threshold 2 wind_calculated m 1 end if noisemean5treated m gt best_threshold 2 amp amp noisemean5treated m lt best_threshold 3 wind_calculated m 1 5 end if noisemean5treated m gt best_threshold 3 wind_calculated m 2 end end 6 Computation of the classification success Ao It separates as well the good from the wrong predictions success 0 for m 1 length WStreated if wind_calculated m WStreated m gt 0 wind_calculated m WStreated m lt 0 5 success success 1 WStreatedright m WStreated m WStreatedw
35. AW It oscillates between 25 or 35 seconds and 1 minute according to the time of the latest update Thus the indicated noise values present a delay between 0 and 25 or 35 seconds with the real time This delay in the noise is in fact quite logical and better comprehensible considering how works the gauge It seems that the gauge analyses all the weight measurements during 25 or 35 seconds and from the variations of these values can provide a noise value Logically this noise can only be indicated when the last measurement is taken in account involving thus some delay In this case it is 25 or 35 seconds according to the number of values analysed for the noise computation 36 Another feature not directly relied with the noise feature is a little bit problematic It appears that during a selected minute where the intensity 1s calculated there are three different noise values because there is a shift between the update of WABS and the time where the intensity is indicated For example for the intensity of 1 226 mm min the noise values of 11 024 15 287 and 4 849 g are considered Looking at all the different measurements made with 5 seconds resolution data this shift occurred quite frequently This shift is problematic for the aggregation of the noise values with 1 minute resolution Unfortunately no information 1s available on how is aggregated the 1 minute noise resolution It is thus possible that the gauge aggregates these three
36. ETH FU Eidgen ssische Technische Hochschule Z rich Institute of Swiss Federal Institute of Technology Zurich Environmental Engineering Projektarbeit High resolution precipitation intensity measurement and analysis December 08 Leonard Murisier Supervising tutor Peter Molnar Acknowledgements This work was a pleasure thanks to the very n ce collaboration w th Bettina and Maurizio as well as the whole team of the suspended corridor Thanks a lot for the good mood and for the help n var ous domains Thanks a lot to Marco Zhe and Konrad for the lending and the explanations concerning the cl mate chamber and different w nd and temperature sensors Finally also a big thanks to Peter Molnar for the supervision and the confidence during this work Table of contents IA AA Vai oussuubesesussuucussucecsss son uvegssecussabasbeussocees voy sevesesevesexeoveusouse 6 USER MANUAL OFTHE TRWS acid 7 ANIE TENERALTOIN T O TN 8 2 2 DESCRIPTION AND INSTALLATION OF THE MECHANICAL PART ssssssecceceessesesaceeceecessesaeeeceeseseesaneeeeseeeeeas 9 AO REGU CAMO OT A E i nee E AE O et a 9 PLA UNO AAA EE EE ce Ota dia E T E E OE 10 2 3 DESCRIPTION OF THE ELECTRONIC PAR atraido 11 Leo DRIOL I SAA ES AA AA AA 11 DIALES AR RAS A ern en a een ener 11 ZI SIE ONV ES 2 aaa a 12 ZS O TN 12 ZINE A ODER ER 12 2 3 0 2 UI CONREE ON Aia 12 2 4 DATA PROVIDED WITH A TEMPORAL RESOLUTION OF 5 SECONDS uuesseesssssssnnnnnnnnsnsssennnnnn
37. al of 1 oe a minute Because of the 200cm orifice 20 g of water falling during minute or corresponds to an intensity of 1 1110 17 i mm min under the assumption that LOT i i the water density always amounts to ie FU A 1000 kg m The indicated intensity is 1116678 14 SSS the value provided by the rain gauge 8 14 36 see table 8 08 14 41 1116 678 08 14 46 0 621 1122 588 0 6209 22 For a small rainfall event these two intensities were compared At the first sight the results are quite identical see figure 19 Table 9 Zoom on the end of this rainfall Ber Intensity mm min 0 14 intensity 0 E pe calculated 0 0055 wei E 0 08 intensity 0 002655 0 0 06 indicated 0 0148 0 023 0 0201 0 0 0135 0 014 Figure 19 Comparison between indicated and calculated 0 0002 0 001 as X S 07 17 07 37 07 47 intensity for a light rainfall event op ol Some discrepancies occur however at the beginning of the rainfall Thanks to other similar analysis t appeared that the rain gauge needs an intensity bigger than 0 01 mm min in order to detect rainfall This feature appears very clearly in this little rainfall event at for example 07 42 see table 9 The intensity derived from weight differences indicates that t is raining However the gauge indicates the beginning of rainfall only a 07 45
38. anical part The TRwS consists of the following parts I Ion l Bucket 2 Strain gauge bridge 3 Base plate with box for electronics 4 200 cm2 orifice collecting r ng with heating Housing Pedestal Support plate 3x adjustable screw bolts Adjustable transport screw bolt 10 Support of sensor 11 Adjusting screw bolts between the pedestal and the guying base ho lt Oo OANA Y Pe hE Figure 1 Mechanical construction of the TRwS 200 Source User s Guide ver 3 1 2 2 1 Required material For the installation the following tools are necessary a thin screwdriver also indispensable to change the place of the wires see the point 1 3 6 two keys 13 a level Moreover the following additional elements should be as well not forgotten Because the heating ring of the gauge is quite energy consuming max 2A it s necessary to be connected to the electrical network It is thus indispensable to take a European Swiss plug adapter for each rain gauge It is as well important to use antifreeze liquid for an optimal utilisation during the winter Common antifreeze liquid should be appropriated For the data transmission a sim card is logically required Specific indications concerning the sim card installation are available in chapter 2 3 2 and 2 6 1 If an intervention on the field 1s necessary for example in order to change certain parameters it 1s important to check t
39. ariations over the orifice of the rain gauge So it 1s not a trivial problem because 1t could be imagined that the wind characteristics such as intermittency or changes in direction and wind speeds could induce more noise than a regular middle wind velocity in regard to the caused dynamic pressure variations Another effect from the wind could be the creation of vibrations of the rain gauge that could be reflected in the weight measurements To assess properly all these effects sophisticated tests should be conducted where the wind features could be precisely controlled Maybe some interesting properties could be discovered but it is probably complex relations For example t would be not surprising to obtain the same amount of noise for different wind conditions Thus in regard to the goal of roughly and simply assess the wind velocity thanks to the noise patterns these tests would be exaggerated For this experiment an anemometer was placed at the same height than the orifice of the rain gauge The anemometer provided data about wind velocity and wind direction with a resolution of 5 minutes a Assessment of the weight influence It was initially planed to conduct different tests aiming at assessing the effect of the initial weight in the bucket Unfortunately the anemometer was only for four days available and because of quite cloudy conditions the solar panel was not able to powering continuously the anemometer Because of that just 2
40. as unfortunately impossible to check that surely because no information on the variance calculation was available on the paper from Marani Another version of computation that didn t select the mean of all the variances was also tried but the same effects appeared again If t can be demonstrated that this effect comes from inadequate calculations 1t would be thus worth to reconsider the 5 seconds resolution data Indeed a part of the observed too high variance for the smallest aggregations scale could come from this computation artefact and not only from the noisy values However the undesirable effect of the evaporation should still be removed It would be maybe interesting to go deeper in these considerations looking for example at the behaviour of the scaling of all the order of moments Maran looked indeed n 6 only at the variance and the second order moment characteristics among the scales Quick calculations showed that indeed all the moments are concerned with this changing behaviour during the transition period see figure 88 65 simple or multiscaling does moment scaling hold Ce observation rmomentl moment moment moments momentd taufg lagiii log delta q Figure 88 Left Plot of the different moments among the aggregation scale A delta value of 10 corresponds to the whole aggregated time series The smallest value of delta corresponds to 2 minutes aggregation Right the behaviour of the expo
41. at are assumed not affected by the data logger warming phase indicated intensity features were comparable In closing there are thus no differences in data processing between the two available resolutions as indicated by the manufacturer The both are computed in the rain gauge Moreover considering all the data up to now it appears that the gauge is always able to provide the intensity with a constant delay of 1 minute Thus the filter software seems to be robust and able to filter the noise even during heavy rainfall or high wind periods 25 3 1 1 3 Elimination of unreal step of weight The compar son between the indicated and the calculated intensity allowed as well demonstrating that the gauge was able to eliminate unreal step of weights As visible on the figure 23 a jump in the absolute weight occurred during a light rainfall at 13 21 The rain gauge was however able to eliminate this improbable variation and simulated thus only an intensity of about 0 04 mm min whereas the indicated one should provide more than 5 mm min a Weight T E E e indicated intensity Figure 23 The jump in the weight is not reflected by the indicated intensity This elimination s very probably possible thanks to the different weights measurements WBEGIN and WCOMP presented in the chapter 2 4 They should probably not present this jump of about 100 g These other weights are probably also the
42. at on this way no wrong rainfall should be created The tests concerning the noise indicated that there are good chances to obtain a reliable wind assessment method from the noise values However further tests should be conduct in order to extend the range of observed wind and initial weights If these tests are made it would be important to pay attention at the height of the anemometer relatively to the rain gauge orifice On the contrary these tests indicated that the attempt to derivate wind speed thanks to the noise patterns during rainfall would be similar as looking for a needle in a haystack The better indication of wind speed n case of rainfall 1s thus probably given by the noise values just before the beginning of the rain Concerning the GPRS transmission this first test phase indicated that it was reliable Only some minutes are missing or contain empty information Roughly estimated it concerns less than 0 2 of the total data The data logger presents an important property If the strength of the GPRS signal decreases it 1s able to store and send again the data when the signal becomes 51 stronger Because the logger has a memory able to store the data for 1 month the risk to loose measurements 1s rather weak However the data logger 1s not able to send the newest measurements after a long time without communication Indeed in this case 1t begins to resent all the historical messages to the server It lasts for many hour
43. ate the range of intensities a sprinkler installation will be used But before executing these tests it is primordial to check that this installation is able to reproduce the same noise values than a natural rainfall otherwise the observed relations would be maybe no more useful on the field Indeed with the sprinkler installation see figure 53 the simulated rain present different properties than a natural one notably concerning the size and the trajectory of the rain drops Because it doesn t correspond to the ideal no wind conditions it was also checked that the drops falling none vertically on the housing and n the bucket don t create additional noise All these effects were successfully controlled the results and discussion of these sprinkler validation tests are presented in the appendix RENEE ols BRASS A Figure 53 The sprinkler installation 47 b Assessment of the weight influence After these precautions the planned experiences were conducted with six different ranges of initial weights ee Looking at the all set of data it appeared 3 1600 20009 that the noise rainfall intensity relation 2 x 2500 3000g presents a quite stabile and linear x 5300 6200g behaviour among a wide spectrum of e intensities see figure 54 It was unfortunately difficult to simulate well greater intensities with the sprinkler intensity mm min installation but it seems however that for the
44. behaviour described by Marani For the data set probably more similar than the one used by Maran the presence of a transition regime at usual scale of hydrological interest was also found The other set of data provided sometimes similar results but the interpretations were quite hazardous because of the great uncertainties considering the relevance of the used data set for these comparisons with the results from Marani However a point is very problematic Indeed the slope increase of the variance curve constantly recorded for all set of data after the first computed aggregation scales seems quite unnatural The biggest clue to pretend that it is possibly an artefact 1s provided by the both sprinkler tests After this first increase the slope was stabilized at a value from 2 illustrating thus the validity of the Marani s observations considering the inner regime Indeed these sprinkler tests should represent an extension of the inner regime because it is supposed that the very small temporal scale variance that could occur during a rainfall 1s reproduced It would be worth to take time to analyse what could be the cause of this increase Artefacts created by the rain gauge can be excluded because the long historical data measured by another rain gauge presents the same features These long time series allowed also excluding an artefact produced by too short data sets Thus very probably the matlab code computes wrong calculations It w
45. ble operative conditions the 1 minute precipitation data provided by the gauge during the different temperature run tests were analysed For the empty case t appears clearly that the filter mechanism was able to avoid simulating wrong rainfall measurements even if the intensity calculated from the weight variations was sometimes greater than 0 01 mm see figure 35 These results are very encouraging because normally intensities bigger than 0 01 mm minute should be sufficient in order that the rain gauge begins to record the rainfall Thus it is a demonstration that the filter mechanism was in this case able to avoid simulation of wrong rainfall Temperature weight dependence 0 02 E Weight g 0 015 E 0 01 2 calculated 0 005 rainfall D c intensity 0 indicated 0 005 E rainfall 0 01 intensity O O O O O O O O O LO 00 o m co O N LO N O O O Figure 35 Comparison between calculated and indicated intensity during the different temperature runs 33 For the case with about 10 kg in the gauge the filter mechanism was also able to avoid simulating intensities greater than 0 01 mm min However for two cases the rain gauge simulated wrong rainfall of about 0 3 mm min These intensities are high and logically linked with some jumps of the weight See figure 36 Temperature weight dependence
46. ble in the figure 76 After the first visual impression it appears that this variance pattern corresponds better to the one expected even f the same strange behaviour still affect the inner regime Indeed a decrease of the slope seems to be present for the highest aggregation scales It could thus correspond to the description of the transition regime made by Maran Logically regarding at the different f values see figure 77 this decrease 1s quite small because the highest available aggregation scale amounts to 12h30 It corresponds thus to only a part of the transition regime According to Marani this transition could extend up to aggregation scales of about 80 hours 1 00E 01 1 00E 01 1 00E 03 sample variance mm22 1 00E 05 Aggregation seconds Figure 76 Aggregation scales between 2 minutes and 12h30 60 10 00E 01 B eee E lt 2 R 0 9981 1 00E 01 100 10000 E ve y 1E 08x1 7748 S R 0 9994 2 01 S 5 1 00E 03 2 y 5E 09x1 9448 S d F S R 0 9999 2 o o a o E 1 00E 05 2 YN G 0 001 Do 4 aggre dation seconds aggregation seconds aggregation seconds Figure 77 Zoom and power law fitting on different aggregation scale ranges Left between 2 and 10 minutes Middle between 10 minutes and 2 hours Right between 2 hours and 12h30 Finally these tests of selected rainfall event wit
47. cated wind velocity will be greater than the one occurring just over the rain gauge orifice 46 3 3 2 3 Wind induced noise during rain To assess this effect it is first necessary to isolate the noise induced by the rain only For this purpose some experiments will be conducted with different initial weights under constant wind and over the same range of rainfall intensity In concrete terms it signifies that a noise intensity relation will be derived for different initial weights When this relation will be better understood it will be tried to isolate the wind effect from the total noise value when it is rainy a Limitations of the experiment and sprinkler test validation Concerning the wind it was unfortunately impossible to find no wind conditions even in the internal garden But even if each sequence of measures for a defined weight occurred at a different moment of the day under different wind conditions it is however assumed that these wind condition differences don t lead to systematic bias A reason for that is that these experiments were conducted more times for each weight on different days Moreover for the majority of the measured data sets the duration should be enough long n order to capture different wind conditions Thus the wind should rather include an additional noise whose amount depends on the wind characteristics and lead thus to create an additional spread for the intensity noise relation To simul
48. cess of maximal 10 minutes where the autocorrelation coefficients regularly decrease However they are always below 0 2 attesting thus of the weakness of the memory process Different factors such as changing wind properties or variations of the water tap flow could explain this little memory signal The whole sprinkler test could thus be interpreted as an artificial prolongation of the really small scale time variance that occurs during a rainfall Therefore this data set should extend indefinitely the inner regime mentioned by Marco Maran The obtained results confirmed this assumption a power law fitting provides indeed a B value very close to 2 see figure 67 However 1t seems that for the smallest aggregation scales the fit with the power law presents some discrepancies In order to m check this impression only the variance aggregated time seconds 100000 y 9E 06x 12563 R 0 9997 1000 sample variance mm22 10000 up to 10 minutes were considered It confirmed that for this domain a power law presenting a smaller B value 1 8 provides a better fit see figure 68 Figure 67 sample variance among aggregation scales from 2 minutes to 7 hours 57 100000 E y 2E 05x 809 a y 8E 06x 1973 E R 0 9996 E R 0 9999 1000 9 3 S S 10 q o 2 E E 5 5 0 1 time seconds time seconds Figure 68 Left sample variance among aggregation
49. ch 67 APPENDIX ocino a dias 68 Ae SPRINKLER TES ES VALIDATION Ss ea E A T eannvenmandes 68 Al Effect of non similar rain drop properties acnesssssssseseenenennensnnssssnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnsnnssnnnnennnnnnnsnnnn 68 A2 Effect of rain particle falling on the housing nnneeseeeeeeeeeennenssnnnnennnnnnnnnnsenennnnnnnnsnnnnnnnnnnnnnnsnssnsnnnnnnnnnennnn 68 B WEA TEAC ODES gt een er Rees 69 Dele PCOVCTION Of WING SV COO Ac 69 BZ CALITO DISCERNIR EET 12 1 Introduction In August 2005 severe floods affected w de regions of Switzerland causing 6 dead and damages of about 3 billion Swiss francs The conducted event analysis 1 indicated among others that the accuracy of the predictions suffers from an insufficient consideration and a poor understanding of the small scale variability in space and time of different processes and their interactions In order to ameliorate these predictions the project APUNCH Advanced Process Understanding and prediction of hydrological extremes and Complex Hazards focuses on the path of a water drop from its formation to its end destination Concretely this multidisciplinary project deals with atmospheric and precipitation processes sediment transport mechanisms and hydraulic as well as geotechnical processes Concerning the precipitation this project aims at investigate the temporal and spatial structure of rainfall in mountainous regions For this purpose a X Band radar was installed a
50. climate chamber was used It was unfortunately impossible to get temperatures below 0 degree The comparison between two button sensors and the external sensor of the rain gauge provide excellent results See figure 28 Just any discrepancies are visible decrease after 40 degree and increase after 0 degree principally for the second sensor During this time the climate chamber was set off and so no more ventilation occurs Because the second sensor was not exactly at the same place than the two others t 1s probable that the temperatures were already different sensor 1 sensor 2 deg cels sensor gauge 01 00 11 00 21 00 07 00 Figure 28 Comparison between two button sensors and the external air sensor of the gauge Further tests were conducted in more natural conditions These tests were conducted because this external sensor is not ventilated It is thus possible that in case of low natural wind the reflected shortwave radiation creates some deviations in measurements To assessing this effect a button sensor was installed at a shaded place that was protected from the incoming and reflected shortwave radiations Just one comparative sensor was used because with the first experiment it appeared that the both button sensors provide the same temperatures On the contrary the external temperature sensor of the gauge is exposed to these radiations It appeared tha
51. cted when it is rainy A problem with this procedure is that the wind is also present during rainfall and thus affects also the noise rainfall intensity relation On this way the corrections also delete in a certain part the wind effect on noise But optimistically it could be thought that the spread around the linear trend could in a certain part reflect the discrepancies with the mean wind conditions that occurred during the rainfall Looking at the total predominance of the rainfall intensity on the noise response 1t seems absolutely necessary to define the noise rainfall intensity relation for each considered rainfall Otherwise the slightest bias affecting the noise rainfall relation would provide corrected noise values that are unusable This procedure was applied for the sprinkler experiment represented n the figure 56 Looking at the corrected noise values see figure 57 it appears directly that they are quite unusable in order to assess the wind velocity during this sprinkler test The oscillations are indeed great and provide sometimes noise values that were never observed with the solely influence of the wind Even if the mean over 5 minutes is considered for the noise values it still provides unusable values The same procedure was applied for rainfall event presenting smaller intensities but without more success y 9 7857x 0 1137 T intensity 7 4 rainfall E mm min ae Linear corrected
52. d a P value from about 1 42 Thus comparing with the before obtained value of 1 48 for the variances between 1 and 10 hours the presence of the transition regime 1s confirmed even 1f the change stays small 1000 10000 9 y 6E 05x 83 S R 0 9997 10 SS EN S 100 z lt gt E gt T 2 E 2E 01 10000 1000000 5 D 1 0 001 1 100 10000 1000000 aggregation minutes aggregation minutes Figure 84 Left the new considered aggregation range Right Interval between 100 hours and 15 days 4 3 4 2 December time series Similar tests were conducted for the time series containing 10 December months Again the considered variance curve was reduced to aggregation scales not bigger than 15 days see figure 85 10000 100 a S o 100 lt O E E 1 Y ae 1 g 100 10000 1000000 gt E c 2E 10000 1000000 S 0 01 s 0 01 0 2 0 0001 S anaes aggregation minutes aggregation minutes Figure 85 Left the whole aggregation range up to 10 months Right Only up to 25 days About the same variance patterns among the aggregation scales than in the July case are visible However with this time series a first increase of the slope after the first aggregation scale occurs The value goes from 1 62 to 1 73 After that the transition regime is well visible and for the scaling regime a value of 1 43 is obtained
53. days of data with empty bucket were usable The experiment made with about 5 kilos weight in the bucket just provided 1 hour of data that are not usable However thanks to other experiments made before that the availability of the anemometer it seems that the total weight in the bucket plays a role concerning the noise The water present in the gauge seems indeed to attenuate the weight variations and thus the noise created by the wind see figure 42 empty 4000g 70009 100009 noise g 14 29 p 15 52 20 02 21 25 17 15 Figure 42 Noise values during a day where the initial way was regularly increased 39 It is unfortunately impossible to prove and to quantify this effect because the variations in the behavior of the noise could also be only explained by different wind conditions that occurred It s however better to take precautions and consider that the experiments that will be now presented in order to find a relation between the noise and the wind speed are only valid for an empty rain gauge b Analysis of the recorded noise and wind data First some preliminary tests on the noise and the wind measurements are made in order to better understand the different connexions between these two sets of data The correlation between the noise velocity the change in wind direction and in wind speed are thus assessed as well for the 5 minute noise mean and the variance
54. e or with the data logger On this way some different important parameters can be changed To start the terminal double click on terminal exe in MPS tools Terminal 2 6 1 Communication with the data logger For that purpose an additional junction cable DB9 that links directly the logger to the computer is required This cable presents quite particular characteristics that not all DB9 cables have All the pins must be wired The cable with article number 671592 and type AA 317 06 from Distrelec works well Moreover the green and yellow wires have to be connected to the data logger place Terminal 1 9b 20041226 by Br y gigi o x Baud rate Data bits Parity r Stop bits Handshaking 1 Bisconnect Ce C600 c 14400 C 57600 es none 1 none Hel 7 C 1200 19200 115200 CE C odd ATS CTS 8 2400 28800 128000 even 1 5 C XON XOFF Spout 4 9 4800 38400 256000 C mark RTS CTS XON XOFF Quit C5 10 C 9600 56000 custom 9 space C2 RTS on TX Settings Set font M Auto Dis Connect Time Stream log custom BR As Clear ASCIltable Scripting cs Eco AutoStart Script J CR LF Stay on Top a600 27 Graph Remote Eos ERAI CLEAR _ResetCounter E Counter 100 Ma Ms J Bin Receive StartLog StopLog 1 at getparam apn gprs swisscom ch at getparam apnlogin a at getparam apnpassword
55. e was plugged for days the first recorded rainfalls are not satisfactory caught It should however not create too big problems for the field work Indeed firstly the gauge will not be too frequently disconnected Secondly even within this warming phase the pattern of the rainfall corresponds to the one that would be indicated by correct intensities The assessment of the small time scale variance is thus not endangered Concerning the calibration of the radar data the observed discrepancies should not lead to big problems because they are rather at 1 minute scale relevant The radar provides information with 5 minute resolution case Anyway there is always the possibility to recalculate the correct intensities thanks to the weight measurements that are also sent with the 1 minute resolution data But on this way some filter mechanism of the data could disappear The experiments conducted about the temperature provide also satisfactory results The accuracy of the extern air temperature sensor is very good Because this sensor will be better protected from the radiations for the next testing phases the provided measurements should be ameliorated On the contrary a constant shift of about 0 6 deg cels affects the intern sensor It is however not so important because this temperature serves to the gauge mainly for internal work It was thus illustrated that the gauge was always able to account for the temperature weight dependence and th
56. ed to the simulation of rainfall intensity But this dependence plays a role for the calculation of short term evaporation There is also a negative correlation between the rate of weight variation and the temperature variation When the temperature increases as already observed the weight decreases It encourages higher evaporation rates On the contrary very low or sometimes even positive variation rates are observed when the temperature decreases See figure 38 Evaporation during about 30 hours Weight variation under evaporation 0 2 20 5700 h O c 45 Sy 5675 3 0 2 2 D a oc Temperature D Temperature O T p E Tio CAN 9650 deg cels 52 0287 deg cels gt a 58 dy 5625 2 5 0452 2 Weight a A Weight 2 5600 zu 5 variation m 0 6 9 min Eg 0 5575 11 00 12 00 13 00 14 00 15 26 21 26 15 26 21 26 03 26 09 26 Figure 38 Left evaporation during about 30 hours Right weight variation during a period with changing temperatures The sun could in part explain this behaviour With sun more energy 1s available enhancing thus higher temperatures The incoming radiation has probably also a non negligible influence on the evaporation rate So when the sun disappears the temperature goes down as well as the evaporation rate But this sun effect can t explain the positive weight variation that occurs during a strong phase o
57. es it is important to relativize the obtained values However according to the choice of these grouped aggregation scales quite different P values can be obtained see figure 74 It is thus very important to first consider the whole aggregation scale and not look only at the small selections of this set otherwise wrong interpretations could be made It is in fact logical that different B values are obtained because the length of the selected data set 1s not so big and thus B should be sensitive to the spread of the variance pattern The different period selected provided however values that already seem to increase among the aggregation scale Before to discuss this behaviour the longer memory event will be analyzed The selected rainfall event presented quite constant behaviour during a day About 12 hours of small intensities provided a total amount of about 6 mm water Logically in this case the memory is quite longer There is also a significant negative autocorrelation after the lag of 200 minutes Probably at this lag the computation leading to the different autocorrelation coefficients begin to take into account the smaller and bigger part of the intensity patterns see figure 75 rainfalldepth mm Sample Autocorrelation 300 600 400 Lag minutes 0 200 Figure 75 Left intensity and cumulated rainfall depth Right ACF The same analysis was conducted and provided the results visi
58. es out of the converter must thus logically be connected to the computer Terminal v1 9b 20041226 by Br y lolx r COM Port 7 Baud rate Data bits Parity gt Stop bitsy m Handshaking ei St Demeter co eeu 1m e son rs roe oy none Once again after opening the sean Ilez e7 1200 e 19200 e ns odd C RTS CTS ers ea 200 e e 128000 even e15 XONAOFF terminal control that the opened About e4 cs eao C 38400 C 256000 C mark C RTS CTS XON KOFF Settings Auto Dis Connect T Time Stream log Set font Teer M AutoStart Script f CR LF J Stay on Top Receive CLEAR Reset Counter 13 Counter 0 customBR Fi Clear ASCII table Scripting 9600 27 Graph Remote CO HEX T Dec Bin e ASCII f Hex StartLog StopLog Docs Eco Oos Zi settings than n the figure 17 Note that for th s case the DTR and the CTS buttons below r ght 11 11 a and above right have to be grey Pay also attention the Baud rate and Handshaking parameters are not the same as in the case where 1 the terminal was used to interact Gor Carts with the data logger Transmit CLEAR _ SendFile CR CR LF Macros Set Macros EMi M2 M3 M4 M5 MG M7 M8 M9 M10 M11 M12 Igetadr Igetadr N Connected Rx 6 Tx 46 Figure 17 Terminal setting for dialoguing w
59. es thus that the selection algorithm 1s quite robust Moreover looking at the selected threshold values for the 3 different cases there are not big differences expect for the one concerning the wind speed class greater than 1 5 m s After a better analyse of the data set 1t appears that this difference is logical looking at the procedure that define the best threshold Indeed the same weight is given to the specificity and the sensitivity Because the data set contains only a few wind observations that are bigger than 1 5 m s there are more chances to make wrong positive bigger than 1 5 m s predictions than to make right positive predictions Thus logically the program selects quite conservative value of the upper threshold in order to increase the number of right positive prediction These low thresholds lead logically to sensitivity value near to 1 and lower specificity values Only for the calibration period that concentrated a higher proportion of wind velocities greater than 1 5 m s the program selected a quite higher threshold value see table 16 Table 16 Comparison of the obtained thresholds between the different calibration periods Corresponding noise threshold g Proportion of wind observation above 1 5 m s Whole data 0 055528 0 078276 0 10748 12 729 Calibration on the 1 0 055528 0 088439 0 19216 8 272 third Calibration on the 24 0 057320 0 076615 0 107630 4 285 third Moreover these little experiments give some intere
60. f temperature decrease So looking at these results it seems more carefully to derivate the evaporation rates at higher ageregation scale or not only during a heavy increasing or decreasing temperature phase Looking at the evaporation over the whole period these variations become negligible 35 3 3 Tests about noise The strain bridge provides weight measurements 24 times pro second In order to provide stabilised weight values different complicated algorithms are implemented to filter the weight measurements These algorithms also provide weight noise values in gram These values are thus an indirect indicator of the different elements that troubled the weight measurement The characteristic of these values will be first briefly presented in the both case of 5 seconds and 1 minute resolution In the second part of this chapter different tests w ll be conducted to find a relation between the measured wind and the recorded weight noise in both cases with and without rainfall On this way only the weight noise values could be able to provide a rough indication about the wind conditions 3 3 1 Features of the noise As already mentioned in the tests for the rainfall intensity the data are processed in the same way for the both resolution cases However looking at the noise features with a resolution of 5 seconds will allow to better understand how these values are calculated Table 13 5 seconds resolution noise PPP ie isl Time
61. firmed with experiments involving other range of welghts see table 12 During these other experiments also strange behaviour in the weight temperature was as well sometimes observed Table 12 Effect on the initial weight on the temperature weight dependence Initial weight Relation temperature weight 0 kg Empty without bucket 0 0884 0 099 g deg cels 1 kg Empty with bucket 0 1239 g deg cels 09 a O 3 8 kg 0 193 10 kilos 0 2993 0 4031 g deg cels 32 This temperature influence is also well observed with natural observations when the gauge 1s placed outside The following example llustrates quite well this effect and provides a more concrete vision of this influence see figure 34 At 17h50 the sun disappeared leading to a higher decrease of temperature The pink curve for temperature presents indeed distinctly two different slopes Thus according to the preceding observations the weight goes up when the temperature decreases Before 17h50 the temperature was also decreasing but not the weight It is probably due to the evaporation because the sun was shining before 17h50 However looking at the eig indicated intensity yellow curve Mewserem luckily these variations were not enough relevant in order to simulate wrong rainfall Figure 34 Temperature weight influence in natural conditions To be sure that this dependence will not create wrong measurements under all possi
62. gth rainfalldepth scales 3 1 for noverlap 1 limit2 j varintermediate2 j noverlap var rainfalldepth noverlap noverlap scales 3 1 end varscale2 j mean varintermediate2 3 1 1imit2 3 end 3 Tests on the moment scaling To length rainfalldepth T scales mom_end 4 2 1 for pas 1 mom_end for k 1 length T delta k T k TO limit3 k length rainfalldepth T k 1 for no 1 limit3 k A no 1 T k sum rainfalldepth no no T k 1 pas 2 end moment pas k mean A end end plot of the different moment gt does the moment scaling hold figure 3 loglog delta moment 1 delta moment 2 delta moment 3 delta moment 4 delta moment 5 113 legend moment0 momentl1 moment2 moment3 moment4 ylabel looM q xlabel log delta title does moment scaling hold plot of the different tau q values gt simple or multiscaling for i 1 9 selecting the first aggregation scale is an approximation of the limit when delta tends LOU value i log moment 1 1 log delta 1 end xaxes2 0 0 5 4 figure 4 plot xaxes2 value o legend observation ylabel tau q xlabel q title simple or multiscaling 73
63. h different structure provided results that are difficult to exploit Indeed especially for the short memory case the results were quite strange Moreover because of the short length of the record only small aggregation scales could be derived Thus it is possible that the observed general tendencies increase of f in the short memory case and decrease for the longer memory case are only small discrepancies that wouldn t play a big role if quite bigger aggregation scale would be assessed However these experiments allow to again remark the same strange features affecting the smallest aggregation scales A comparison between the two types of memory effect is difficult to do because theoretically the influence of the memory should be visible in the asymptotic behaviour of the aggregation scales belonging to the scaling regime It is also difficult to comment these results because such tests with so short period of measurement are not mentioned n the paper of Marani On the contrary quite longer data set are used These data sets present also logically numerous periods where no rain occur 4 3 3 Tests on short extended rainfall event In order to have data set that could be a little bit more similar to the one used by Marani the selected rainfall events are extended to longer observation period It will allow exploring further the behaviour of the variance in the assumed transition regime thanks to aggregation scales of 32h30 and 62h30
64. hat the computer that will be used to make these changes has the good cable connections possibilities Otherwise adequate converters are necessary For more information see the chapter in 2 3 3 and 2 6 1 2 2 2 Installation For this testing phase instead using a base a metallic stake was pressed in the ground The pedestal was stabilised on this stake thanks to different wood flakes On this way a very good stability ofthe rain gauge was guarantied Once this pedestal is installed the base plate can be set on it For the precision of the measurement t s absolutely necessary that this base plate 1s set perfectly horizontal Thanks to the 3 different adjustable screw bolts see figure 2 1t 1s really easy to do 4 Figure 2 One of the adjustable screw that fix the base plate on the pedestal Use the level tool at least n two different directions to ensure that the horizontality is perfect Proceed in an iterative way with the level tool see figure 3 When it is horizontally in one direction recheck in the other one Or simpler use a circular level tool able to directly indicate the horizontality in all directions When it is satisfactory block the base plate gripping the different bolts Figure 3 The level tool The strain gauge that measures the total weight of fallen precipitation is installed on this base plate This sensor is very sensitive so that in order to avoid any damage during transport it
65. hmM ooooonoooccoccconononanononononnnncononnnnnnccnnnnnnncnnnos 45 Ey Conelu dns TOTALS a a a ee en seele 46 32325 Wind induced noiserdurins Tanz ea aan ae 47 a Limitations of the experiment and sprinkler test Validation coccnooococcconooonnncnononnonnncnnnononononononnnnnnccnonnnnnnonnos 47 b Assessment of the weieht influence uses rules 48 c Assessment of the wind during rainfall from the noise values ccoooooocncnnoooonnnnnnonoannnnnnnnnnnnncnonnnnnnncnnnonononcn n 49 d Concludmerremarks teens Sessel ence Meet in steht eig 50 gt A GENERADA PPRECTATION OF THE RAIN GAUGE u a ders 51 DATA ANALYSIS 2a ee a 53 AN EN ERA POINT 9 ers ss Sen Meeres 53 4 2 PREBIMINARY TES VS cai 54 4 2 1 Influence of the data reson Oo hrc A SS dai 54 4 2 2 Influence of the calculation Of the scale variance nnaeeeseeeeeeeeeenenensnnnnnnennnnnnnnennnnnnnnnnnnsnennnnnnnnnnnsnnnnnnnn 56 E E A 57 SMS A E ee E T o EE S 4 3 2 Tests on short isolated rainfall events nnnnnneeneeeeeeeneneeeessnsnsnnnnnnnnnnnnnnnnnnnnnnnnnnnnennnnnnnnnnnnnsnsnnnnnnn 59 Ael Shorr 01h 9 068 aw 0 ae are TR te iT ere tee ee eae eee 59 4322 2 OMT MEMO Evei enana enee ii rci 60 AIR Less ON SOLE extended TUNA MEVA 61 4 5 5 1 het Short memory A EN se se 61 A Ns se a tee eee ues aude 62 ASTM ESOO LOMO N stonie GAO rss sen cies een lee aS 63 A A ee 63 43 42 December ME UA A A RS 64 A CONCLUDING REMARRS Ia RI E a 65 REFERENCES A ea
66. ht So in order to isolate the effect of each factor as well the effect of the rainfall intensity as the one of the wind characteristic will be assessed under different initial weight conditions However before starting these experiments it 1s first important to estimate the ground noise of the rain gauge 37 3 3 2 1 Assessment of the ground noise In order to become an idea about this ground noise the gauge was kept inside during many week ends where 1t was assumed that no perturbations occurred No wind effect no rainfall no human disturbance Thus the noise should mainly stem from the sensors or from the electronic components These experiments were also conducted with different initial weights see figure 40 ground noise for 3800 g ground noise for 9800 g 0 12 0 12 0 1 0 1 0 08 0 08 gt 0 06 U N ul PE a 0 06 0 04 che hl i 0 04 0 0 AAA MONA LLL Figure 40 Left ground noise for 3800 g Right ground noise for 9800 g Looking at the table 14 it appears that the mean and the standard deviation of the ground noise increase a little bit when the weight as well increases But these changes stay small and it is not worth to take this phenomenon into account for the continuation of these experiments The distribution of these noise values diverges from a normal one for the lowest and the highest quantiles See figure 41 Moreover the weight doesn t seem to influence the
67. ided a f value of 1 22 As 2 goes A for the non extended case a B value smaller than 2 was A eS obtained for the inner regime 1 0080 e aggregation seconds Figure 81 Aggregation scales between 2 minutes and 62h30 Finally looking at the result concerning the extending set of data there are rather encouraging signs Indeed with data sets that are a little bit more comparable with the one used by Marani for the both considered events 1t seems that already a tendency for decreasing slope of the variance pattern exists inside the transition regime Apparently the scaling regime characterized by a constant power law behaviour of the highest aggregation scales was never reached It is also difficult to assess the influence of the memory process Moreover the two set of data didn t present the same length and it could thus skew a comparison 62 4 3 4 Tests on long historic data In order to have set of data that should be n all features comparable to the one of Marco Maran some historical data of the meteoswiss station in Zermatt was downloaded The resolution amounts to 10 minutes All the months of December and July of the last 10 years were put together n order to create two different time series supposed to present different behaviour looking at the memory Indeed 1t was assumed that in summer convective events presenting high intensity and short duration would rather occur These characteristics shou
68. ion 1s first to assess how strong 1s this dependence and secondly to check that the filter algor thms developed by MPS are always able to eliminate this dependence so that the gauge doesn t simulate wrong rainfall 30 First some experiments were conducted inside the office with some objects inside the gauge instead of water in order to remove the effect of the evaporation on the weight All these experiments indicated that there is a strong negative correlation between the temperature and the weight See figure 30 But the correlation only said how well the temperature weight relation follows a linear relation In this case 1t appears that the influence of the temperature on the weight 1s rather weak About 0 124 g deg cels 1038 5 a 1038 2 2 3 E D 1037 5 2 san y 0 1239x 1040 6 p R 0 8995 Weight 1037 0 10 20 30 oO O O oO oO oO 22920909209 oO 00 Temperature deg cels oO o O O Figure 30 Temperature weight relation To test this dependence under all possible operative condition the same experiment was conducted with other ranges of weights and wider temperature fluctuations For this purpose the rain gauge was differently lasted during the different temperature runs executed in the climate chamber see figure 31 5 O 4 o 3 Temperature 32 z 65
69. ith the rain gauge It is the same principle as when the computer was speaking with the data logger There are also an operating and a maintenance mode The different commands of the operating mode are available in the chapter 5 2 of the user s guide ver 3 1 Pay attention they have to be surprisingly entered in the white field and not in the grey field as in the case where the terminal were used to interact with the data logger In this user guide the requests are not given in an explicit way For example to ask the address of the gauge it is written to type lt ENQ gt GETADR lt CR gt With this command the terminal understands that an enquiry lt ENQ gt that 1s getting the address 1s made At the end it 20 is necessary to make understand the terminal that t s the end of the enquiry using the Carriage Return command lt CR gt Lar Gee Scripting But these commands have not to be entered 3600 127 Graph emote ha End era on this way in the white window It 1s first Ascii table necessary to translate the commands that are N es d between the lt gt with their ASCII code The 000 000 000 00000000 NUL Null char ASCII table can be consulted directly from 001 001 001 00000001 SOH Start of Header g f 002 002 002 00000010 STX Start of Text the terminal clicking on the corresponding 003 005 003 00000011 ETX End of Text 004 004 004 00000100 EOT End of Transmission button Take onl
70. l in the same minute than WABS Moreover they don t react to the very small weight variations and are only updated each minute They are thus probably the last part of the filter mechanism that treats the raw weight in order to provide right rainfall intensities The weight noise WNOIS is an interesting value It is a result coming from the different algorithms that filter the raw data It provides thus an indirect indication about the elements that perturbed the weight measurements as for example the wind Under the influence of wind over the bucket dynamic pressure variations occur above the gauge orifice causing thus weight fluctuations The noise values could thus be used as indicator about the wind strength Some specific tests about the noise are presented in the chapter 3 3 16 Noticed problems Sometimes a lap of 6 seconds between two measurements is indicated It is normal because looking at the figure 13 1t appears that the program send not exactly each 5 seconds 5 006 or 5 007 a query to get the data from the gauge It can also sometimes occur that no information 1s sent for up to 20 seconds but it s really rare 2 5 Data provided with a temporal resolution of 1 minute This resolution is available when the data transit through the data logger Because it works as well as modem the data logger 1s able to send the measurements to a server thanks to a GPRS General Packet Radio Service connection It enables fas
71. lained In a further part the different data provided by the gauge as well as the way to recuperate it will be briefly presented for the both possible temporal resolutions Indeed the gauge can provide data each 5 seconds when it is directly connected to a computer On the contrary when the data transit through the data logger the measurements are sent to a server with a resolution of 1 minute Finally a presentation of the terminal that allows dialoguing either with the rain gauge or with the data logger will be made Warning If you don t want to spend time to read the following treasure of literature just remember to ALWAYS set the white plastic transport rest during each manipulation that could induce shock on the strain gauge bridge transport and emptying or installing manipulations More information concerning this transport rest is available in the chapter 2 2 2 2 1 General points Different types of rain gauges are available to measure the precipitation intensity The final report about WMO laboratory intercomparison of rainfall intensity gauges 2 proposes a detailed classification and description of different available type of rain gauges To summarize there are two main types of rain gauges the catching and the non catching instruments The first group measures the water equivalent volume or mass of the precipitation falling through an orifice of precisely known dimension The intensity is derived considering the amount of acc
72. ld lead to short memory On the contrary longer memory should characterize the December months when rather frontal rain presenting long duration and low intensity occurs 4 3 4 1 July time series According to the attempts the selection of 10 of July months provided an autocorrelation function that decreases rapidly After about 75 lag of 10 minutes the coefficients fluctuate around O see figure 82 The scale based analysis was conducted on the whole data set allowing obtaining a maximal aggregation period of about 10 months sample Autocorrelation Function ACF 100000 a 5 0 8 Se San ae WON ENED Oe we HOG A EE Ee Dek te Ei ie PS EL aa dee E 1000 amp 0 6 ei Tad DI Oa AAS hue geen Mie mrad ERES wanted RA She ate bie 5 8 ig A A A A A E E S MEEI a es a 0 1 a 5 0 0 001 0 2 i i aggregation minutes O 200 400 600 800 1000 Figure 82 Left ACF Right Aggregation scales between 20 minutes and 10 months The analysis of the variance patterns illustrated different effects First for the first lags the before observed decrease seems not to be present Indeed the two power law fits obtained for the aggregat on scale considering 20 minutes to 1 hour and the 1 hour to 10 hours see figure 83 provide quite similar B values Moreover these B values are quite small but with 20 minutes aggregation the inner regime should be already behind 10 1 sample
73. le set of _ data it appeared that these discrepancies ns W ABS Noise concerned the first rainfalls recorded by the Time am mm min data logger after each time that the gauge So 28 L was disconnected Otherwise it appears that ia ec the gauge works in the same manner than in 20211 0l 0 057 1050 46 1 638 the 5 seconds resolution case Because a lot of manipulations were done during this test 20 23 0 266 0 235 1059 27 phase and that each week end the gauge 20 24 0 196 0 3185 1065 64 2 681 should be disconnect and kept inside a great 20 25 0 348 0 2315 1070 27 2 184 proportion of the recorded data suffers from 20 26 0 188 0 2415 1075 1 3 these discrepancies 2 946 It is difficult to say if this problem only concerns the intensity value Some clues tend to indicate however that the other measurements are not affected First the time indication stays always correct Moreover the noise and the absolute weight values seem not to be concerned 24 by this problem Indeed looking at the table 10 1t appears that the higher noise value at 20 20 attests that 1t was already raining as indicated by the difference of the absolute weights The laps of time as well as the 0 12 amount or the characteristic of 04 precipitation needed to again obtain good results is not clear On the pa ene figure 21 three distinct rainfall intensity events prese
74. mm min g g g 16 54 41 0 217 1647 719 1652 259 3 975 16 54 46 1647 719 1655 398 3 975 16 54 51 1647 719 1655 259 3975 16 54 56 1647 719 1656 934 3 975 16 55 01 1647 719 1660 025 3975 16 55 06 1660 464 31975 1650 72 1662 947 MAI 16 55 16 1650 72 1664 016 MMM 16 55 21 1650 72 1666 907 M 16 55 26 1650 72 1666 847 MMM 16 55 31 1668 865 B 16 55 36 1672 232 11 024 16 55 41 0 593 1659 582 1672 232 11 024 16 55 46 1659 582 1675 258 11 024 16 55 51 1659 582 1680 252 11 024 16 55 56 1659 582 1682 716 11 024 16 56 01 1659 582 1683 724 11 024 16 56 06 1685 791 11 024 16 56 11 11687 319 157887 16 56 16 1668 756 1688 092 SES 16 56 21 1668 756 1688 854 BEST 16 56 26 1668 756 1689 102 15287 16 56 31 1689 943 SEST 1689 489 Looking at the 5 seconds resolution noise data 1t appears that the noise value stays constant for a period where WABS also stays constant Moreover 1t seems that the noise values describe the conditions that occurred 25 or 35 seconds just before that WABS s updated See the different colours in table 13 linking a jump in WABS with the noise values Indeed logically great differences between two consecutive WABS signifying thus high intensities are linked with higher noise values But as already mentioned WABS presents a delay with the real time WR
75. n It would be interesting to look at the wind features that occurred at this time Finally maybe the resolution provided by Climap is not exactly the same than the rain gauge and could explain thus some differences However generally the fit is good The ameliorations provided by the MPS gauge concerning the assessment of the small scale rainfall patterns in time are huge 28 3 2 Tests about temperature The rain gauge provides two different temperature measurements An external sensor indicates the air temperature Another sensor situated on the strain bridge indicates the temperature inside the gauge It 1s an important information needed by the filter mechanism of the rain gauge in order to delete the influence of the temperature on the weight measurements Indeed the resistance measured by the strain bridge presents a sensitivity to the temperature This internal temperature serves as well to engage the heating ring when the ambient temperature is below a threshold defined by the user It could seem strange that the external temperature doesn t serve as indicator for engaging the heating but this external temperature is only an option that is not available on all the gauges 3 2 1 Temperature accuracy In this part the values provided by these two sensors are verified over a wide range of temperature thanks to different additional button temperature sensors To simulate temperature ranges between O and 40 degree Celsius a
76. n Payerne see figure 27 0 04 meteo swiss moving average MPS 0 03 mm min 0 02 0 01 Y U 0 100 200 300 400 500 600 700 800 900 1000 minutes Figure 27 Comparison between the records of the both rain gauges The measurements from the meteoswiss gauge were artificially extended to 1 minute resolution It seems that the MPS rain gauge is able to recognize before the beginning of the rain It appears logical because the meteoswiss device is a tipping bucket rain gauge that probably has a lower detection resolution The tipping bucket has first to be filled This delay didn t appear for the second snowfall probably because the bucket was already almost full The MPS indicated for this event an accumulated rainfall depth of 5 882 mm The meteoswiss indicated only 5 1 mm It represents thus a positive difference of about 15 To interpret these discrepancies more information is necessary about the rain gauge used by meteoswiss as well as the characteristics of the records provided by Climap Different factors could explain these discrepancies As indicated in the introduction the tipping bucket gauges can suffer from underestimation but it is also possible to correct these errors Different catching properties of the both devices could be also mentioned Indeed some discrepancies between the two gauges seem to be present between the 700 and the 800 minute of the observatio
77. nd end intensity intensity rainfalldepth rainfalldepth x x 6 plot of the accumulated rainfall depth and the intensity figure 1 haxes depth int plotyy x rainfalldepth x intensity axes haxes 1 ylabel rainfalldepth mm datetick x 13 xlim x 1 x length x axes haxes 2 axis ij Sif you want to inverse this axe ylabel intensity mm min datetick x 13 xlim x 1 x length x 6 plot of the autocorrelation function ACF Lags Bounds autocorr intensity length intensity 1 figure 2 autocorr intensity length intensity 1 2 Statistical analysis sample variance at different aggregation scales scales 12 353 46 Ve Sy Ov Wp Wy 12 15 10 21 24 30736041870 Ol Ao x axes for 5 seconds resolution data Sxaxes 5 scales 0 10 gt 4 x axes for 1 minute resolution data xaxes 60 scales 12 calculation of the sample variance for each aggregation time case without overlapping for i 1 length scales limit i fix length rainfalldepth scales i fix a limit in order to avoid that the program reads values that don t exist begin 0 for nseparate 1 limit i varintermediate i nseparate var rainfalldepth begin 1 nseparate scales 1 begin nseparate scales 1 end varscale i mean varintermediate 1 1 limit 1 end Ao Case with overlapping for j 1 length scales limit2 3 len
78. nected to the GPRS What is the logger internal time cclk yy mm dd hh mm ss What is the last measured data attlast attention without short or long response More information in the original user guide time What is the strength of the GPRS A number between 0 and 32 Below 10 signal some problem for the data transmission can occur There s also a maintenance mode that s only accessible with a password To switch to the operating to the maintenance mode the following command must be entered at mtnc password The password 1s 1949 The terminal will indicate that the maintenance mode is now activated trough the following answer maintenance logged mode When the maintenance is open it 1s possible to change the time settings or to obtain some information about the preceding maintenances See table 5 Table 5 Some commands available in the maintenance mode Operation or request Command to type Change the time setting at time yy mm dd hh mm ss Attention the logger restarts and all data records are deleted What is the number of restarts a getparam restart What is the date and time of the at getparam stamp last restart Moreover in the maintenance mode it is possible to access to different parameters for consulting their current value and f needed to set new values The procedure for consulting changing and saving the parameter values is simple as visible in the table 6 Don t
79. nent tau q among the different order of moments q This example is not very good because it considers the extended long memory event It is indeed not sure that the use of this short data set 1s relevant On the contrary 1t seems to be rather a strange data set looking at the obtained behaviour of the exponent tau q among the different order of moment Indeed in the most data set presenting multi scaling the observation points were rather above the straight line Computing this graph could thus maybe provide a kind of control about the validity of the used data set Other indications about the validity could surely be found analyzing the autocorrelation function The small data sets tend to often present some negative autocorrelation after a defined number of lags Such patterns are probably not to find in longer data sets In closing 1t would also probably be interesting to assess 1f all the different moments present an inner a transition and a scaling regime Moreover 1t would be probably indispensable to check that the aggregat on scales at which these different regimes appear are always located at the same place for the all different moments and that the observed changes are similar Otherwise f the multi scaling s assessed only thanks to parameters that take into account the behaviour of the tau q exponents 1t could maybe lead to supplementary extrapolation errors 66 References 1 2 3 Eidg Departeme
80. nnnnnnsennnnnnennnnnnnenn 14 2 5 DATA PROVIDED WITH A TEMPORAL RESOLUTION OF 1 MINUTE csssscccceecessesececeeecesseseneceeceesseseeeeeeeeees 17 2 6 PRESENTATION OF THE TERMINAL V TB aii ati 18 2 01 COMMUNICATION with The data LOGO CF WAS AAA Ai 18 2 0 22C0oMMUNLEANON WHN THE BOW GS A A A 20 PERSP TESTS ti au 22 SLES TSA BOUT RAINFALE INTENSITY 2 en na 22 III COLCULAIION OF META nee sich ates oO ac aoe ieee dO O 22 Dobe bel ate With seconds FeSO MILO Ma aos 22 31 12 Dua lo AAA la ee nee 24 3 1 1 3 Elimination of Unreal step of weisht ae a u aparece asec 26 Id 2 TEN SUN ECU A A i n 26 32 TESTS ABOUT TEMPERATURE is 29 eL EMPERO ACCUTANE AA 29 32 TOMEI AMINE ILENE ON ANC WEA A A assay uy AER 30 DOES FS ABOUT NODE ra ee ea ee ee eh 36 II COLUTES Of The NOS Cda 36 I 2 WIN ROSE VEAN near A Codecs welds 37 3 3 2 1 Assessment Of he groundnoisser ana a a E ee 38 32 EW UG ICC CG MOISE ann e es e 39 a Assessment olhe we ohl influence we ale er 39 b Analysis of the recorded noise and wind data e cceseeeessssssneenessnnnnnnnnnnsnnnnnnnnnnnnnnnnnnnnnnnsnnnennnnnsnnnnennsssnenen 40 c Prediction of the wind velocity from the 5 minute mean noise Values oooooccccnonocnnncnnnononnnnnnnonnnnnnnncnononnnnnnns 41 d Assessment of the robustness of the threshold selection oooconnooococcnonoooonncconononnnnccnnnnnononocnnnonononccnnnononcnnn 43 e Possible future amelioration in the threshold selection algorIt
81. nt f r Umwelt Verkehr Energie und Kommunikation UVEK 2008 Hochwasser 2005 in der Schweiz Synthesebericht zur Ereignissanalyse L Lanza M Leroy C Alexandropoulos L Stagi W Wauben 2005 WMO Laboratory intercomparison of rainfall intensity gauges A Molini 2007 WMO Field Intercomparison of rainfall intensity gauges Data Manager report 4 Duchon C E and G R Essenberg 2001 Comparative Rainfall Observations from Pit and Aboveground Rain Gauges with and Without Wind Shields Water Resour Res 37 12 3253 3263 5 Marani M 2002 On the correlation structure of continuous and discrete point rainfall Water Resour Res VOL 39 NO 5 1128 doi 10 1029 2002WR001456 2003 6 Marani M 2005 Non power law scale properties of rainfall in space and time Water Resour Res 41 W08413 doi 10 1029 2004WR003822 Internet General information about the Apunch Project http www swiss experiment ch index php APUNCH Home 67 Appendix A Sprinkler tests validation Ai Effect of non similar rain drop properties One reason of inadequate sprinkler tests could be that when small intensities are simulated t is rather thanks to isolated large drops that are falling intermittently on the gauge On the contrary under natural condition such intensities could also be obtained through small water drops that fall regularly and frequently in the gauge These different drop structures could create different noise
82. nting sometimes high indicated intensities were not sufficient On 0 02 en the contrary as visible on the figure o Mahn 22 and the table 11 after a light 06 00 07 00 08 00 09 00 10 00 precipitation during about 2 hours the data logger was able to self Figure 22 self recalibration of the data logger recalibrate Table 11 Zoom on different time of the rainfall indicated calculated MES Intensity ne As already mentioned for the good part of Pen 0 cm a ey EA T transmitted data it appeared that the m a h bae de he E d to t e one obtained in the seconds resolution case Indeed for this example the indicated intensity from 08 43 corresponds to the calculated one The same little Doae A a ee discrepancies as the 5 seconds resolution osa ol 5667 35 0 069 ase concerning the beginning of the storm 08 44 o 0 0065 5667 48 0 108 are also well visible the gauge needs again a threshold value of to begin recording the rain Finally even if the data logger needs some time to calibrate itself after a restart looking at the general features of the indicated precipitation it appears that there are quite comparable with the one provided by the variation of weight Moreover the total amount of precipitation is the same in the both cases So it seems not to be a major problem all the more that it is possible to find the correct intensity making the calculations from the weight differences th
83. observed one On this way the program would provide wrong results display press enter to quit pause break end selection of the best noise value used to separate the different wind ranges textnoise noiseall noise_05 noise_10 textwind windall wind_05 wind_10 noisemeandtreatednew noisemean5treated WStreatednew WStreated for i 1 length wind_range namenoise textnoise i txt namewind textwind i txt save in a text file each noise and wind values corresponding to the current wind range csvwrite namenoise noisemeandtreatednew csvwrite namewind WStreatednew wind_obs textread namewind wind_obs2 wind_obs lt separation of the wind velocities that are greater than the current wind range wind_obs wind_obs2 gt wind_range 1 1 wind_obs wind_obs2 lt wind_range 1 0 choice of threshold gt building a ROC limit_min min noisemean5treated limit_max max noisemean5treated 300 step step limit_max limit_min 299 threshold limit_min step limit_max nb_thres length threshold for j 1 nb_thres Prediction of the wind velocity according to the noise values all the values over the threshold gt bigger than wind_range i all the values under the threshold gt smaller than wind_range i windspeed_pred textread namenoise sf windspeed_pred2 windspeed_pred windspeed_pred windspeed_pred2 gt threshold 3
84. of the threshold selection In a first attempt the 1 third of data was selected as calibration period The selected threshold values allow obtaining a success rate of 74 1 see figure 48 l HE Corresponding noise threshold 0 055528 0 088439 0 19216 Wind velocity m s OSS i i g Specificity Sensitivity MO predicted right wrong wind speed mis observed O 1 5 1 1 5 2 predicted Figure 48 Above chosen threshold value for the calibration period Below comparison between the so predicted and observed wind speed classes After that the obtained threshold values were applied to the validation period providing a success rate of 69 see figure 49 43 MO predicted right wrong wind speed m s observed 0 5 1 1 5 2 predicted Figure 49 Validation period comparison between the predicted and observed wind speed classes It provides on this case very good results that were even better than the one obtained considering the whole data set In order to better understand why better threshold values are obtained the correlation between the 5 minutes mean noise and the wind speed was analysed for this case see figure 50 It amounts to 0 7775 and s thus more elevated than for the all set of data 0 7363 The correlation characterising the data set could thus play a role in the success rate of the selection To confirm the influence of the quality
85. or thms n order to filter these variations produced for example by the wind or temperature fluctuations or other perturbations After this treatment the green WABS value is obtained This absolute weight value presents logically a more stabile behaviour over time The pink data represents the sensor temperature and the yellow one the one minute intensity that 1s directly derived from the weight changes 1 mm rain corresponds to 20 gram water It is logical regarding at the 200 cm of the orifice In this case the absolute weight represents a rather conservative value of the weight but it doesn t play for the calculation of the intensity because only the changes of weight are taken into account As already mentioned the rain gauge provides in fact more data On the other window usm exe all the data transmitted by the rain gauge 1s indicated see figure 13 The program usm send S a request about each 5 seconds to the rain gauge address 11 The response R contains all the measurements done by the rain gauge IPS iS 26 09 08 10 16 01 215 gt d1B E 26 09 08 10 16 01 585 gt G1160 6A 2286 1038142 2 A 165805 1038273 1037841 1038 261 2368 62 9 8 361 4 A A S668 30 Figure 13 The usm exe window All these data are also stored n a separate text file under the directory MPS tools usm ETH APS data Every time that the program usm is started a new file is created It contains only the measurements made during
86. ping case always allows considering the whole data set For smaller aggregation scales the results are however similar Thus in order to exploit all the data set the case with overlapping will be chosen for calculating the variance pattern among the aggregation scales variance mre with overlapping Variances variances 10 10 seconds seconds Figure 65 Above comparison of the variance curves over the different aggregation scales Below number of considered intermediate variances for each aggregation scale 56 4 3 Results In th s chapter different experiments nvolving always longer data set w ll be presented Moreover it was tried to select event presenting different memory properties in order to compare these results with the observations and the theory from Maran 4 3 1 Sprinkler tests It is particularly interesting to conduct this analysis on sprinkler tests They should indeed represent a field of constant rainfall intensity affected by a noise that is assumed random To check this assumption the sample autocorrelation was computed see figure 66 0 8 Sample Autocorrelation Function ACF nn en IO AN rainfalldepth mm intensity mm min Sample Autocorrelation 0 4 RAR 09 00 00 12 00 00 15 00 00 g Y ae a hom 300 30 400 40 Figure 66 Sprinkler test Left intensity and cumulated rainfall depth Right ACF It appears that the sprinkler test present a little memory pro
87. r statistical test Al nn m o on because it logically decreases with the time see figure 63 rainfalldepth mm intensity mm min Figure 63 Accumulated rainfall depth and intensity 5 seconds resolution data 4 2 2 Influence of the calculation of the scale variance Two different ways of calculating the variance were tested On the first manner overlapping between the different rainfall depth values was allowed leading thus to more numerous intermediate variances than n the case where overlapping was avoided For the both cases the final variance just considers the mean of all these intermediate values see figure 64 as well as the matlab code in the appendix B1 Aggregation with overlapping Aggregation without overlapping Varint scale 2 Varint scale 2 Varint scale 1 Varint scale 1 2 final calculation Varfinal scale mean varscale 1 limit Varfinal scale mean varscale 1 limit2 Figure 64 The both ways tested to calculate the scale variance In this example the scale equals logically 2 comparison with and without overlaping Looking at the results provided by the both ways it appears that the overlapping of rainfall depth provides more convenient final scale variances Indeed for large aggregation scales the case without overlapping begins to not consider all the data set leading thus to quite strange behaviour of the general pattern gt a a my o see figure 65 On the contrary the seconds overlap
88. racy As already mentioned the World Meteorology Organisation WMO organised very accurate tests in order to have an intercomparison of different rain gauges In the first phase different laboratory tests were conducted For each rain gauge precisely known constant water flow was generated and the relative error between the generated and the measured rainfall intensity was assessed All the weighing gauges obtained very good results See figure 24 the devices in green The MPS TRwS rain gauge presented a very small average relative error over the all 26 range of tested intensities confirming thus the 0 1 precipitation range intensity indicated by the manufacturer Welghing gauges 300 E Yokogawa Geonor Serosi i A ee ee l l I measured mm 3 reference mmh Figure 24 Left comparison between generated and indicated intensity for all weighing rain gauges Right average relative error over all the different range of generated intensity The weighing rain gauges are represented in green Source 2 Just to confirm roughly these results we conducted similar experiments Different known weights were inserted in the rain gauge The intensity indicated by the rain gauge was always correct Because the laboratory tests only provide indication about the counting and not about the catching errors of the gauges the WMO did field tests to assess their performance under several climatic conditions As
89. rainfall l a noise g 0 0 2 0 4 0 6 intensity mm min Figure 57 Left assessment of the noise intensity relation for the rainfall event Right correction of the noise values according to the derived relation 49 d Concluding remarks It appears thus that during rainfall the noise characteristics don t allow to derive the wind velocity The obtained results are totally insufficient and prove that the made assumptions were false The spread in the rainfall intensity noise relation can therefore not be caused only by the wind characteristics Thus different other parameters should explain this spread It 1s difficult to assess surely what kind of characteristics could explain this spread because no information is available on the different filter algorithms that provide the noise data However probably that the rainfall variance occurring inside the considered minute plays an important role in the noise response Indeed the tests made about the ground noise attest of a quite oscillating behaviour These oscillations were also observed in other tests during rainfall where also a great increase of the raw noise was followed by an also impressive decrease These observations seem to indicate that a short rainfall pulse creates more weight variations than a regular one and could thus explain the spread of the intensity noise relation sample Autocorrelation Function ACF A er er 0 4 1 0 3
90. re available each 25 and 35 seconds they can not provide a smaller resolution of intensity because the other weight values WCOMP are updated only each minute These other weight values present probably the last stage of the filter mechanism This not exploitable data is however not a critic of the gauge because 1t was explicitly mentioned that this 5 seconds resolution measurements are not a commercial application This resolution was moreover very useful and indispensable to better understand the working style of the gauge In closing this rain gauge should provide 1 minute data of high quality without high supervision needs The whole system presented furthermore a high reliability during the all duration of this first test phase 32 4 Data analysis In this chapter the data provided by the gauge will be analysed with scaling based methods This analysis is inspired by the one described by Marani in the both papers on the correlation structure of continuous and discrete point rainfall 5 and non power law scale properties of rainfall in space and time 6 Before starting with the analysis these both papers will be first roughly and non exhaustively summarized Secondly some preliminary tests are made on the available data in order to conduct the analysis in the more appropriate way Particularly 1t will be assess 1f the 5 second resolution data should be used spite of 1ts noisy behaviour Indeed for statistical analy
91. re considering noise variance values provides success rates that are just a little bit lower see table 17 Moreover looking at all the different wind speed classes the results provided by the both predicator are quite similar It is so difficult to take directly advantage of the variance value 45 Table 17 Success rate of the different predicator for each wind class Predicator Wind speed class 0 67353 0 77704 0 56667 0 67742 0 1875 65706 0 71723 0 57143 0 59322 It should be however possible to further ameliorate the selection comparing where in the data set the both predicators are not able to well predict the right wind speed class Because the noise mean predicator provides better results 1t could be imagined that 1t serves as first principal predicator Looking precisely at the noise variance patterns of the wrong predictions made by the noise mean indicator 1t could be conceivable to implement an additional filter or criterion based on the noise variance properties It could so reduce partly these errors f Concluding remarks After these first tests t appears that the wind velocity can be successfully derived from the noise values with a resolution of 5 minutes Indeed if the best calibration period is chosen more than 70 of the total predictions are right Even if non optimal threshold values are chosen the success rate doesn t go below 66 for the whole period It assesses of the robustness of the selection mechanism
92. re observed are present for the smallest aggregations scales Moreover for the biggest aggregat on scales 1t as well seems that the 0003 slope of the variance pattern becomes a little bit oie E steeper T T 100 y 10000 de 0 1 sample variance mm 2 Figure 72 Variance curve for the aggregated rainfall process Aggregation scales between 2 minutes and 2 hours i 10 These first observations were confirmed fitting y 2E 06x 5 a R 0 9996 different power law functions to the variance curve E 100 10000 among different groups of aggregation scale For T the aggregation between 2 and 10 minutes a value of 1 59 was fitted 2 5 0 001 aggregation seconds Figure 73 Aggregation between 2 and 10 minutes 59 100 y 7E 07x 17828 R 0 9998 y 2E 07x 8707 R 0 9992 sample variance mm22 sample variance mm 2 1 100 0 01 aggregation seconds aggregation seconds 10000 1E 06 100 y 4E 07x 83 u 6E 07x 83 y R 0 9987 R 0 9998 100 sample variance mm42 sample variance mm42 1 100 0 01 aggregation seconds aggregation seconds 10000 1E 06 Figure 74 Above between 10 min and 2 hours and between 2 and 10 hours Below between 10 minutes and 1h15 and between 1h15 and 10 hours 4 3 2 2 Longer memory event Concerning the bigger aggregation scal
93. redicted 0 1 5 0 100 200 300 400 500 600 700 Figure 46 Comparison between the predicted and observed wind speed classes The results are quite good The success rate defined as the percentage of observations that are contained in the predicted wind speed range amounts to 67 35 It is worth to progressively eliminate in the program the predicted and observed values that are above the best selected threshold Otherwise the computation of the sensitivity and the specificity is skewed and lead thus to the selection of non optimal threshold values even if looking at the so obtained threshold better sensitivities and specificities can be obtained see figure 47 In this case the classification would only present a success rate of 64 34 42 Wind velocity m s ificity ivity Specificity Sensitivity DS selection of the best threshold thanks to the ROC plots 1 1 Fe A T US FARA Ihe Cc Cc d d d on uw Ln 0 0 O 0 5 1 0 0 5 1 O 1 5 1 specitity specitity specitity Figure 47 Characteristics of the 3 different selected thresholds without elimination Below The related ROC plots In order to test the robustness of the prediction it 1s interesting to separate this data set in a calibration and a validation parts It is indeed interesting to understand if these results are only good because they consider the all set of data and thus know all the reality d Assessment of the robustness
94. robably the 5 minute resolution 1s not sufficient in order to assess their effect and their influence could thus be diluted among the 7200 weight measurements that occurred in 5 minutes Thus according to these preliminary tests the 5 minute noises mean will be used in order to predict different wind speed ranges c Prediction of the wind velocity from the 5 minute mean noise values The goal of this part is to select threshold noise values that are able to class properly the observed wind in different classes For this set of data presenting low wind velocities the wind will be classed in 4 different classes as indicated n the figure 42 0 5 m s 1 m s 1 5 m s CLASS 1 CLASS 2 CLASS 3 E O Figure 44 Separation in 4 classes of the observed wind A matlab program see appendix B1 is responsible for selecting the noise threshold values that predict in the best way the wind velocity classes The procedure s easy Thanks to the wind measurements the program knows the observed values It can thus easily assess if a proposed noise value would represent a good threshold regarding at different comparisons between the predicted data from noise thresholds and the observed true data So for each class of wind the program considers a lot of possible threshold noise values and just selects the one that maximises the specificity proportion of positive case correctly predicted and the sensitivity proportion of negative case correctly predic
95. rong m NaN else WStreatedwrong m WStreated m WStreatedright m NaN end end successrate success length WStreated 7 Plots and other small calculations B2 Scaling based data analysis PROGRAM FOR BUILDING THE VARIANCE PATTERN AMONG DIFFERENT AGGREGATION SCALES 1 Download and read data derivation of useful values Attention to well pre process the file in excel replace all the by 999 and define the format cell of the date column in the excel sheet as number with 10 decimals date intensitylmin raindur senstemp WABS STATID DEV interntime WRAW bweight cweight exttemp NOISE textread sprinklerwithoutbias txt Sf Sf Sf Sf Sf Sf SE SE SE SE SE Sf S headerlines 4 n length WABS MATLABDate x2mdate date 0 SConverts Excel serial date number to MATLAB serial date number Calculation accumulated rainfall depth mm and intensity mm min from the different WRAW 6 values 5 seconds resolution with noise for 3 2 n rainfalldepth j 1 WRAW j WRAW 1 20 intensity 3 1 WRAW 3 WRAW 3 1 20 12 5 x j 1 MATLABDate j 1 end Calculation accumulated rainfall depth mm and intensity mm min from the different indicated intensity values 1 minute resolution without noise c 0 for j 1 n if intensitylmin j 999 c c 1 intensity c intensitylmin 3 rainfalldepth c sum intensity x c MATLABDate j e
96. s This feature 1s not very pleasant because on this way 1t 1s difficult to control that the gauge works well after an installation To check that the installation was successful 1t s thus necessary to use the additional cable allowing to dialogue with the data logger from a computer Asking for the last measurements allows assessing 1f the gauge works properly The installation of the gauge is easy and doesn t require a lot of time and tools However the solutions should always be adapted to the chosen place and could thus require more time or persons For each situation 1t should be possible to obtain a perfect horizontality of the gauge because the adjustable screw bolts allow a great breathing space Moreover the need of maintenance is really reduced The bucket can store up to 12 kilos of water which corresponds to 600 mm rainfall In the reality because of evaporation the gauge should be able to record quite more precipitation before that the emptying manipulation 1s necessary The dialogue with the rain gauge or the data logger thanks to the terminal 1s not user friendly and presents some times quite strange characteristics It s unfortunately the single way to change different parameters or to install the simcard However with a little bit practice the most important commands are easy usable Concerning the resolution of 5 seconds the measurements are quite noisy and thus unusable Even 1f treated absolute weight values WABS a
97. same that are able to store the non simulated rainfall below 0 01 mm min before that the gauge already recognizes rain Because these additional weights are also corrected according to the evaporation indicated by the variation of WABS the rain gauge is also able to simulate correctly the rain falling after evaporation period Unfortunately it is difficult to completely understand the role of these other weights because they are only available with 5 seconds data resolution and in the conducted tests with this lower resolution no similar case occurred But it is probable that the intensity is directly derived from these other weight differences that surely represent a kind of ameliorated weight measurement It is indeed comprehensible that the manufacturer decided to not provide these additional weight values with the 1 minute resolution data If as above assumed the intensity is directly derived from the variation of these other weights they would give the same information than the intensity Moreover because of all the included filter mechanism in these measurements they maybe could sometimes not correctly represent the present amount of water in the gauge and thus could make problem for example for the decision concerning the emptying time So the choice to give the WABS and the intensity 1s interesting because it allows as well to detect where anomalies appeared and thus to check if the corrections were Justified 3 1 2 Intensity accu
98. scales from 2 minutes to 10 minutes Right from 10 minutes to 7 hours This result is really strange because it doesn t respect the observational and theoretical conclusions of Marani concerning the inner regime In order to check if this strange effect is maybe related only to this sprinkler experiment another shorter sprinkler test with less intensity was also conducted Looking at the autocorrelation coefficients characterising this sprinkler test 1t seems that for this second sprinkler test no memory effects are present see figure 69 60 0 5 sample Autocorrelation Function ACF Ea o 08 o rainfalldepth mm intensity mm min Sample Autocorrelation 0 2 9 20 40 e 50 100 120 140 O 16 00 00 17 00 00 18 00 00 i Minutes Figure 69 Second sprinkler test Left intensity and cumulated rainfall depth Right ACF Looking at the values obtained for the different scale variances t appears that exactly the same behaviour as before is present among the different aggregation scales Also for the small aggregations the fit with a power law provides a B value smaller than 2 The only difference is in the absolute values of the sample variance It s a little bit smaller but it is logical looking at the weaker intensity see figure 70 10000 y 1E 05x 1 8096 a _ R 0 9996 9 100 E o E 2 sprinkler 8 2 test 5 aa S j m smaller S o E intensity 2 a 10000 1000
99. see figure 86 1 10 10000 1 6202 y 5E 06x 2 or 2 N a y 7E 06x R 0 9999 E 3E 05x 19 lt R 0 9996 z E ir E E R 0 9999 E 100 Q g 100 0 01 2 Dj E E S 2 2 gt a Q E no 1 5 0 1 100 10000 100000 0 0001 0 001 0 aggregation minutes aggregation minutes aggregation minutes Figure 86 Different intervals of the whole aggregation range Left 20 minutes to 1 hour Middle 1 hour to 10 hours Right 100 hours to 15 days 64 The value fitted for the highest aggregation scales s quite s milar to the one obtained for the month of July It appears a little bit surprisingly considering the first assumptions about the memory process that should be longer for the month of December However looking at the autocorrelation function of th s time series see figure 87 large differences with the one from July are not iin visible The decrease 1s less rapid for the smallest lags but O 200 400600601000 also after about 100 lags of 10 minutes the coefficients Figure 87 ACF for the December begin fluctuating around 0 Thus the small difference time series between the values fitted for the highest aggregations scales of the both time series 1s not absolutely an anomaly Sample Autocorrelation Function ACF Sample Autocorrelation 4 4 concluding remarks The obtained results correspond only partly to the theoretical and observational
100. sis this high resolution 1s very interesting Finally the results conducted on different data sets will be presented Indeed for comparison purposes sprinkler tests and two ra nfall events presenting different behaviour are analyzed Because the testing phase was too often interrupted in order to obtain long and good set of data the historical rainfall measured in Zermatt will also be used 4 1 General points In the last 40 years numerous researchers studied the relationship between the statistical characteristic of rainfall measured at relatively long aggregation scale and the properties observed at shorter aggregation scales 5 The motivation of this research is easy comprehensible Indeed f a relation 1s found the meteorological data at larger aggregation scale for example daily data could be exploited in order to become information about rainfall properties at smaller aggregation scales Thus the daily data which is already from a long time available in numerous regions could for example be used to generate thanks for example to downscaling approaches hourly resolution data that is absolutely required for a lot of different hydrologic purposes Different scientists found a power law structure over the different range of aggregation of the rainfall statistical moment Marco Marani showed however that this power law scaling can t hold over all the aggregation scales He made detailed research about the variance patterns of the
101. station In this case just one station with the following characteristics is available see figure 10 To start collecting the data click on the ok button The measurements will be graphically a presented in real time with a resolution of 5 seconds a ir moo 4 oO I Figure 10 The address per default of the station is 11 To set a new address see point 2 6 2 In the per default configuration only 4 different measurements are graphically represented It is possible to select in the config menu other or more data It is as General Graph Database well good to increase in the settings menu see figure Monin ips 11 the maximum and optimum values for example Optimum 1100 Fe elne 1200 and 1100 otherwise only the measurements of the ve two last minutes are v s ble on the graph Note that all these changes have to be done before clicking on new Cancel Save settings otherwise they will not be applied Figure 11 The settings menu The figure 12 presents the graphic of the measurements obtained with the default configuration The grey value WRAW represents the raw weight measured by the gauge This value is derived from 120 other measurements 24 measurements pro second It has a very fluctuating behaviour because of the high sensitivity of the weight sensor fuer Figure 12 Measurements with 5 second resolution 14 The MPS Company developed a lot of complicated alg
102. sting indication about the choice of the calibration period In order to obtain the best thresholds values a wind class should not be represented in quite negligible proportion A period presenting a high correlation seems as well to ensure a better selection for the whole period Logically this calibration period should also contain the highest wind measurements Concerning the program 1f the maximal observed wind velocity 1s smaller than the highest chosen wind class an error message will appear and the program will quit Indeed in this case the program would provide non optimal threshold values for this last class leading thus to a less good wind classification from the noise values For more information see the matlab code in the appendix Bl Thus even if it is a little bit disappointing that the program is not able to select the best general solution considering all the data set these experiments showed that choosing a calibration according to defined criterion allow ameliorating the success rate of the whole classification e Possible future amelioration in the threshold selection algorithm There are however also other means in order to increase this success rate It would be for example maybe possible to ameliorate the selection of these threshold values looking as well as the noise variance patterns Surprisingly even if the correlation the correlation is less pronounced between the noise variance and the wind speed the same procedu
103. t are already sent to the data logger After this time without communication the logger will resent all the historical messages and it takes a few hours But no data will be lost Moreover sometimes some part of the information sent by the data logger is missing or incomplete It occurred only during a week end where on the total 10 minutes were missing and 26 minutes presented incomplete or empty information The MPS Company said that this type of problem can occur but stays really rare Finally it appeared as well during this test phase that the data logger seems to need time after an unplug of the gauge before sending correct values Indeed the first precipitations indicated by the data logger present a curious behaviour Detailed information about this problem is available in the chapter 3 1 1 2 17 Lethe 2 AA Concerning the server the provided graphs AJ just link the different measurement points with a line So for example if for a certain time no data were sent to the data logger the BFA TA graphs on the server will link the last old Figure 15 Problem with the graphs provided by the server 2 6 Presentation of the terminal v 1 9b f and the first new measurements On the figure 15 1t appears that wrong temperature values are indicated on the straight line The terminal has several functions Through different commands it is possible to either communicate directly with the rain gaug
104. t significant deformation the behaviour of the raw weight values WRAW It is supposed that the characteristics of the real intensity are well catch by the behaviour of the raw weight Even for the biggest observed rainfall peak of 1 2 mm min that occurred moreover very quickly the filtered weights were able to reproduce without any deformations this peaky pattern see figure 20 23 D D O eb 5 1 61 121 181 241 301 1 61 121 181 241 301 seconds seconds Figure 20 Left Raw weight and treated weight in real time Right Removing the delay of 1 minute affecting the WABS values it appears that the rainfall patterns are very well reproduced 3 1 1 2 Data with 1 minute resolution The same comparison was made for the 0 8 case with one minute resolution data 0 7 For numerous rainfall events the 06 indicated and calculated intensities were E takbutted quite different as for example for the a2 intensity one represented in the figure 21 The 0 2 dicated general feature of the rainfall is well 0 1 intensity caught but looking at the individual 0 O E am N 22 19 intensities quite high differences occur see table 10 Figure 21 Discrepancies between indicated and calculated intensities Table 10 Discrepancies between caluculated and indicated rainfall intensity After looking carefully at the who
105. t the Klein Matterhorn in the Visp valley In order to calibrate and validate the measurements performed by this radar rain gauges with high temporal resolution are necessary Moreover thanks to smaller stations connected to a main rain gauge 1t will be possible to analyze the small scale variance of precipitation The goal of this Projektarbeit 1s to present and assess the performance of the device that will be used as main rain gauge On the first part of this work a small user manual will briefly present how to use the gauge and the different delivered programs Secondly different tests on the measurements provided by the gauge will be conduct They will improve the general understanding of the gauge and allow as well to assess and test the accuracy of the different sensors In the last part of this Projektarbeit the data provided by the gauge will be analysed with scaling based methods Looking at different types of data set a particular interest will be paid to the non power law scale properties in time of rainfall 2 User Manual of the TRwS The following manual s based on the already existing user s guide provided with the gauge ver 3 1 It tries to describe in a more detailed way some features of the rain gauge TRwS Total Rain weighing Sensor First the installation of the different mechanical elements of the rain gauge will be shortly presented After that the role of the different electronic components will be exp
106. t the radiation play an important role in the measurements of temperature It leaded to an increase of some degrees when the sun is present There were also 29 more temperature variations according to the instantaneous characteristics of the sun light cloud effects For the next experiments a protection housing of the external sensor temperature was built Some holes should allow the air to circulate around the sensor It will logically not provide ventilated temperatures but at least the effect of the sun will be reduced To assess the improvement carried by this protection comparisons will be available when the gauge will be installed nearby existing meteo swiss stations where also ventilated temperature measurements are avallable During the same experiment another button sensor was used to check the accuracy of the internal rain gauge sensor The results of the comparison indicate that there s a constant difference between the two measurements Because 1t was not possible to place the button sensor at the same place than the one from the gauge this test was redone inside On this way Annan without the bucket and the housing t was sensor possible to place the button sensor exactly at the same place It provided however the same results The temperature sensor of the strain br dge constantly underestimates the Figure 29 Bias affecting the intern sensor temperature with a bias of about 0 6 0 7 degree see figure 29
107. t to be properly applied an anemometer is thus absolutely necessary 50 3 4 General appreciation of the rain gauge Globally the rain gauge provided always very satisfactory results concerning the rainfall intensity On all the different tests made only rare and very small deformations of the rainfall were observed Moreover these deformations didn t lead to under or overestimation of the total amount of rainfall The WMO conducted a lot of accurate laboratory tests that confirm the very high accuracy of the indicated intensity Some rough tests just confirmed these excellent results concerning the counting errors Unfortunately the made also by the WMO comparing the performance of different rain gauges n field conditions are not yet available It is therefore difficult to become an idea about the possible catching errors of the gauge The next test phases in Payerne and Zermatt will allow comparisons with other gauges However during all this first test phase the gauge was always able to provide consistent intensity data with a delay of 1 minute over all different climatic conditions The single problem concerning the intensity was linked to the data logger Indeed after each restart of the rain gauge the data logger seems to need a warming phase whose characteristics and duration are not well known before indicating correct intensities This phase seems rather to be linked with rainfall features because even if the gaug
108. ted These best predictors are also graphically visible thanks to the ROC plots provided by the program see figure 43 41 selection of the best threshold thanks to the ROC plots 1 4 Poth ajaj 1 A 0 5 0 5 sensitivity sensitivity sensitivity O 1 5 1 0 1 5 1 0 0 5 1 specitity speciity speciity Figure 45 ROC plot 3 different plots are obtained for the selection of each threshold noise value corresponding to the wind velocities of 0 5 1 and 1 5 m s These plots also allow assessing about the validity of the general model chosen to predict the wind speed classes It appears that the model is not perfect because the area below the blue points is smaller than 1 It is indeed logically because here a very simple relation between noise and wind speed is assumed whereas the noise values are quite complex and depend from a lot of other factors But looking at the red right line representing a hazard prediction it appears that this procedure is however able to choose threshold values presenting good specificity and sensitivity patterns Table 15 Characteristics of the 3 different selected thresholds Specificity Sensitivity 0 69 083 058 Ranking all the mean noise values according to the above three obtained thresholds it is now possible to rebuild the wind speed n classes See figure 46 predicted right predictions wrong predictions Wind speed m s observed 0 5 x 1 15 2 p
109. ter data transfer on the GSM Global System for Mobile Communications network If the strength of the GPRS signal is not sufficient more information in chapter 2 6 1 the transmission of data will begin to encounter problem But the logger will not loose these data they will be sent when the signal will become stronger The data logger stores anyway the data even 1f they are successfully sent Only 1f during a long time the strength of the GPRS signal is not sufficient some data will be lost because the data logger begins to rewrite on its memory when it s full The autonomy 1s about 1 month For this test phase the measurements are sent to a server http metnet mps system sk 20002 eth index php belonging to the MPS System Company All the measurements are available with a resolution of 1 minute The data on the server are updated every ten minutes This server is really user friendly All the data can be downloaded for example on an excel sheet csv file It is as well possible to plot online the different values of interest for a selected period Because the data are sent trough GPRS less information is available as in the case with 5 second resolutions Only the time of record the 1 and 5 minute precipitation intensity the absolute weight and the noise values are sent Noticed problems When the data logger didn t have for a long time communication trough GPRS it 1s not possible to have instantaneously the new data tha
110. the rain drop on the housing is quite weak see figure 90 In this case the biggest noise values were to find when the orifice was open because the dynamic pressure fluctuations engendered by the wind again occurred Thus it seems that the housing protects very well the gauge noise phase 1 noise phase 2 1 16 31 46 61 76 Figure 90 Noise pattern comparison 68 B Matlab codes B1 Prediction of wind speed PROGRAM FOR EXPLOTING THE NOISE VALUES IN ORDER TO OBTAIN WIND INFORMATION Because for this experiment the solar panel couldn t provide enough energy some lacks are present in the wind data Thus first find these lacks and eliminate the corresponding noise values provided by the rain gauge clear all 1 Download the data WS WD time textread filename txt Sf amp Sf Sf headerlines 0 time number preciplmin temp WABS precip5m noise textread filename txt Sf sf sf SL Sf Sf f headerlines 0 2 data treatment It eliminates the wrong measurements of the anemometer as well as the noise measurements related to the wrong wind measurements oe oP oO ao Ao Ao ngauge length WABS limit fix ngauge 5 c 0 for i 1 limit if WS 1 0 amp amp WD i 0 wrong measurement c eL WStreated c WS i WDtreated c WD i noisetreated c 1 5 1 5 c noise i 1 5 1 5 i end end
111. the time when the program was running The name of the file indicates when began the measurement datal0242 yyyymmddhhmm It is possible to open and close this text file not the program during the measurements without any precaution the text file is automatically updated Let put an object in the gauge to simulate a pulse of precipitation n order to better present all the different data provided by the rain gauge The object was placed between 10 36 25 and 10 36 30 in the gauge Note that the indicated time is the one from the clock of the computer so check that it 1s right Logically the raw weight reacts directly to this input On the contrary the absolute weight reacts one minute later at 10 37 30 because of the delay involved by the different filter mechanisms The gauge indicates an intensity of about Ilmm for the minute between 10 37 and 10 38 It represents thus a delay of 1 minute because in the reality this input occurred between 10 36 and 10 37 WRAW 1306 8259 Figure 14 simulation of a pulse rainfall 15 The table 3 presents all the recorded data for the example of the pulse rainfall simulation Table 3 data of the text file related to the pulse rainfall simulation 1min Statistic Comp min rain Sensor Total De Devi Internal Raw Base arated Weight Cam Sara ETa 9 TN ht FE En rv FETA ght weight a 0 A PRm Bew mo was _ Sam EY Im RAW _ eran FE a anos hms a le ae Te et at te er an 10 36
112. tion devices considered the transition regime 1s located at the temporal scales of usual hydrologic interest 6 So the use of temporal downscaling based on power law scaling assumption may result into gross extrapolation errors due to the presence of the transition regime 6 Thus n order to avoid these errors the downscaling should take into account the different observed variance patterns among the aggregation scales 4 2 Preliminary tests In this part the results obtained with the both available resolution will be presented Indeed the noise value could be usable because the white noise contained in this data should disappear with increasing aggregation scales Moreover different calculation of the variance for each aggregation scale will be tested These results will allow selecting the data set and the calculation method that provide the most relevant results 4 2 1 Influence of the data resolution It was possible to compare the results provided by these both resolutions for the same data set Indeed in the first case the raw weights indicated each 5 seconds were used to compute the intensity and the accumulated rainfall depth In the second case the treated intensity value provided each minute served to compute the accumulated rainfall depth The figure 60 allows a good comparison of the both set of data mn rainfalldepth mm rainfalldepth mm ensity mm min nt 7 6 5 4 3 2 1 O 1
113. ty on the computer or a adequate USB converter Figure 7 The female part ending of the 9 pin junction cable DB9 2 3 4 OVP2 This part s an Over Voltage Protection device protecting thus the other components from too high voltage on the power supply line 2 3 5 Fuse This little and thin grey component is a device providing protection to the other components f an uncontrollable amount of current flows In this case the fuse would melt breaking thus the path of current flow After that a new fuse has to be installed For that 1t 1s just necessary to unplug the grey box pulling on the little grey handle See picture 8 Figure The fuse inside the grey box 2 3 6 Multiconnector This part s composed from different single connectors To send the measurements via GPRS the data have to transit through the data logger and thus the yellow and green wires should be connected on the places MPSS D and D To have directly the data with a computer just connect these two wires on the PC D and D places See figure 9 For this purpose just take a thin screwdriver and place t on the hole behind the cable and do a little lever movement on your direction During the manipulation it is not necessary to take any precaution such as unplugging the gauge Logically because of these two different places 1t 1s not possible to have in the same time the 5 seconds and the 1 minute data resolution 12 On the figure 9 other wires are
114. umulated precipitation during specified laps of time The non catching instruments measure for example the drop size distribution and the velocity of falling particles and can thus calculate the rainfall intensity or amount by mathematical integration The table 1 tries to highlight the most important characteristics of the different rain gauges types Table 1 The different rain gauges types Type Measuring element particularity _ _ _ _ catching Tipping bucket rain gauges Level measurement rain gauges Weighing rain gauges Drop counters Impact disdrometers Non catching Optical disdrometers There is a tipping balance with two buckets Each tip produces an impulse The intensity is calculated over the period of time between 2 tips The water is collected in a tube of specified diameter The water level provides information about the volume Without funnel the precipitation is directly collected into a bucket and weighed thanks for example to a tensiometer With funnel the water is first collected in a funnel that fills a cylinder whose weight is determined These instruments use a thin nozzle in order to produce single uniform droplets which are counted by an optical system A membrane of plastic or metal is used as measurement surface to sense the impact of single precipitation particles One or two thin laser light sheets detect particles crossing it Each particle blocks the transmitted
115. variance mm 22 oO sample variance mm 22 y 5 05x 14781 2 a u R 0 9998 time minutes time minutes Figure 83 Left sample variance among aggregation scales from 20 min to 1 hour Right 1 to 10 hours For the aggregation scale bigger than 10 hours t seems that the slope character sing the variance patterns presents a tendency to decrease It was confirmed checking at the provided B values that presented as well a decreasing behaviour However for very high aggregation scales this slope begins again to increase This last increase feature is in fact quite logical Indeed it should be not too much aggregated otherwise the variance will begin to mix events of different sorts It seems to be here as well the case even if data coming always from the same month was chosen It is not so surprisingly because the rain patterns can be different from one year to another and as well inside a single month Indeed it would be totally 63 possible that frontal rain occurs as well in July For these reasons it is necessary to not too much aggregate and the variance pattern will be analysed up to 15 days which however represents a quite big aggregat on scale see figure 84 Considering only these smaller ageregation scales the results look quite better Indeed at the first sight a zone where a change in the slope occurs is well visible The power law fit based on the interval 100 hours 15 days indicate
116. y the first value Dec 005 005 005 000001 01 ENG Enquiry i NNE AE ANR MANMAN Ark rarknnwlerment indicated on the left of the Ascii table see Figure 18 The Ascii table of the terminal figure 18 Note that before the corresponding ASCII number the symbol has to be inserted So for lt ENQ gt GETADR lt CR gt 1t must be entered 005getadr 013 The answer will be 11 It 1s the default address Note as well that it 1s not possible to copy and paste the codes the terminal doesn t recognize the command each letter has to be entered However it s very easily to create macros that avoid always taping the codes thanks to the Set Macros button A command of the operating mode in the user guide is not very clear It is the one concerning the time synchronization It s written that when the TRwS 1s used with the datalogger 1t s recommended to synchronise every minute 3 seconds before sending request for data But in fact this synchronisation is done automatically so that nothing special has to be done As for dialoguing with the data logger there is a service mode The access doesn t require a password it is just necessary to enter the following request OPEN lt ADR gt WS lt CR gt it corresponds to OPEN11WS 013 To see the list of all the different possible commands in the service mode enter the letter H On this way it is for example possible to change the heating temperature threshold temperature Note that in

High resolution precipitation intensity

Contents

Download Pdf Manuals

Related Search

Related Contents