Home

SDSM 4.2 - co-public - Loughborough University

Contents

1. TMAXCCF61 90 0UT Figure 9 2 Quantile Quantile plot of maximum daily temperature at Blogsville versus the ensemble mean downscaled from HadCM3 for the period 1961 90 Chart settings can be adjusted selecting the Settings button at the top of the screen Refer to Section 11 3 for information on customising charts in this way PDF Plot The PDF plot provides a Probability Density Function of the selected data files The data are first sorted into order then into categories as defined by the User the default being 20 A count is made of the number of data points in each category The resultant density is plotted on a line chart as shown for example in Figure 9 3 In this case the TMAX DAT data have been distributed into 20 categories Figure 9 4 shows a graph of the same data this time with only ten categories SOSM PDF Chart x axis label Figure 9 3 PDF plot of observed maximum daily temperature at Blogsville for the period 1961 1990 20 categories the ge eb o 9 amp Wack Reset sattings Copy Pine reb SDSM POF Chart x axis label Figure 9 4 PDF plot of observed maximum daily temperature at Blogsville for the period 1961 1990 10 categories Line Plot This produces a simple time series chart of the selected data Note that only a maximum of ten year s of data can be plotted using this option Figure 9 5 provides an example of such a plot In this c
2. D o k Zz 2 5 20 2 S 2 e o o el ar Api Jan Feb Mar Apr May Jun Jul Aug Sep Figure 11 6 The same as Figure 11 3 but with customised bar colour scheme tick marks scale and legend Finally to incorporate Line or Bar charts in a Word document first use the Copy button at the top of the screen then in Word use Paste Special Picture 12 TIME SERIES ANALYSIS 12 1 Time series chart The Time Series Analysis screen allows the User to produce a time series plot of chosen data file s Up to a maximum of five files can be plotted simultaneously on the same chart This screen employs files that contain a single column of data so when using downscaled ensemble output some prior data handling must be undertaken using the Frequency Analysis see Section 9 screen to extract individual members To access the plotting facility click on the Time Series Analysis button at the top of any main screen The following screen will appear Time Series Analysis File Edit Help o Home Quality Control Transform Data J Screen variables J Calibrate Model Summary Statistics Frequency Analyses Scenario Generator Compare Results Time Series Analysis eR s To 01 01 1961 BE 31 12 1990 File Figure 12 1 The Time Series Analysis screen File Selection Using the Drive Directory and File Selection boxes the User can select up to five files to pl
3. Dependent on the realism of GCM Dependent on the realism of GCM boundary forcing e Choice of domain size and location affects results e Requires significant computing resources Ensembles of climate scenarios seldom produced e Initial boundary conditions affect results e Choice of cloud convection scheme affects precipitation results e Not readily transferred to new regions or domains e Typically applied off line therefore results do not always feedback into the host GCM Table 1 1 Main strengths and weakness of statistical and dynamical downscaling The consensus of model inter comparison studies is that dynamical and statistical methods have comparable skill at estimating surface weather variables under present climate conditions However because of recognised inter variable biases in host GCMs assessing the realism of future climate change scenarios produced by statistical downscaling methods is problematic This is because uncertainties exist in both GCM and downscaled climate scenarios For example precipitation changes projected by the U K Met Office s coupled ocean atmosphere model HadCM2 were found to be over sensitive to future changes in atmospheric humidity Overall the greatest obstacle to the successful implementation of both statistical and dynamical downscaling is the realism of the GCM output used to drive the schemes However because of the parsimony and low tech advantag
4. Un conditional Set model Transform process structure variables Calibrate Station and model NCEP data Do wnscale GCM predictand predictors NCEP predictors Weather Scenario generator generator Model output Summary Frequency Compare Time series Statistics analysis results analysis Chart results Figure 2 1 SDSM Version 4 2 climate scenario generation Full technical details of SDSM and downscaling prototypes are provided in the Bibliography Within the taxonomy of downscaling techniques SDSM is best described as a hybrid of the stochastic weather generator and transfer function methods This is because large scale circulation patterns and atmospheric moisture variables are used to condition local scale weather generator parameters e g precipitation occurrence and intensity Additionally stochastic techniques are used to artificially inflate the variance of the downscaled daily time series to better accord with observations To date the downscaling algorithm of SDSM has been applied to a host of meteorological hydrological and environmental assessments as well as a range of geographical contexts including Africa Europe North America and Asia The following sections outline the software s seven core operations along with the UKSDSM data archive and recommended file protocols 2 1 Key function
5. wet days a simple way of classifying trace rainfall days as dry days Therefore different values for the Event Theshold will yield different results in Screen Variables correlation values and scatterplots are both affected will produce different parameters in Calibrate Model and different results from the two Analyse Data operations Note however that the Weather Generator and Scenario Generator operations will still produce values in the range 0 to 0 3 even if the threshold is set at 0 3 Q What are the advantages and disadvantages of using the monthly seasonal or annual Model Type in Calibrate Model The Model Type button in Calibrate Model determines whether individual downscaling models will be calibrated for each calendar month climatological season or entire year The monthly button should be selected whenever the predictand is known to have a strong seasonal cycle noting that even the annual button can produce the same result provided that one or more predictor variables have strong seaonality Annual models are more parsimonious in the sense that they have only one set of regression weights instead of twelve in the case of the monthly models Seasonal models might be used in situations where data are too sparse at the monthly level for model calibration for example a low incidence of precipitation in semi arid regions Q I am trying to model precipitation and have chosen the fourth root transformation in Advanced
6. whether or not data transformation s may be needed and the importance of outliers For the Blogsville example select TMAX as the predictand p_u as the predictor file and February under Select analysis period following the results in Figure 5 2 Check that all other predictors have been deselected and that Unconditional is selected under Process Note that if Conditional is selected all values less than or equal to the Event Threshold in Settings are excluded from the plot Click on the Scatter button at the top of the Screen Variables menu The results are shown in Figure 5 4 Scatter Plot Eile Edit Help February 0 missing value s 2 ncepp__uxx dat Figure 5 4 The Scatterplot for the Blogsville example showing the association between TMAX and p__u in February The results suggest that during February higher maximum daily temperatures are associated with stronger westerly airflows The presentation quality of the Scatterplot may be customized as required by doubling clicking on any of the axis legends titles or data points Additional windows enable changes to be made to chart font style size colour etc To incorporate the Scatterplot in a Word document first use the Copy button at the top of the screen then in Word use Paste Special Picture 6 MODEL CALIBRATION The Calibrate Model process constructs downscaling models based on multiple regression equations given daily weather data the
7. Grid Scale g 3 5 RCM 5 2 S bb E on 5 lt va Precipitation SDS Topography Vegetation Figure 1 1 A schematic illustrating the general approach to downscaling Statistical downscaling methodologies have several practical advantages over dynamical downscaling approaches In situations where low cost rapid assessments of localised climate change impacts are required statistical downscaling currently represents the more promising option In this manual we describe a software package and accompanying statistical downscaling methodology that enables the construction of climate change scenarios for individual sites at daily time scales using grid resolution GCM output The software is named SDSM Statistical DownScaling Model and is coded in Visual Basic 6 0 As far as the authors are aware SDSM was the first tool of its type freely offered to the broader climate change impacts community Most statistical downscaling models are generally restricted in their use to specialist researchers and or research establishments Other software although more accessible produces relatively coarse regional scenarios of climate change both spatially and temporally For example SCENGEN blends and re scales user defined combinations of GCM experiments and then interpolates monthly climate change scenarios onto a 5 latitude x 5 longitude global grid Weather generators such as WGEN LARS WG or CLIGEN see b
8. e a S e Print Help Back Reset E Copy Period Annual Fit Generalised Extreme Value lt TMAXDAT paT MAREE 40 50 60 Return period years Figure 9 8 Generalised Extreme Value plot of maximum and downscaled temperature for Blogsville for the period 1961 1990 For values of k approaching zero 0 005 lt k lt 0 005 in SDSM the two parameter Gumbel distribution is applied using the following equations see Kysely 2002 F x exp exp oa where y is the Euler constant 0 577215655 Gumbel Fits a Gumbel Type 1 distribution to the data using the annual maximum series after the method of Shaw 1994 F x l e 7 Thus the annual maximum for a return period of T years can be calculated from Or 0 K T S o v6f T X K T 2 i m e In which Q is the mean of the annual maximums S is the standard deviation of these maximums K T is a frequency factor T X is the return period in years and y is the Euler constant 0 577215655 Figure 9 8 shows an example of a Gumbel plot using the same data as in Figure 9 6 Stretched Exponential Fits the data to a Stretched Exponential distribution of the form P R gt r expl It is used to calculate the probability that an event is greater than a threshold r RO is the mean of all events and c is determined from the data fitting The data are truncated according to the User specified threshold value Figure 9 10 provides an examp
9. saon LER eter WA Wales a i EE Eastern England l PIE SW Southwest England 510N H ssw se SE Southern England e NI SB IR 2 NE SB EE 2 SW WA SE 2 480N F SW eW sW 30E Figure 2 2 Location and nomenclature of the UK grid boxes in the SDSM archive For model calibration the source is the National Centre for Environmental Prediction NCEP re analysis data set The data were re gridded to conform to the grid system of HadCM3 Figure 2 2 All predictors with the exception of the geostrophic wind direction see below were normalised with respect to the 1961 to 1990 average However daily predictors are also supplied for the period 1961 2000 For downscaling future climate scenarios four sets of GCM output are available HadCM2 HadCM3 CGCM2 CSIRO Three emission scenarios are available the greenhouse gas only experiment with CO2 compounded annually by 1 per year HadCM2 only the two SRES scenarios A2 and B2 produced by greenhouse gas sulphate aerosol and solar forcing HadCM3 CSIRO CGCM2 2 3 UKSDSM predictors Table 2 1 lists the daily predictor variables held in the UKSDSM data archive Ideally candidate predictor variables should be physically and conceptually sensible with respect to the predictand strongly and consistently correlated with the predictand and realistically modelled by GCMs For precipitation downscaling it is also recommended that the predictor suite
10. the Analyse Data function Section 8 check the box under Create SIM File If the User wishes to extract a single ensemble member from a multi column data file check the Extract box on this screen Enter the number of the ensemble member required and the data will be written to the selected Save File Note in this case no transformation is applied to the extracted member Select the Transformation by checking the appropriate button Available transformations include natural logarithms and log10 squares cubes fourth powers inversion lag interval and binomial together with the inverse transformations of the above where appropriate If Wrap is selected for Lag n the last value is used as the first value in the lag transformation otherwise the Missing Data Identifier is inserted note that a negative lag value will shift the data forward a positive lag value will shift the data back The Backward change button is used to compute differences between successive days All transformations can be applied to standard predictor variables prior to Model Calibration Section 6 to produce non linear regression models e g use power transformations for polynomial models For the Eastern England data select Lag n enter 1 in the box and check the Wrap box which will produce a lag I series of the variable with no missing data Click on the Select Output File button An Open file window will appear browse through until the required
11. with the same inputs Even with the same inputs i e PAR file Settings and data period the Weather Generator and Scenario Generator operation is not expected to produce identical results if the Random Number Seed is checked in Settings This is because of the stochastic random component that is applied to each downscaled series to compensate for the fact that the deterministic component of the model due to the chosen predictor variables does not explain all of the observed variance Differences between individual runs and or Ensemble Members is likely to be greater for poorly determined predictands such as precipitation than in better constrained predictands such as temperature Q Does SDSM produce realistic results for multiple sites Also what if I m interested in preserving relationships between variables Both of these questions are the subject of ongoing research However results from previous studies suggest that regression based downscaling does preserve some of the observed inter site correlations provided that models calibrated on a site by site basis are forced by a common set of predictors In other words inter site correlations are implicitly reproduced by virtue of correlated predictor variables rather than by the model structure Alternatively inter site behaviour may be reproduced by employing a conditional resampling approach in which case SDSM is used to downscale a predictand at a benchmark site This serie
12. 1 923 5 459 0 047 Spring 0 134 13 534 0 000 2 621 6 721 0 058 Summer 0 193 23 919 0 000 5 862 8 415 0 071 Autumn 0 150 11 803 0 000 3 112 6 480 0 107 Annual 0 076 22 666 0 000 1 937 14 465 0 030 Figure 10 10 Summary statistics for downscaled precipitation using GCM predictors Using Compare Results Section 11 Figure 10 11 shows for example that the downscaling produced similar monthly mean daily totals under observed NCEP and GCM HadCM2 forcing for the present climate Eile Edit Help a 8 Back Reset Settings Copy Print Help Mean 24 hour totals NCEP HadCM3 Jan Feb Mar T apr May Tun I Jul Taug T Sep T oct Tow bec Figure 10 11 Monthly mean daily precipitation totals at Blogsville for the present climate downscaled using observed NCEP predictors 1961 1990 and GCM HadCM3 predictors 1961 1990 The Scenario Generator operation was implemented for a second time using HadCM3 predictors under present 1961 1990 and future 2070 2099 climate forcing Figure 10 12 shows dry spell lengths plotted by the Compare Results option The results signal a shift to longer dry spells in late summer and autumn File Edit Help Reset Settings Copy Dry spells 1961 90 207099 Figure 10 13 Monthly mean dry spell lengths at Blogsville downscaled using HadCM3 predictors under present 1961 1990 and future 2070 2099 forcing 11 GRAPHING MONTH
13. 26 1315 1337 Huth R 1999 Statistical downscaling in central Europe evaluation of methods and potential predictors Climate Research 13 91 101 Kidson J W and Thompson C S 1998 A comparison of statistical and model based downscaling techniques for estimating local climate variations Journal of Climate 11 735 753 Mearns L O Bogardi I Giorgi F Matayasovsky I and Palecki M 1999 Comparison of climate change scenarios generated daily temperature and precipitation from regional climate model experiments and statistical downscaling Journal of Geophysical Research 104 6603 6621 Mearns L O Mavromatis T Tsvetsinskaya E Hays C and Easterling W 1999b Comparative responses of EPIC and CERES crop models to high and low spatial resolution climate change scenarios Journal of Geophysical Research 104 6623 6646 Murphy J 1999 An evaluation of statistical and dynamical techniques for downscaling local climate Journal of Climate 12 2256 2284 Murphy J 2000 Predictions of climate change over Europe using statistical and dynamical downscaling techniques nternational Journal of Climatology 20 489 501 Salath E P 2003 Comparison of various precipitation downscaling methods for the simulation of streamflow in a rainshadow river basin International Journal of Climatology 23 887 901 Schoof J T and Pryor S C 2001 Downscaling temperature and precipitation A comparison of regression based
14. File Not selected od c 01 01 1961 Figure 8 1 The Summary Statistics screen The first step in the analysis is to select the Data Source click on either Modelled for downscaled data analysis or Observed for observed data analysis The second step is the selection of an appropriate data file Click on the Select Input File button on the left hand side An Open file window appears browse through until the correct directory and file are reached then click on the appropriate file name for example actual maximum daily temperatures at Blogsville are held in TMAX DAT The name of the file will then appear beneath the button If using Modelled output click on View Details to check basic information about the downscaling experiment such as the number of predictors start date etc Next specify the sub period required for analysis using the Analysis start date and Analysis end date windows under the Analysis Period header The default values are the Standard Start Date and Standard End Date held in the global Settings Section 3 1 The default Use Ensemble Mean box produces mean diagnostics for all ensemble members and the standard deviation of the ensembles see Figure 8 4 However diagnostics for individual members may be extracted by deselecting Use Ensemble Mean and by entering the required Ensemble Member in this case integers 1 to 100 in the Ensemble Member box To save the analysis results it is necessary to se
15. Group I Report Terms in italics are found elsewhere in this Glossary Aerosols Airborne solid or liquid particles with a typical size between 0 01 and 10um that reside in the atmosphere for at least several hours Aerosols influence the climate directly through scattering and absorbing radiation and indirectly through the formation and optical properties of clouds Akaike s Information Criterion AIC A measure used to distinguish between two competing statistical models that takes into account the goodness of fit of the model whilst penalising models with larger numbers of parameters See BIC Airflow index Trigonometric measures of atmospheric circulation obtained from surface pressure or geopotential height fields Commonly derived indices include vorticity zonal flow meridional flow and divergence Certain indices have been used to replicate subjective classifications of daily weather patterns or as predictor variables in statistical downscaling schemes Anthropogenic Resulting from or produced by human beings Atmosphere The gaseous envelope surrounding the Earth comprising almost entirely of nitrogen 78 1 and oxygen 20 9 together with several trace gases such as argon 0 93 and greenhouse gases such as carbon dioxide 0 03 Autocorrelation A measure of the linear association between two separate values of the same random variable The values may be separated in either space or time For time series the autoc
16. P 2004 Hydrologic implications of dynamical and statistical approaches to downscaling climate model outputs Climatic Change 62 189 216 Zorita E and von Storch H 1999 The analog method as a simple statistical downscaling technique Comparison with more complicated methods Journal of Climate 12 2474 2489 Miscellaneous Kysely J 2002 Probability estimates of extreme temperature events stochastic modelling approach vs extreme value distributions Stud Geophys Geod 46 93 112 Shaw E 1994 Hydrology in Practice 3 Edition Chapman amp Hall London APPENDIX 1 ENHANCEMENTS SINCE SDSM VERSION 3 1 SDSM 4 2 includes a number of enhancements to version 3 1 sponsored by the Environment Agency of England and Wales Frequency analysis for extremes e Allows the User to fit distributions to observed and downscaled data as either a whole data set or by isolating particular seasons or months GEVs stretched exponential Empirical and Gumbel distributions Results can be viewed in either tabular form or as line charts e User can also plot PDFs of observed and modelled data and Quantile Quantile plots settings allow all charts to be changed e The user can save these analysed results to a text file and a threshold can be applied e A line plot can be made allowing the user to compare observed data with ensembles either as means all ensembles or individual ensembles Step wise regression e Examines all poss
17. Settings What else must I do Nothing The fourth root button in Advanced Settings tells the software that this transformation is to be used throughout including calibration weather generation and scenario generation If checked there s no need to apply any further transformations as this is all backed out automatically So when calibrating the model with fourth root checked you should supply the model with untransformed rainfall data making sure that the Conditional process button is checked in the Calibrate Model screen Q Is it OK to model precipitation as an unconditional process As a general rule precipitation should be modelled as a Conditional process It does not make much sense to neglect the occurrence process 1 e sequences of wet or dry days are first modelled then the amounts if it is a wet day If you are being swayed by higher R sq values of an unconditional model during calibration beware the result is probably seriously biased by the large number of zero values entered in the multiple regression Remember daily precipitation amount is the most problematic daily variable to downscale Q When I use the Weather Generator I get unrealistically large maximum daily precipitation values What s going wrong Unrealistically large values generally imply that the variance inflation and or bias correction in Advanced Settings are too high Q Why do I get slightly different results every time I run the Weather Generator
18. WordPad E TMAXNCEP76 90 0UT WordPad DER Edit View Insert Format Help oH 24 Bo B 113 6 871 12 015 108 11 070 7 764 941 15 645 13 199 7 022 8 132 10 827 11 576 6 391 8 273 8 176 476 10 682 286 10 952 562 4 512 623 617 704 082 3 055 145 871 482 671 264 795 463 629 805 496 451 162 863 284 984 0 407 547 649 953 253 338 905 550 865 633 0 890 3 006 6 148 6 211 4 904 NEOAOMEPUSENHHWAMKEAINDAD wo PHO HOUAALBYHADHANY 8 9 8 6 9 7 1 9 8 8 5 1 0 5 6 Ta 2 IOANHNHHOSHOKOH SS HID IMOMPHORPNYHWT HHP OHH PPE AMO For Help press F1 Figure 7 3 An example of the format of the simulated data file OUT E TMAXNCEP76 90 SIM EBR File Edit View Insert Format Help Cael 64 5 12 366 01 01 1976 5479 FALSE 20 12 1 1 TNAX DAT ncepp__ uxx ncepp__ ZXxx ncepp500xx neepylagxx neepzlagxx For Help press F1 Figure 7 4 The SIM file produced by the Weather Generator operation for the Blogsville example The output is in line order 1 the number of predictor variables 2 the number of regression models used l annual 4 seasonal 12 monthly 3 the maximum number of days in a year here a calendar year is used so there are up to 366 days in leap years 4 the start date of the data used for model calibration 5 the number of days simulated 6 whether or not the predictand is a conditional TRUE or uncon
19. centre of the screen The directories available on this drive will then appear in the window directly above the drive window Browse through again until the appropriate directory is located All DAT files in this directory are then listed in the window above To select a predictor simply click on the file name it will be highlighted in blue A brief definition of the chosen variable is given in the Predictor Description window To deselect a file click on it again and it will no longer be highlighted The number of predictor variables chosen is shown beneath this window up to a maximum of 12 The Data menu on the left hand side of the Screen Variables screen allows the start and end dates of the analysis period to be changed The default dates are held in the Settings screen see Section 3 1 in this case 1961 1990 If the start and end dates lie outside the permissible range the User will be prompted to enter new values The User must also choose the seasonal subset from the pull down window under Select analysis period The available options are Annual no seasonal sub setting Winter December February Spring March May Summer June August Autumn September November and individual months Three more actions are necessary before the analysis can take place Firstly the type of Process must be specified If the predictor predictand process is not regulated by an intermediate process as in the case of maximum temperature
20. contain variables describing atmospheric circulation thickness stability and moisture content In practise the choice of predictor variables is often constrained by data availability from GCM archives The predictors in Table 2 1 therefore represent a compromise between maximum overlap between NCEP and GCM archives as well as a range of choice for downscaling Daily variable Code NCEP HadCM2 HadCM3 CGCM2 CSIRO 1961 GG SRES SRES SRES 2000 1961 1961 1961 1961 2099 2099 2099 2099 Precipitation mm prec x x x x Maximum temperature K tmax x x x x Minimum temperature K tmin x x x x Mean temperature temp x x x x Mean sea level pressure mslp x x x x 500 hPa geopotential height p500 x x x x 850 hPa geopotential height p850 x x x x Near surface relative humidity rhum x x x x Relative humidity at 500 hPa height r500 x x x x Relative humidity at 850 hPa height r850 x x x x Near surface specific humidity shum x x x x Geostrophic airflow velocity KEF x x x x x Vorticity eZ x x x x x Zonal velocity component u x x x x x Meridional velocity component y x x x x x Wind direction th x x x x x Divergence zh x x x x x Table 2 1 Daily variables held in the UKSDSM data archive denoted by x Bold type indicates variables that have not been normalised and are provided for comparative purposes Italics indicate secondary airflow variables derived from p
21. dates are held in the Settings screen Section 3 1 in this case 1961 1990 If the start and end dates lie outside the permissible range the User will be prompted to enter new values Ideally the model should be calibrated using part of the available data withholding the remainder for independent model validation see Sections 7 and 8 To specify the name of the output parameter PAR file click on the Output File button An Output PAR file window appears For maximum convenience make sure that the parameter file is saved in the same directory as the predictand files in this case C SDSM Blogsville Observed1961 90 Enter an appropriate file name in the File name box then click on the Save button The name of the parameter file will then be displayed beneath the Output File button for example TMAX61 75 PAR if data from 1961 75 are used for calibration 6 2 Model type To determine the temporal resolution of the downscaling model check either Monthly Seasonal or Annual under the Model Type box In Monthly models different model parameters are derived for each month In Seasonal models all months in the same season e g December January and February for winter have the same model parameters In Annual models all months have the same parameters i e there is no attempt to specify intra annual variations in parameter values Next indicate whether the downscaling process should be Unconditional or Conditional by checking th
22. example TMAXNCEP76 90 0UT The name of the file will then appear beneath the button Click on the View Details button and the predictand followed by predictor files used in model calibration are listed in the window below The Record Start date and available Record Length number of days are also displayed NOTE PAR files generated by earlier versions of SDSM can still be handled The SDSM version number of the PAR file is displayed along with the process type conditional or unconditional and whether autoregression was selected The User must specify the sub period required for weather generation using the Synthesis Start and Synthesis Length boxes respectively In this case the synthesis spans 1976 90 so Synthesis Start is 1 1 76 and Synthesis Length is 5479 days The default values for Synthesis Start and Synthesis Length are used to simulate the period of record used for model calibration If however model verification is to be undertaken using a set of independent data withheld from the calibration process then the two values should be amended accordingly If simulation of observed data based on the complete predictor record is needed then the Record Start and Record Length values should be used 7 2 Ensemble size Decide how many ensemble members are needed up to a maximum of 100 and enter the appropriate value in the Ensemble Size box at the bottom right hand corner of the screen the default is 20 Individual ense
23. if selected for each statistic for the base period Likewise V2020s is the mean of all ensembles or a specific ensemble for each statistic for period A and so on for V2050s and V20g0s 8 3 The Statistics The diagnostics that can be produced by the Summary Statistics screen are derived for each time period i e Generic Tests Mean Maximum Minimum Sum Variance Median Count Extreme range Minimum range Peaks over threshold Peaks below threshold Percentile Inter quartile range Autocorrelation Skewness Maximum N day total Conditional Tests Percentage wet month season annual as follows Average of all values Largest of all values Smallest of all values Total sum of all values Variance of all values in each time period Median of all values in each time period Count of the total number of values Maximum range of values within a given period Minimum range of values within a given period Number of values greater than or equal to the User specified threshold Number of values less than or equal to the User specified threshold Value of the User specified percentile Difference between the 25 and 75 percentiles Correlation coefficient for successive days Skewness of the data Maximum total accumulated over N days Percentage of days that exceed the threshold Mean dry spell length Mean wet spell length Maximum dry spell length Maximum wet spell length SD dry spell le
24. methods and artificial neural networks International Journal of Climatology 21 773 790 Takle E S Gutowski Jr W J Arritt R W Pan Z Anderson C J Silva R Caya D Chen S C Christensen J H Hong S Y Juang H M H Katzfey J J Lapenta W M Laprise R Lopez P McGregor J and Roads J O 1999 Project to Intercompare Regional Climate Simulations PIRCS Description and initial results Journal of Geophysical Research 104 19 443 19 462 Wilby R L Hay L E and Leavesley G H 1999 A comparison of downscaled and raw GCM output implications for climate change scenarios in the San Juan River basin Colorado Journal of Hydrology 225 67 91 Wilby R L Hay L E Gutowski W J Arritt R W Takle E S Pan Z Leavesley G H and Clark M P 2000 Hydrological responses to dynamically and statistically downscaled climate model output Geophysical Research Letters 27 1199 1202 Wilby R L Wigley T M L Conway D Jones P D Hewitson B C Main J and Wilks D S 1998 Statistical downscaling of general circulation model output A comparison of methods Water Resources Research 34 2995 3008 Winkler J A Palutikof J P Andresen J A and Goodess C M 1997 The simulation of daily temperature series from GCM output Part I Sensitivity analysis of an empirical transfer function methodology Journal of Climate 10 2514 2532 Wood A W Leung L R Sridhar V and Lettenmaier D
25. performance and utility of regional climate models in climate change research Reducing uncertainties in climate change projections the PRUDENCE approach Climatic Change in press Dibike Y B and Coulibaly P 2005 Hydrologic impact of climate change in the Saguenay watershed comparison of downscaling methods and hydrologic models Journal of Hydrology 307 145 163 Frei C Sch ll R Fukutome S Schmidli J and Vidale P L 2006 Future change of precipitation extremes in Europe An intercomparison of scenarios from regional climate models Journal of Geophysical Research Atmospheres 111 D06105 doi 10 1029 2005JD005965 Hanssen Bauer I Achberger C Benestad R E Chen D and Forland E J 2005 Statistical downscaling of climate scenarios over Scandinavia Climate Research 29 255 268 Hay L E and Clark M P 2003 Use of statistically and dynamically downscaled atmospheric model output for hydrologic simulations in three mountainous basins in the western United States Journal of Hydrology 282 56 75 Hay L E Wilby R L and Leavesley G H 2000 A comparison of delta change and downscaled GCM scenarios for three mountainous basins in the United States Journal of the American Water Resources Association 36 387 397 Hewitson B C and Crane R G 2006 Consensus between GCM climate change projections with empirical downscaling Precipitation downscaling over South Africa International Journal of Climatology
26. predictors to model weights The User must also specify the period of record to be synthesised as well as the desired number of ensemble members Synthetic time series are written to specific output files for later statistical analysis graphing and or impacts modelling 2 1 5 Data analysis SDSM provides means of interrogating both downscaled scenarios and observed climate data with the Summary Statistics and Frequency Analysis screens In both cases the User must specify the sub period output file name and chosen statistics For model output the ensemble member or mean must also be specified In return SDSM displays a suite of diagnostics including monthly seasonal annual means measures of dispersion serial correlation and extremes 2 1 6 Graphical analysis Three options for graphical analysis are provided by SDSM 4 2 through the Frequency Analysis Compare Results and the Time Series Analysis screens The Frequency Analysis screen allows the User to plot extreme value statistics of the chosen data file s Analyses include Empirical Gumbel Stretched Exponential and Generalised Extreme Value distributions The Compare Results screen enables the User to plot monthly statistics produced by the Summary Statistics screen Having specified the necessary input file either bar or line charts may be chosen for display purposes The graphing option allows simultaneous comparison of two data sets and hence rapid assessment of downscaled
27. several layers in the atmosphere defined by a regular grid of about 200km resolution Couple ocean atmosphere general circulation models O AGCMs also include ocean land surface and sea ice components See climate model Geopotential height The work done when raising a body of unit mass against gravity i e acceleration due to gravity at a given level in the atmosphere multiplied by distance divided by the value of gravity at the Earth s surface Greenhouse gas Gaseous constituents of the atmosphere both natural and anthropogenic that absorb and emit radiation at specific wavelengths within the spectrum of infrared radiation emitted by the Earth s surface the atmosphere and clouds The primary greenhouse gases are water vapour H20 carbon dioxide CO2 nitrous oxide N20 methane CH4 and ozone 03 Grid The co ordinate system employed by GCM or RCM to compute three dimensional fields of atmospheric mass energy flux momentum and water vapour The grid spacing determines the smallest features that can be realistically resolved by the model Typical resolutions for GCMs are 200km and for RCMs 20 50km Meridional flow An atmospheric circulation in which the dominant flow of air is from north to south or from south to north across the parallels of latitude in contrast to zonal flow NCEP The acronym for the National Center for Environmental Prediction The source of re analysis climate model assimilated data widel
28. study region All data files including NCEP predictors may then be downloaded directly to Users PC for immediate deployment by SDSM 3 GETTING STARTED To launch SDSM click on the Start button on the Windows desktop then on Programs and then on SDSM which will appear as a small rain cloud on the list of available programs The following screen will appear Figure 3 1 The SDSM splash screen Click on Start to continue to the SDSM main menu Figure 3 2 If you do not wish the splash screen to appear in future ie the main menu screen will appear upon starting SDSM click the tick box by Do not show this splash screen again If further information is required at any time click on the Help button at the top of each screen the User may then search the Help Contents by key word or task SDSM is navigated by selecting appropriate buttons from the bar at the top of each screen These are arranged in the same logical order as key functions of SDSM Figure 3 2 Main menu of SDSM 4 2 Before downscaling the User should check the date ranges type and integrity of all input data To establish the working environment click on the spanner symbol at the top of the main menu or at the top of any other screen to access the Settings screen Figure 3 3 3 1 Settings The Settings screen may be accessed throughout SDSM The following global preferences are available Year Length The default Calendar 366 allows 2
29. then click on Unconditional otherwise select Conditional as with precipitation where amounts depend on wet day occurrence Secondly amend the Significance Level as required This value is used to test the significance of predictor predictand correlations The default is p lt 0 05 5 Finally if the User wants an autoregressive term to be included in the calculations the Autoregression option should be selected Once the above have been specified SDSM is ready to analyse the chosen predictor predictand relationship s for specified sub period s 5 2 Temporal variations in predictor strength The Analyse button is used to investigate the percentage of variance explained by specific predictand predictor pairs The strength of individual predictors often varies markedly on a month by month basis see Figure 5 2 The User should therefore be judicious concerning the most appropriate combination s of predictor s for a given season and predictand As stated above the local knowledge base is also invaluable when determining sensible combinations of predictors Results RESULTS EXPLAINED VARIANCE Analysis Period 01 01 1961 31 12 1990 Significance level 0 05 Total missing values 0 Predictand TMAX DAT Predictors JAN neepp__uxx dat 0 348 neepp__vxx dat 0 031 ncepp__zxx dat 0 030 neepp500xx dat 0 088 neepulagsx dat 0 284 ncepylagsx dat 0 030 ncepzlagxx dat 0 013 Figure 5 2 The Results screen for the Blog
30. ty ot gOS NE oN SERS IAN RNa 41 ST SASS T s 3 n D Predicted Value Y Figure 6 3 The Scatter Plot screen Click on the Back button to return to the Calibrate Model screen 6 4 The PAR file During model calibration a PAR file is generated that stores various parameters relating to the structure of the model NOTE Information held in the PAR file can often be used to diagnose the cause of any unexpected model results or behaviour Figure 6 4 provides an example of such a file produced using the Blogsville data set In this file the data are stored in line order as follows 1 2 BW 0 1 2 3 17 m TT I OT OT OT I 5 6 7 8 9 1 1 1 1 The number of predictors The season code 12 months 4 seasons annual model The year length indicator 366 365 or 360 Record start date Record length days Model fitting start date Number of days used in the model fitting Whether the model is conditional True or unconditional False Transformation 1 none 2 fourth root 3 natural log 4 lognormal Ensemble size Autoregression indicator True or False Predictand file name Predictor filenames in this case five 18 29 30 Model parameters the first 6 columns in this example are the parameters including the intercept the last two columns are the SE and r squared statistic The root directory of the predicta
31. 07 Page 72 of 94 The following points explain how each of these options work Lines The User can adjust the width and legend text of each line in the chart After making the required changes the User must click the Make Changes button to apply the changes to the chart Legend The User can choose to show the legend on the chart or remove the legend by clicking the appropriate buttons at the bottom of this screen Y axis ticks These refer to the tick lines drawn across the chart on the y axis The default is no tick lines except for the line y 0 If the User wishes to apply y axis tick lines enter the number required in the text box here and click the Make Changes button Clicking the Clear Y Ticks button removes these tick lines Y axis range The User can adjust the extent of the Y axis by entering appropriate minimum and maximum values on this page and clicking the Make Changes button X axis labels For analysed data year markers are shown on the X axis For raw data it is possible to apply a number of data markers a counter on the X axis X axis label spacing can be specified by entering an appropriate value in the X axis labels gap text box This specifies the interval between successive X axis labels markers The default for analysed data is 1 i e markers appear every year The default for raw data is 0 i e no markers appear Note that when plotting SPI data series the X axis labels are determined by the total number o
32. 1 November 9 723 18 615 0 017 10 517 December 7 004 17 880 4 363 14 309 Winter 6 903 18 647 5 979 13 768 Spring 12 630 25 061 1 008 17 797 Summer 20 145 30 094 8 326 10 930 Autumn 14 026 26 351 0 017 20 249 Annual 13 455 30 094 5 979 37 818 Standard Deviations of Results January 0 105 1 517 1 736 0 902 February 0 114 0 841 0 895 0 740 March 0 114 1 155 1 045 0 860 April 0 111 1 324 0 924 0 525 May 0 099 0 832 1 596 0 502 June 0 070 0 956 1 169 0 652 July 0 074 0 700 0 646 0 317 August 0 104 1 029 0 957 0 575 September 0 102 0 948 0 963 0 502 October 0 081 1 188 0 906 0 557 November 0 108 0 753 1 263 0 451 December 0 110 1 113 1 443 0 628 Winter 0 062 1 315 1 553 0 504 Spring 0 073 0 832 1 045 0 464 Summer 0 053 0 766 1 165 0 319 Autumn 0 060 0 944 1 263 0 555 Annual 0 037 0 766 1 553 0 393 Figure 8 4 Example output of Summary Statistics Modelled showing the mean and standard deviation of diagnostics for a 20 member ensemble 8 2 Delta Statistics Click on the Delta Stats button to calculate Delta Statistics Delta statistics take the form A2020s V 0205 E Vase ij 100 base A205 Os a Vao505 Vase 3 100 base Vase 100 A2080s Cros0s 7 base if Percentage Difference is selected in Statistics or A2020s V9 V base A20508 Vasos V base A2080s V9 V base if Absolute Difference is selected Voase is the mean of all ensembles or a specific ensemble
33. 13 194 232 280 Winter 6 796 17 900 6 700 14 337 575 100 Spring 12 393 26 100 1 600 18 087 1140 193 Summer 20 120 34 800 0 000 14 673 1851 033 Autumn 13 900 27 100 0 600 20 362 1264 873 Annual 13 331 34 800 6 700 39 332 4869 540 Figure 8 3 Default statistics for observed daily maximum temperatures at Blogsville during the validation period 1976 1990 The Results screen lists the name of the input file along with the start and end dates of the analysis Monthly seasonal and annual mean statistics are listed beneath for the chosen tests Comparison of the results obtained from the Weather Generator see below gives an indication of model skill See Section 11 for graphical comparisons of monthly statistics Figure 8 4 shows the summary statistics for the modelled maximum daily temperatures of Blogsville during the validation period 1976 1990 The summary results of these statistics are saved to TMAXNCEP76 90 TXT Results File Help SUMMARY STATISTICS FOR TMAXNCEP 6 390 0UT Analysis Start Date 01 01 1976 Analysis End Date 31 12 1990 Ensemble Member s ALL Month Mean Maximum Minimum Variance January 6 851 17 879 5 319 13 999 February 6 850 16 377 3 060 12 878 March 10 060 20 541 1 008 13 623 April 12 034 21 264 2 775 10 394 May 15 777 25 061 4 632 12 259 June 19 025 29 755 8 329 14 247 July 20 699 28 302 12 132 8 175 August 20 675 29 245 11 757 8 661 September 18 057 26 347 9 359 8 073 October 14 288 23 047 6 323 7 72
34. 17 6 995 15 287 0 000 February 6 710 18 464 4 076 13 280 0 000 March 9 078 21 566 5 422 20 089 0 000 April 12 475 23 133 1 203 12 286 0 050 May 15 934 26 839 4 511 12 038 3 900 June 19 729 31 363 8 519 12 673 63 950 July 20 701 30 477 12 284 8 548 69 200 August 20 194 30 450 9 802 11 794 74 600 September 18 152 28 373 5 173 13 756 22 600 October 13 390 23 705 1 890 13 754 0 100 November 98 910 20 561 2 602 13 391 0 000 December 6 749 18 310 5 606 15 262 0 000 Winter 6 471 19 602 7 427 14 747 0 000 Spring 12 495 26 839 5 422 22 643 3 950 Summer 20 208 31 951 8 403 11 166 207 750 Autumn 13 484 28 373 2 602 27 875 22 700 Annual 13 165 31 951 7 430 42 849 234 400 Standard Deviations of Results January 0 093 1 178 1 583 0 486 0 000 February 0 073 1 561 0 953 0 548 0 000 March 0 092 0 998 1 133 0 683 0 000 April 0 064 1 053 1 118 0 486 0 218 May 0 064 1 028 1 710 0 411 1 670 June 0 080 1 452 1 029 0 361 5 912 July 0 062 1 075 0 659 0 334 5 501 August 0 066 0 748 1 103 0 441 6 184 September 0 065 0 664 1 258 0 552 4 236 October 0 078 0 990 1 042 0 356 0 300 November 0 070 1 153 1 083 0 514 0 000 December 0 080 1 036 1 464 0 554 0 000 Winter 0 060 1 184 1 400 0 240 0 000 Spring 0 039 1 028 1 133 0 393 1 687 Summer 0 042 1 111 0 934 0 211 9 889 Autumn 0 049 0 664 1 083 0 247 4 291 Annual 0 029 1 111 1 398 0 225 11 868 Figure 10 3 Example results for Blogsville using GCM predictors 1961 1990 Second predictors from the HadC
35. 30 204 220 Wilby R L and Harris I 2006 A framework for assessing uncertainties in climate change impacts low flow scenarios for the River Thames UK Water Resources Research 42 W02419 doi 10 1029 2005WR00406S Downscaling overviews and general guidance Christensen J H and Hewitson B C 2007 Regional climate projections IPCC WG1 Fourth Assessment Report Chapter 11 forthcoming Fowler H Blenkinsop S and Tebaldi C 2007 Linking climate change modelling to impacts studies recent advances in downscaling techniques for hydrological modelling International Journal of Climatology in press Goodess C Osborn T and Hulme M 2003 The identification and evaluation of suitable scenario development methods for the estimation of future probabilities of extreme weather events Tyndall Centre for Climate Change Research Technical Report 4 http Avww tyndall ac uk research theme3 final_reports it _16 pdf Leung L R Mearns L O Giorgi F and Wilby R L 2003 Regional climate research needs and opportunities Bulletin of the American Meteorological Society 84 89 95 Lu X 2006 Guidance on the Development of Climate Scenarios within the Framework of National Communications from Parties not Included in Annex I NAT to the United Nations Framework Convention on Climate Change UNFCCC National Communications Support Programme NCSP UNDP UNEP GEF in press Mearns L O Giorgi F Whetton P Pab
36. 5 121 133 Benestad R E 2004 Tentative probabilistic temperature scenarios for northern Europe Tellus Series A Dynamic Meteorology and Oceanography 56 89 101 Burger G 1996 Expanded downscaling for generating local weather scenarios Climate Research 7 111 128 Birger G and Chen Y 2005 Regression based downscaling of spatial variability for hydrologic applications Journal of Hydrology 311 299 317 Cavazos T and Hewitson B C 2005 Performance of NCEP NCAR reanalysis variables in statistical downscaling of daily precipitation Climate Research 28 95 107 Charles S P Bates B C Smith I N Hughes J P 2004 Statistical downscaling of daily precipitation from observed and modelled atmospheric fields Hydrological Processes 18 1373 1394 Crane R G and Hewitson B C 1998 Doubled CO2 precipitation changes for the Susquehanna Basin downscaling from the GENESIS general circulation model International Journal of Climatology 18 65 76 Hay L E McCabe G J Wolock D M and Ayers M A 1991 Simulation of precipitation by weather type analysis Water Resources Research 27 493 501 Huth R 1999 Statistical downscaling in central Europe evaluation of methods and potential predictors Climate Research 13 91 101 Huth R 2005 Downscaling of humidity variables A search for suitable predictors and predictands International Journal of Climatology 25 243 250 McCabe G J and Dettinger M D 199
37. 5 Relations between winter precipitation and atmospheric circulation simulated by the Geophysical Fluid Dynamic Laboratory General Circulation Model International Journal of Climatology 15 625 638 Schmidli J Frei C and Vidale P L 2006 Downscaling from GCM precipitation A benchmark for dynamical and statistical downscaling methods International Journal of Climatology 26 679 689 Schubert S and Henderson Sellers A 1997 A statistical model to downscale local daily temperature extremes from synoptic scale atmospheric circulation patterns in the Australian region Climate Dynamics 13 223 234 Wilby R L 1997 Non stationarity in daily precipitation series implications for GCM downscaling using atmospheric circulation indices International Journal of Climatology 17 439 454 Wilby R L and Wigley T M L 2000 Precipitation predictors for downscaling Observed and general circulation model relationships International Journal of Climatology 20 641 661 Weather generators Katz R W and Parlange M B 1998 Overdispersion phenomenon in stochastic modeling of precipitation Journal of Climate 11 591 601 Kilsby C G Cowpertwait P S P O Connell P E and Jones P D 1998 Predicting rainfall statistics in England and Wales using atmospheric circulation variables International Journal of Climatology 18 523 539 Kilsby C G Jones P D Burton A Ford A C Fowler H J Harpham C James P Smith A
38. 8 35 336 0 001 17 751 43 606 May 3 413 40 340 0 001 24 804 49 815 June 4 072 50 800 0 001 34 787 52 503 July 4 348 48 217 0 001 41 302 50 664 August 4 168 56 362 0 001 41 588 57 456 September 3 750 40 965 0 001 28 262 46 117 October 3 080 34 071 0 001 17 324 43 990 November 3 186 37 392 0 002 19 918 49 407 December 3 282 35 233 0 001 20 626 50 504 Winter 3 090 39 488 0 001 17 917 138 622 Spring 3 024 43 743 0 001 18 371 135 429 Summer 4 191 64 568 0 001 39 286 160 623 Autumn 3 315 45 188 0 001 21 598 139 513 Annual 3 374 65 833 0 001 23 922 578 808 Standard Deviations of Results January 0 150 6 110 0 00 February 0 236 7 738 0 00 March 0 118 5 199 0 00 April 0 151 8 829 May 0 209 10 826 June 0 279 14 049 July 0 231 13 791 August 0 341 15 253 September 0 276 8 306 October 0 155 8 923 November 0 164 7 946 December 0 212 6 692 Winter 0 116 6 728 Spring 0 070 9 800 Summer 0 186 12 154 Autumn 0 113 7 197 Annual 0 055 10 472 2 560 2 511 3 719 3 673 1 753 2 192 2 789 2 369 5 119 3 062 8 536 3 488 8 678 2 944 10 446 5 030 5 904 4 040 3 664 2 601 2 602 3 238 4 002 3 222 1 819 5 653 2 149 3 442 5 613 6 697 2 388 5 599 1 525 8 560 eoooooooeoeeoo Ssooo0o0o00000000 Sseqcesesossoqoesss ececq0000 e00 0 500 Figure 10 9 Summary statistics for downscaled precipitation using observed NCEP predictors Results File Help SUMMARY STATISTICS FOR PRCPCCF61 390 0UT Analysis Start Date 01 01 1961 Analysis E
39. 9 days in February every fourth year i e leap years and should be used with observed data The alternatives allow for different numbers of days in GCM data For example CGCM2 and CSIRO have 365 days and no leap years whereas HadCM2 and HadCM3 have model years consisting of 360 days WARNING Failure to set this parameter correctly can lead to system errors due to insufficient data or the production of non sensible output Standard Start End Date Enter the global default start and end date for all input data These dates will appear throughout the operation of SDSM but may be updated from any screen Allow Negative Values The default allows simulation of negative values by unconditional processes in the downscaling model e g for minimum temperature deselection truncates values at zero e g for sunshine hours Conditional processes e g rainfall amounts are unaffected by this button Event Threshold For some variables it is necessary to specify an event threshold For example when calibrating daily precipitation models the parameter might be set to 0 3 mm day to treat trace rain days as dry days Similarly the threshold for sunny versus cloudy days might be set at 1 0 hours day to discriminate between overcast and sunny conditions Missing Data Identifier This is the code assigned to missing data in all input series Whenever SDSM encounters this code the value will be skipped e g during model calibration or calculation o
40. 90 0UT clearly under estimate the intensity of the observed TNAX DAT hot days with return periods exceeding 3 years Frequency Analysis Plot File Edit Help Pee Back Reset Settings Copy Period Annual Fit Empirical gt TMAX DAT TMAXNCEPE61 90 0UT 12 15 18 Return period years Figure 9 6 Example of an empirical fit to observed and downscaled maximum daily temperature at Blogsville for the period 1961 1990 Frequency Analysis Results File Help Obs 23 800 24 400 24 700 25 600 25 700 26 600 26 700 26 700 26 700 26 900 26 900 27 100 27 400 27 700 27 800 28 000 28 200 28 500 28 700 28 800 28 900 29 400 29 400 30 500 30 600 30 600 31 400 32 200 32 400 34 800 Figure 9 7 Same results as Figure 9 5 presented in a tabular format GEV This fits a three parameter P k Generalised Extreme Value GEV distribution to the data of the form 1 y i F a exp KES p The parameters p k are estimated using the method of L moments in which the first three L moments l l2 13 are estimated from the data see Kysely 2002 The parameters are then calculated according to k 7 8590z 2 955z Lk P gt ras r 1 k 1 E p In which 2 S In2 l In3 l 3 The results are plotted up to a return period of 100 years Figure 9 8 shows the GEV plot using the same data as in Figure 9 6 Frequency Analysis Plot File Edit Help e e
41. C Harpham C Haylock M R Hundecha Y Maheras P Ribalaygua J Schmidli J Schmith T Tolika K Tomozeiu R and Wilby R L 2007 An intercomparison of statistical downscaling methods for Europe and European regions assessing their performance with respect to extreme temperature and precipitation events Climatic Change in press Guangul S G 2003 Modelling the effect of climate and land use changes on hydrological processes An integrated GIS and distributed modelling approach Published PhD Thesis Vrije Universiteit Brussels Belgium Harpham C and Wilby R L 2005 Multi site downscaling of heavy daily precipitation occurrence and amounts Journal of Hydrology 312 235 255 Haylock M R Cawley G C Harpham C Wilby R L and Goodess C M 2006 Downscaling heavy precipitation over the UK a comparison of dynamical and statistical methods and their future scenarios International Journal of Climatology 26 1397 1415 Khan M S Coulibaly P and Dibike Y 2006 Uncertainty analysis of statistical downscaling methods Journal of Hydrology 319 357 382 Khan M S Coulibaly P and Dibike Y 2006 Uncertainty analysis of statistical downscaling methods using Canadian Global Climate Model predictors Hydrological Process 20 3085 3104 Lines G S and Pancura M 2005 Building climate change scenarios of temperature and precipitation in Atlantic Canada using the Statistical DownScaling Model SDSM M
42. CM uses the GCM to define time varying atmospheric boundary conditions around a finite domain within which the physical dynamics of the atmosphere are modelled using horizontal grid spacings of 20 50 km The main limitation of RCMs is that they are as computationally demanding as GCMs placing constraints on the feasible domain size number of experiments and duration of simulations The scenarios produced by RCMs are also sensitive to the choice of boundary conditions such as soil moisture used to initiate experiments The main advantage of RCMs is that they can resolve smaller scale atmospheric features such as orographic precipitation or low level jets better than the host GCM Furthermore RCMs can be used to explore the relative significance of different external forcings such as terrestrial ecosystem or atmospheric chemistry changes 1 1 2 Weather typing Weather typing approaches involve grouping local meteorological data in relation to prevailing patterns of atmospheric circulation Climate change scenarios are constructed either by re sampling from the observed data distributions conditional on the circulation patterns produced by a GCM or by generating synthetic sequences of weather patterns and then re sampling from observed data Weather pattern downscaling is founded on sensible linkages between climate on the large scale and weather at the local scale The technique is also valid for a wide variety of environmental variables as
43. Climate Impacts Scenarios CCIS Group though the Climate Change Action Fund Assistance in kind was provided by A Consortium for the Application of Climate Impact Assessments ACACIA at the National Centre for Atmospheric Research NCAR NCAR is sponsored by the U S National Science Foundation NCEP Re analysis data were provided by the NOAA CIRES Climate Diagnostics Center Boulder Colorado USA from their Web site at http www cdc noaa gov The Climate Impacts LINK Project is funded by the UK Department of the Environment Transport and the Regions Contract Reference EPG 1 1 124 CONTENTS i Preface ii Acknowledgements 0 TECHNICAL INFORMATION 1 INTRODUCTION 1 1 1 2 1 3 Downscaling techniques 1 1 1 Dynamical 1 1 2 Weather typing 1 1 3 Stochastic weather generators 1 1 4 Transfer functions Relative skill of statistical and dynamical downscaling techniques Manual outline 2 OVERVIEW OF SDSM STRUCTURE AND UKSDSM ARCHIVE 2 1 2 2 2 3 2 4 2 5 Key functions of SDSM 2 1 1 Quality control and data transformation 2 1 2 Screening of downscaling predictor variables 2 1 3 Model calibration 2 1 4 Weather generator 2 1 5 Data analysis 2 1 6 Graphical analysis 2 1 7 Scenario generation UKSDSM data archive UKSDSM predictors SDSM file protocols Obtaining SDSM predictors online 3 GETTING STARTED 3 1 3 2 Settings Advanced settings 4 QUALITY CONTROL AND DATA TRANSFORMATION 4 1 4 2 Qu
44. LY STATISTICS The Compare Results operation enables the User to plot monthly statistics produced by the Summary Statistics screen Section 8 Graphing options allow the comparison of two sets of results and hence rapid assessment of downscaled versus observed or present versus future climate scenarios To access this facility click the Compare Results button at the top of any main screen The following screen will appear Compare Results File Edit Help A Home 5 Quality Control Oo Transform Data 5 Screen Variables 5 Calibrate Model 5 Weather Generator Oo Summary Statistics gt Frequency Analyses 5 Scenario Generator Oo Compare Results 5 Time Series Analysis Reset Bar Line Settings Input File 1 Input File 2 File Not selected File Not selected Select statistic Figure 11 1 The Compare Results screen 11 1 Line chart To choose a results TXT file click on Select First File button An Open file window appears browse through until the correct directory and file are reached then click on the appropriate file name for example observed statistics for maximum daily temperature at Blogsville might have been stored in TMAXOBS60 91 TXT The name of the file will then appear beneath the button along with a list of available statistics Repeat the process by clicking on the Select Second File button Then click on the required statistic listed under Select Statistic Finall
45. M3 experiment for the period 2070 2099 were used to downscale future climate forcing Note that as the Blogsville data set contains future GCM data for 2070 99 it was necessary in the Summary Statistics screen to set the analysis period to 1961 90 to cover the thirty year period of the 2070 99 data set Ordinarily SDSM predictors would be supplied for the full period 1961 2100 so this tweak to the dates would not be needed Figure 10 4 shows the Results screen for this scenario using the Summary Statistics operation Results File Help SUMMARY STATISTICS FOR TMAXFCF 0 99 0UT Analysis Start Date 01 01 1961 Analysis End Date 31 12 1990 Ensemble Member s ALL Month Mean Maximum Minimum Variance POT January 6 773 18 997 5 153 14 528 0 000 February 7 853 19 307 3 899 13 239 0 000 March 11 043 24 147 3 155 19 343 0 300 April 14 070 25 194 3 350 11 906 0 750 May 18 161 29 610 7 147 13 987 32 300 June 21 925 33 767 10 570 15 013 195 850 July 22 702 32 339 13 487 9 076 201 550 August 22 911 33 662 12 981 11 243 237 550 September 20 375 31 484 7 374 13 952 94 550 October 15 451 25 610 3 773 13 663 1 100 November 10 009 21 380 1 015 13 055 0 000 December 7 360 20 160 5 427 15 792 0 000 winter 7 329 20 681 5 084 14 719 0 000 Spring 14 425 29 610 3 155 23 588 33 350 Summer 22 513 34 584 10 429 11 960 634 950 Autumn 15 278 31 484 1 015 31 481 95 650 Annual 14 886 34 584 6 140 49 350 763 950 Standard Deviations of Results Janu
46. O Screen Variables Oo Calibrate Model 5 Summary Statistics QO Frequency Analyses 5 Scenario Generator oO Compare Results 5 Time Series Analysis a X Reset Synthesize Settings Input File Data Select Parameter File View Details i File Select Predictor Directory Ae Valo x Ses No pre ors 5 EJSDSM Blogsville oregressio False Unconditional 4 1 01011976 sis Length Figure 7 1 The Weather Generator screen 7 1 File handling The first step in the synthesis process is the selection of the appropriate model parameter file Click on the Select Parameter File button in the top left hand corner An Open file window appears browse through until the correct directory and file are reached then click on the appropriate file name for example the parameters for maximum daily temperature at Blogsville are held in TMAX61 75 PAR The name of the file will then appear beneath the button Next specify the location of the predictor variable files by choosing the correct drive and directory from the window in the bottom left hand corner of the screen under Select Predictor Directory To write synthetic data to a results file it is necessary to select an appropriate directory and output file name Click on the Save To OUT File button in the top right hand corner An Open file window appears browse through until the correct directory is reached then enter a suitable file name for
47. SDSM 4 2 A decision support tool for the assessment of regional climate change impacts User Manual Robert L Wilby and Christian W Dawson August 2007 Department of Geography Lancaster University UK Science Department Environment Agency of England and Wales UK Department of Computer Science Loughborough University UK Preface General Circulation Models GCMs suggest that rising concentrations of greenhouse gases will have significant implications for climate at global and regional scales Less certain is the extent to which meteorological processes at individual sites will be affected So called downscaling techniques are used to bridge the spatial and temporal resolution gaps between what climate modellers are currently able to provide and what impact assessors require This manual describes a decision support tool for assessing local climate change impacts using a robust statistical downscaling technique SDSM 4 2 Statistical DownScaling Model facilitates the rapid development of multiple low cost single site scenarios of daily surface weather variables under present and future climate forcing Additionally the software performs ancillary tasks of data quality control and transformation predictor variable pre screening automatic model calibration basic diagnostic testing statistical analyses and graphing of climate data This manual also describes the UKSDSM archive a set of daily predictor var
48. ability and to the unanticipated effects that changes to precipitation occurrence may have on secondary variables such as temperature 1 1 4 Transfer functions Transfer function downscaling methods rely on empirical relationships between local scale predictands and regional scale predictor s Individual downscaling schemes differ according to the choice of mathematical transfer function predictor variables or statistical fitting procedure To date linear and non linear regression artificial neural networks canonical correlation and principal components analyses have all been used to derive predictor predictand relationships The main strength of transfer function downscaling is the relative ease of application coupled with their use of observable trans scale relationships The main weakness is that the models often explain only a fraction of the observed climate variability especially in precipitation series In common with weather typing methods transfer methods also assume validity of the model parameters under future climate conditions and the downscaling is highly sensitive to the choice of predictor variables and statistical form see below Furthermore downscaling future extreme events using regression methods is problematic since these phenomena by definition tend to lie at the limits or beyond the range of the calibration data set 1 2 Relative skill of statistical and dynamical downscaling The wide range of downscaling t
49. al and Modelling Software 17 145 157 Wilby R L Tomlinson O J and Dawson C W 2003 Multi site simulation of precipitation by conditional resampling Climate Research 23 183 194 Example applications of SDSM Abraham L Z 2006 Climate change impact on Lake Ziway watershed water availability Ethiopia Unpublished MSc Thesis University of Applied Sciences Cologne pp123 Aspen Global Change Institute AGCI 2006 Climate Change and Aspen An Assessment of Impacts and Potential Responses Appendix B p107 111 Aspen Global Change Institute Colorado pp147 Bootsma A Gameda S and McKenney D W 2005 Impacts of potential climate change on selected agroclimatic indices in Atlantic Canada Canadian Journal of Soil Science 85 329 343 Crawford T Betts N L and Favis Mortlock D T 2007 Issues of GCM grid box choice and predictor selection associated with statistical downscaling of daily precipitation over Northern Ireland Climate Research under review Diaz Nieto J and Wilby R L 2005 A comparison of statistical downscaling and climate change factor methods impacts on low flows in the River Thames United Kingdom Climatic Change 69 245 268 Fealy R 2006 An assessment of the relationship between glacier mass balance and synoptic climate in Norway Likely future implications of climate change Unpublished PhD Thesis University of Maynooth Ireland Goodess C M Anagnostopoulo C Bardossy A Frei
50. aling model parameter file Click on the Select Parameter File button in the top left hand corner An Open file window appears browse through until the correct directory and file are reached then click on the appropriate file name for example the parameters for maximum daily temperature at Blogsville are held in TMAX61 75 PAR The name of the file will then appear beneath the button Next click on the View Details button and the predictand followed by predictor files used in model calibration are listed in the window below In addition information on the number of predictors autoregression process type and SDSM version are also presented Next select the appropriate drive location and directory for the GCM predictors under the GCM Directory header For best practice GCM predictors originating from different experiments or time slices e g 1961 1990 or 2070 2099 should be held in separate folders This is because SDSM will load only files with the same predictor names i e characters 5 to 8 as those used in model calibration see Table 2 1 Note that in order to proceed with the Blogsville example two lagged files of one day will need to be created using the Transform Data screen in the Blogsville gcmx1961 90 and the Blogsville gcmx2070 99 directories gcmxp__vxx dat to produce gcmxvlagxx dat and gcmxp_zxx dat to produce gcmzlagxx dat As in the Weather Generator Section 7 decide how many ensembles members are needed
51. ality control Data transformation 5 SCREENING OF DOWNSCALING PREDICTOR VARIABLES 5 1 5 2 5 3 5 4 Setup Temporal variations in predictor strength Correlation matrix Scatterplot 6 MODEL CALIBRATION Page mere O O O o o0 12 13 14 14 14 14 15 15 15 15 16 17 19 20 21 22 24 24 25 27 27 28 29 32 6 1 File handling 6 2 Model type 6 3 Blogsville example 6 4 The PAR file 7 WEATHER GENERATOR 7 1 File handling 7 2 Ensemble size 7 3 Blogsville example 8 ANALYSIS OF OBSERVED AND DOWNSCALED DATA 8 1 Overview 8 2 Delta Statistics 8 3 The Statistics 9 FREQUENCY ANALYSIS 9 1 Setup 9 2 Diagnostics and plots 9 3 Extreme value analysis 10 SCENARIO GENERATION 10 1 Check settings 10 2 Setup 10 3 Blogsville example temperature 10 4 Blogsville example precipitation 11 GRAPHING MONTHLY STATISTICS 11 1 Line chart 11 2 Bar chart 11 3 Customizing charts 12 TIME SERIES ANALYSIS 12 1 Time series chart 12 2 Adjusting chart appearance 13 FINAL CAUTIONARY REMARKS BIBLIOGRAPHY APPENDIX 1 ENHANCEMENTS SINCE SDSM VERSION 3 1 APPENDIX 2 FREQUENTLY ASKED QUESTIONS GLOSSARY 32 33 34 35 37 37 38 39 41 41 44 45 47 47 48 50 55 55 56 57 60 65 65 66 67 69 69 12 75 76 84 86 90 0 TECHNICAL INFORMATION SDSM version 4 2 runs on PC based systems and has been tested on Windows 98 NT 2000 XP Note on older machines some statistical ana
52. and Wilby R L 2007 A daily weather generator for use in climate change studies Environmental Modelling and Software in press Palutikof J P Goodess C M Watkins S J and Holt T 2002 Generating rainfall and temperature scenarios at multiple sites Examples from the Mediterranean Journal of Climate 15 3529 3548 Qian B Hayhoe H and Gameda S 2005 Evaluation of the stochastic weather generators LARS WG and AAFC WG for climate change impact studies Climate Research 29 3 21 Richardson C W 1981 Stochastic simulation of daily precipitation temperature and solar radiation Water Resources Research 17 182 190 Semenov M A and Barrow E M 1997 Use of a stochastic weather generator in the development of climate change scenarios Climatic Change 35 397 414 Wilks D S 1992 Adapting stochastic weather generation algorithms for climate change studies Climatic Change 22 67 84 Wilks D S 1999 Multisite downscaling of daily precipitation with a stochastic weather generator Climate Research 11 125 136 Wilks D S and Wilby R L 1999 The weather generation game a review of stochastic weather models Progress in Physical Geography 23 329 357 Downscaling comparisons Charles S P Bates B C Whetton P H and Hughes J P 1999 Validation of downscaling models for changed climate conditions case study of southwestern Australia Climate Research 12 1 14 Christensen J H 2007 Evaluating the
53. application of SDSM particularly from the wider climate change impacts community The authors would also appreciate copies of any publications or reports arising from the use of SDSM This helps share experience with other Users and adds to the knowledge base of projected climate changes in different regions BIBLIOGRAPHY This Bibliography cites papers containing full technical details of SDSM followed by example case studies Additional overview material is also recommended for other downscaling methods as well as selected review papers in which various downscaling methods have been compared Note that an JPCC TGCIA Guidance Document on Statistical Downscaling is available via the Data Distribution Centre Technical basis of SDSM Conway D Wilby R L and Jones P D 1996 Precipitation and air flow indices over the British Isles Climate Research 7 169 183 Hassan H Aramaki T Hanaki K Matsuo T and Wilby R L 1998 Lake stratification and temperature profiles simulated using downscaled GCM output Journal of Water Science and Technology 38 217 226 Narula S C and Wellington J F 1977 An algorithm for linear regression with minimum sum of absolute errors Applied Statistics 26 106 111 Rubinstein R Y 1981 Simulation and the Monte Carlo method Wiley New York Wilby R L Dawson C W and Barrow E M 2001 SDSM a decision support tool for the assessment of regional climate change impacts Environment
54. ary 0 091 1 350 0 946 0 528 0 000 February 0 057 1 032 0 923 0 443 0 000 March 0 086 1 011 1 235 0 820 0 458 April 0 096 0 792 0 847 0 406 1 043 May 0 067 0 946 1 063 0 383 3 565 June 0 051 1 320 0 755 0 634 8 089 July 0 075 1 017 1 036 0 487 10 749 August 0 069 1 173 1 115 0 411 8 303 September 0 062 0 922 1 357 0 457 6 152 October 0 059 0 992 1 051 0 585 0 943 November 0 080 1 087 0 654 0 448 0 000 December 0 063 0 983 1 557 0 576 0 000 Winter 0 039 1 120 1 261 0 316 0 000 Spring 0 057 0 946 1 235 0 456 3 732 Summer 0 039 1 152 0 732 0 269 16 394 Autumn 0 037 0 922 0 654 0 528 6 436 Annual 0 020 1 152 1 205 0 310 16 651 Figure 10 4 Example results for the Blogsville using GCM predictors 2070 2099 Using the Compare Results operation see Section 11 it is possible to compare the frequency of hot days at Blogsville downscaled using observed NCEP and GCM HadCM3 predictor variables For example Figure 10 5 shows the respective monthly mean frequencies produced by each set of predictors with an ensemble size of 20 This was achieved by comparing peaks over threshold POT statistics for TMAXNCEP61 90 TXT and TMAXCCF61 90 TXT File Edit Help Back Reset Settings Copy Frequency of hot days NCEP HadCM3 Q E a gt a a Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Figure 10 5 Monthly frequency of hot days gt 25 C at Blogsville for the presen
55. ase the ensemble mean of the maximum temperature TMAXNCEP61 90 downscaled from NCEP has been plotted against the observed maximum temperature for Blogsville for the period 1961 1970 Time Series File Edit Help 0 8 amp 4 9 Back Reset Settings Copy Print Help Maximum daily temperature at Blogsville 61 o a o 2 a a a a 5 Date 1961 70 Figure 9 5 Time series plots of maximum daily temperature at Blogsville observed data blue line and ensemble mean downscaled from NCEP red line 9 3 Extreme value analysis The remaining four statistical measures allow the User to fit distributions to observed and downscaled data as either a whole data set or by isolating particular seasons or months in order to interpret extreme events The available distributions are Generalized Extreme Value GEV stretched exponential empirical and Gumbel Results can be viewed in either tabular format by selecting FA Tabular or as line charts by selecting FA Graphical from the menu buttons at the top of the screen Empirical This option fits a simple empirical distribution to the data by sorting the annual maximums into ascending order and plotting these according to the return period Figure 9 6 provides an example of an Empirical line plot while Figure 9 7 shows the same results presented in a tabular format FA Tabular In this case the maximum temperatures downscaled from NCEP TMAXNCEP61
56. ations on request All UK data have been re gridded to a standard co ordinate system 2 5 latitude x 3 75 longitude and normalised with respect to the 1961 1990 climatology The User must simply select the nearest grid box es to the site in question For all other regions including the UK gridded predictor variables are available online courtesy of the Canadian Climate Impacts Scenarios Group The web site is accessed from http www cics uvic ca scenarios index cgi Scenarios Q Can I use observational data that lie outside the standard period 1961 to 2000 No Observed predictor variables for SDSM archives are obtained from NCEP and normalised only for the period 1961 to 2000 Station meteorological data prior to Ist January 1961 or after 31st December 2000 will have no pre prepared predictor variables The software also assumes that meteorological data provided by the User commences on 1st January 1961 1 e has the same start date as the predictors If this is not the case the User should pad the station data with the Missing Data Identifier Q How important is the selection of predictor variables Identifying sensible predictor predictand relationships is the most critical procedure in all statistical downscaling methods The Screen Variables screen is designed to assist the User in the choice of appropriate downscaling predictor variables for model calibration via seasonal correlation analysis partial correlation analysi
57. axim Legend 2 title TMaxNCEP61 90 TXT Maxi Make Changes Apply Ticks Clear Ticks Show Legend Clear Legend Figure 11 4 An illustration of the Chart Settings screen Enter the required values text then click on the Make Changes button to change text and or click on the Apply Ticks button to change tick marks Similarly click on the Clear Ticks Show Legend or Clear Legend buttons as required Then click on the Back button to return to the plot To change the colour scheme of the lines or bars double click on the object required A Colour palette will appear Select the desired colour then OK to return to the chart Similarly to change the Font double click on title and or y axis title To change the position of the title single click then drag to the required location on the chart By applying the following design preferences in Figure 11 5 it is possible to customize the bar chart in Figure 11 3 to that shown in Figure 11 6 Chart Settings File Help Q 8 Close Reset Help Enter number of tick points 7 Chart title Maximum temperature at Blogsville Enter new Y axis maximum 40 Y axis label Temperature deg C Enter new Y axis minimum 0 Legend 1 title Observed Legend 2 title Downscaled Make Changes Apply Ticks Clear Ticks Show Legend Clear Legend Figure 11 5 Design preferences entered into Chart Settings screen ra a i 3i 25 HB Observed f Hi Downscaled a
58. d SIM files to ensure that data are consistently handled during subsequent scenario and data analysis routines Variance Inflation Controls the magnitude of variance inflation in downscaled daily weather variables This parameter changes the variance by adding reducing the amount of white noise applied to model estimates of the local process The default value produces approximately normal variance inflation prior to any transformation Larger values increase the variance of downscaled properties Variance inflation is de activated by setting the parameter to zero Note that for Fourth root and Natural log Model Transformation see above this parameter also affects changes in the mean estimate Bias Correction Compensates for any tendency to over or under estimate the mean of conditional processes by the downscaling model e g mean daily rainfall totals The default value is 1 0 indicating no bias correction Conditional Selection Adjusts the way in which conditional processes e g rainfall amounts are sampled The default Stochastic allows the outcome to be entirely based on chance Fixed Threshold allows the User to increase the chance of a conditional event by setting the threshold closer to zero or reducing the chance by setting closer to 1 0 Optimisation Algorithm SDSM 4 2 provides two means of optimising the model Dual Simplex as in earlier versions of SDSM and Ordinary Least Squares Although both ap
59. d as a percentage Resolution The grid separation of a climate model determining the smallest physical feature that can be realistically simulated Scenario A plausible and often simplified description of how the future may develop based on a coherent and internally consistent set of assumptions about driving forces and key relationships Scenarios may be derived from projections but are often based on additional information from other sources sometimes combined with a narrative story line Specific humidity The ratio of the mass of water vapour in grams to the mass of moist air in kilograms in a given volume of air Station The individual site at which meteorological measurements are systematically observed and recorded Stochastic A process or model that returns different outcomes from repeat experiments even when presented with the same initial and boundary conditions in contrast to deterministic processes See weather generator Transfer function A mathematical equation that relates a predictor or set of predictor variables to a target variable the predictand The predictor s and predictand represent processes operating at different temporal and or spatial scales In this case the transfer function provides a means of downscaling information from coarse to finer resolutions Tropopause The boundary between the lowest part of the atmosphere known as the troposphere and the highly stratified region of the atmosphere
60. directory is located enter the Filename for transformed data in this case ncepzlagee dat i e vorticity on previous days then click on Save WARNIG The name used for transformed files MUST comply fully with the protocol described in Section 2 4 1 To activate the procedure click on the Transform button at the top of the screen The following confirmation will appear Transformation complete s 10957 rows processed i No missing values in file Figure 4 4 The Transformation complete dialogue box Click on the OK button to return to the Transform Data File screen Click on the Reset button to clear the screen entries or to perform a new transformation 5 SCREENING DOWNSCALING PREDICTOR VARIABLES Identifying empirical relationships between gridded predictors such as mean sea level pressure and single site predictands such as station precipitation is central to all statistical downscaling methods and is often the most time consuming step in the process The purpose of the Screen Variables option is to assist the User in the choice of appropriate downscaling predictor variables for model calibration Section 6 SDSM performs three supporting tasks seasonal correlation analysis partial correlation analysis and scatterplots Ultimately however the User must decide whether or not the identified relationships are physically sensible for the site s and predictands in question In this matter there is no substitute for l
61. distinctive meteorological properties e g chance of rainfall sunshine hours wind direction air quality etc Examples of subjective circulation typing schemes include the European Grosswetterlagen and the British Isles Lamb Weather Types Zonal flow An atmospheric circulation in which the dominant flow of air follows the lines of latitude e g the westerlies in contrast to meridional flow
62. ditional 4FALSE variable 7 the number of ensemble members 8 the variance inflation parameter see Advanced Settings 9 the transformation code for conditional variables 1 none 2 fourth root 3 natural log 4 inverse normal 10 the bias correction parameter see Advanced Settings 11 the predictand file name 12 onward the predictor file name s 8 ANALYSIS OF OBSERVED AND DOWNSCALED DATA 8 1 Overview Statistical analyses of observed and downscaled weather data are handled in slightly different ways by SDSM but both are performed in the Summary Statistics screen Common diagnostic tests are available for both observed and synthetic data These statistics include the variable mean maximum minimum variance peaks above below thresholds percentiles percent wet days and wet dry day spell lengths computed on a calendar month seasonal or annual basis To evaluate either downscaled data or observed data click on the Summary Statistics button at the top of any main screen The following screen will appear Summary Statistics File Edit Help Home Quality Control E Transform Data Screen Variables Calibrate Model Weather Generator Summary statistics Frequency Analyses E Scenario Generator Compare Results Time Series Analysis z Reset Statistics Analyse Settings Select Input File Select Output File Select File Save Summary File As File Not selected
63. e appropriate option in the Process box Note that for conditional processes in which the distribution of predictand values is skewed it is possible to apply one of several transformations in Advanced Settings see Section 3 2 For example the Fourth root might be selected for daily precipitation amounts If an autoregressive component is required in the model i c a lagged predictand is used as a predictor the User should select Include within the Autoregression box SDSM 4 2 can calculate residual statistics and display these on either a scatter diagram or in a histogram The scatter diagram plots the residuals against the modelled predictor while the histogram shows the distribution of the residuals These two charts are generated after the summary statistics of the modelling are presented to the User The number of bars in the histogram can be adjusted by altering the value in the Histogram Categories box The User can also view the Chow test statistics for model stationarity by checking the appropriate box The Chow test is an optional test as it can slow down the modelling process significantly particularly if a Dual Simplex optimisation is selected Finally click the Calibrate button at the top of the screen 6 3 Blogsville example For the Blogsville example five predictor files pu p_z p500 vlag and zlag might be selected to downscale daily maximum temperatures TMAX see Figure 5 3 There is clearly a seasonal cycle in the
64. e to maintain the same nomenclature of the file but with the GCM extension Q Why do I keep getting an error message when I use GCM data The most likely explanation is that the Year Length in Settings has not been set correctly with respect to the number of days in the GCM simulation For example HadCM2 and HadCM3 have year lengths of 360 days whereas CGCM1 has 365 days in every year i e no leap years Version 4 2 prompts the User to double check the number of days before proceeding Q What s the best way of handling SDSM files outside the software All SDSM output files are written in ASCII format and therefore accessible by any word processor Model results OUT files are tab delimited if the number of Ensemble Members is greater than one and as such can be imported into commercial spreadsheets for further analysis or graphing Q I ve looked at the predictor variable files and the values only range between 5 Is there something wrong with the data No All predictor variables NCEP and GCM are normalised using their respective 1961 1990 means and standared deviations The result is that each predictor variable is dimensionless and will typically vary between 5 and 5 GLOSSARY Where appropriate the following definitions were drawn from the Glossary of terms in the Summary for Policymakers A Report of Working Group I of the Intergovernmental Panel on Climate Change and the Technical Summary of the Working
65. echniques both dynamical and statistical has prompted a growing number of model comparisons using generic data sets and diagnostics Until recently these studies were restricted to statistical versus statistical or dynamical versus dynamical model comparisons However some studies are now undertaking statistical versus dynamical model comparisons and Table 1 1 summarises the relative strengths and weaknesses that have emerged Statistical downscaling Dynamical downscaling boundary forcing e Choice of domain size and location affects results e Requires high quality data for model calibration e Predictor predictand relationships are often non stationary e Choice of predictor variables affects results e Choice of empirical transfer scheme affects results e Low frequency climate variability problematic e Always applied off line therefore results do not feedback into the host GCM Strengths Station scale climate information from 10 50 km resolution climate GCM scale output information from GCM scale output e Cheap computationally undemanding Respond in physically consistent ways and readily transferable to different external forcings Ensembles of climate scenarios permit e Resolve atmospheric processes such as risk uncertainty analyses orographic precipitation e Applicable to exotic predictands such Consistency with GCM as air quality and wave heights Weakness
66. ence Durham UK Whitehead P G Wilby R L Butterfield D and Wade A J 2006 Impacts of climate change on nitrogen in a lowland chalk stream An appraisal of adaptation strategies Science of the Total Environment 365 260 273 Wilby R L 2003 Past and projected trends in London s urban heat island Weather 58 251 260 Wilby R L 2005 Constructing wet season precipitation scenarios for a site in the Anti Atlas Mountains Morocco Proceedings of the Conference on Optimising Land and Water Resources in Arid Environments Agadir Morocco Wilby R L 2007 Constructing climate change scenarios of urban heat island intensity and air quality Environment and Planning B Planning and Design under review Wilby R L and Dettinger M D 2000 Streamflow changes in the Sierra Nevada CA simulated using a statistically downscaled General Circulation Model scenario of climate change In McLaren S J and Kniveton D R Eds Linking Climate Change to Land Surface Change Kluwer Academic Publishers Netherlands pp 99 121 Wilby R L Hassan H and Hanaki K 1998b Statistical downscaling of hydrometeorological variables using general circulation model output Journal of Hydrology 205 1 19 Wilby R L Whitehead P G Wade A J Butterfield D Davis R and Watts G 2006 Integrated modelling of climate change impacts on the water resources and quality in a lowland catchment River Kennet UK Journal of Hydrology 3
67. er environmental variables e g air quality parameters wave heights snow cover etc In any event these data must be supplied by the User in SDSM format see also Section 2 4 2 This is single column text only data beginning Ist January 1961 if necessary padded with the Missing Data Identifier Assembly of the candidate predictor suite can be a far more involved process entailing data extraction re gridding and normalisation techniques For this reason SDSM is supplied with a prepared set of daily predictor variables for selected grid boxes covering the British Isles Figure 2 2 and globally for all land areas via the web Section 2 5 The User simply locates the required grid box and data source from the UKSDSM or online archive As Figure 2 2 shows the UK is represented by nine grid boxes each measuring 2 5 latitude by 3 75 longitude corresponding to the grid co ordinate system of the Hadley Centre s coupled ocean atmosphere GCMs see below Of the nine cells six are land and three are ocean To obtain more realistic estimates of forcing over land areas that are represented by ocean grid boxes in the GCM data from the two nearest land cells were averaged For example predictor variables for Southwest England SW are the average of data from the Wales WA and Southern England SE grid boxes ie ae f SC Scotland NI Northern Ireland F SB Scottish Boarders iai E NE Northeast England Nt NE IR Ireland
68. es of statistical downscaling over dynamical downscaling Table 1 1 a hybrid conditional weather generator method was chosen as the basis of the decision support tool SDSM 1 3 Manual outline The rest of this manual is organised in seven main parts Section 2 provides a brief overview of the key operations of SDSM For a complete description of the model specification interested readers should refer to the articles listed in the Bibliography see below Descriptions of the UKSDSM and Canadian Climate Impacts Scenarios CCIS data archives and file nomenclature are also provided in Section 2 Sections 3 to 12 provide guidance on the practical implementation of the key functions in SDSM for downscaling regional climate change scenarios Application of SDSM is illustrated using a hypothetical case study for Blogsville Section 13 provides a few cautionary remarks concerning the limitations of SDSM and appropriate usage Users are strongly recommended to consider the issues raised here before developing local scenarios using SDSM Next a comprehensive Bibliography is supplied This provides a general overview of downscaling as well as more detailed discussions of the technical basis of SDSM example applications and comparisons with other downscaling methods Enhancements to SDSM since version 3 1 are listed in Appendix 1 A trouble shooting guide and outline of the most common pitfalls is provided in the form of a Frequently Asked Questi
69. eteorological Service of Canada Atlantic Region Science Report series 2005 9 Dartmouth Canada pp41 London Climate Change Partnership 2002 A climate change impacts in London evaluation study Final Technical Report Entec UK Ltd MacDonald O 2004 Coupling glacier mass balance and meltwater yields in the European Alps with future climate change downscaling from integrations of the HadCM model Unpublished PhD Thesis University of Salford UK Reynard N Crooks S Wilby R L and Kay A 2004 Climate change and flood frequency in the UK Proceedings of the 39th Defra Flood and Coastal Management Conference University of York UK Scibek J and Allen D M 2006 Modeled impacts of predicted climate change on recharge and groundwater levels Water Resources Research 42 W11405 Wetterhall F B rdossy A Chen D Halldin S and Xu C 2007 Daily precipitation downscaling techniques in three Chinese regions Water Resources Research 42 W11423 doi 10 1029 2005WR004573 Wetterhall F Halldin S and Xu C Y 2007 Seasonality properties of four statistical downscaling methods in central Sweden Theoretical and Applied Climatology 87 123 137 Whitehead P G Futter M and Wilby R L 2006 Impacts of climate change on hydrology nitrogen and carbon in upland and lowland streams Assessment of adaptation strategies to meet Water Framework Directive objectives Proceedings of the British Hydrological Society Confer
70. f months available In this case it may be better to remove the labels entirely to avoid overcrowding on the X axis The User implements chart settings by clicking the Make Changes button X axis labels are removed by clicking the Clear X Labels button Labels The User can adjust the text appearing on the X and Y axis and also the chart title by typing in the appropriate text on this screen and clicking the Make Changes button The User can also make adjustments to the chart directly For example by double clicking on the lines the User can adjust their colour Double clicking on the title and axis labels allows the User to change the text font The axis labels chart title or legend are removed by clicking on them and hitting delete or backspace To incorporate the Chart in a Word document first use the Copy button at the top of the window then in Word use Paste Special Picture Figure 12 4 below provides an example of a time series plot generated using Largest n day total statistic 3 Time Series Plot File Edit Help Largest 5 day total for PRCP DAT 197611977 11978 11979 11980 1981 lis8211983 1198411985 11986 11987 11988 1989 11990 Year Figure 12 4 An example of an annual time series plot of the largest 5 day total rainfalls at Blogsville 1976 1990 13 FINAL CAUTIONARY REMARKS SDSM is a Windows based decision support tool for the rapid development of single site ensemble scenarios of daily weather variables unde
71. f summary statistics The default is 999 Random Number Seed Ensures that the random sequences produced by Weather Generator Section 7 and Scenario Generator Section 10 are different each time the model is run If replicate experiments are preferred the check box should be deselected Default File Directory Allows the user to select a default directory that is accessed by all screens when first searching for files Settings Eile Help ol ae a x 9 Back Reset Load Save Advanced Help Default File Directory Documents and Settings E Christian Dawson EJ Desktop Sm TestData J Blogsville C NewTestData Figure 3 3 The Settings screen 3 2 Advanced settings The advanced settings are accessed from the Settings screen by clicking on the Advanced button at the top of the screen The Advanced Settings screen allows the User to change and save further downscaling model preferences Figure 3 4 Model Transformation Specifies the transformation applied to the predictand in conditional models The default None is used whenever the predictand is normally distributed as is often the case for daily temperature The alternatives Fourth root Natural log and Inverse Normal are used whenever data are skewed as in the case of daily precipitation Note that the Inverse Normal transformation employs conditional resampling of the observed predictand see Wilby et al 2003 The transformation type is recorded in PAR an
72. he ensemble mean and are accessed by bar line chart options The data format also enables convenient export to other graphing software and spreadsheets Table 2 2 SDSM file names and recommended directory structure Extension Explanation Directory DAT Observed daily predictor and predictand files employed by SDSM Scenarios the Calibrate and Weather Generator operations input Calibration PAR Meta data and model parameter file produced by the SDSM Scenarios Calibrate operation output and used by the Weather Calibration Generator and Generate Scenario operations input SIM Meta data produced by the Weather Generator and SDSM Scenarios Generate Scenario operations output Results OUT Daily predictand variable file produced by the Weather SDSM Scenarios Generator and Generate Scenario operations output Results TXT Information produced by the Summary Statistics and SDSM Scenarios Frequency Analysis operations output Results 2 5 Obtaining SDSM predictors online SDSM predictors may be obtained for any global land area courtesy of a data portal maintained by the Canadian Climate Impacts Scenarios Group The web site is accessed from http www cics uvic ca scenarios index cgi Scenarios Having registered by e mail address the User then selects predictors from the available GCMs currently HadCM3 and CGCM2 given the latitude and longitude of the nearest grid box es to the
73. hold box is ticked only data which are above the global threshold see Settings are included in the analysis If the User wishes to plot a Probability Density Function PDF as part of the analysis the number of categories can be entered in the PDF Categories box the default is 20 9 2 Diagnostics and plots Following the initial set up process the User can perform a number of diagnostics on the data These diagnostics are discussed in turn Quantile Quantile Q Q Plot A Quantile Quantile plot is used to compare a modelled data set with an observed data file The procedure works by sorting each of the data files into order and calculating the percentiles 1 to 99 These are then plotted against one another on a scatter chart with observed data on the y axis and modelled data on the x axis Note that it is assumed that observed data are always based on calendar years while the length of the modelled data year is set using the Settings screen Figure 9 2 provides an example of a Quantile Quantile plot In this case TMAX DAT has been selected as the observed data TMAXCCF61 90 OUT as the modelled data maximum daily temperature at Blogsville downscaled using HadCM3 output for the period 1961 1990 In this case all the data have been analysed and the ensemble mean has been chosen to represent the modelled data Scatter Plot Eile Edit Help 8 9 Back Reset Settings Copy Print Help Quantile Quantile Plot x
74. iables prepared for model calibration and downscaling at sites across the UK The archive contains variables describing atmospheric circulation thickness stability and moisture content at several levels in the atmosphere under climate conditions observed between 1961 and 1990 Equivalent predictor variables are provided for four GCM experiments of transient climate change between 1961 and 2099 Users seeking to apply SDSM to regions outside the UK may obtain predictor variables online by visiting http Awww cics uvic ca scenarios index cgi Scenarios Application of SDSM is illustrated with respect to the downscaling of daily maximum temperature and precipitation scenarios fat a hypothetical location Blogsville under present 1961 90 and future 2070 99 climate forcing Acknowledgements SDSM Version 4 2 was supported by the Environment Agency of England and Wales as part of the Thames Estuary 2100 project SDSM Version 3 1 was supported by the Environment Agency of England and Wales as part of the Climate Impacts and Adaptation Research Programme The UKSDSM data archive was updated by Ian Harris Climate Research Unit and now includes data kindly supplied by the UK Hadley Centre CSIRO Atmospheric Research and the Canadian Centre for Climate Modelling and Analysis SDSM Version 2 2 was sponsored by the Environment Agency through the National Centre for Risk Analysis and Options Appraisal SDSM Version 2 1 was supported by the Canadian
75. ian dry spell Median wet spell SD dry spell SD wet spell Spell length correlation Dry day persistence Wet day persistence Maximum dry spell Maximum wet spell Nth largest value Largest n day total Percentage of precipitation above annual percentile Percentage of all precipitation from events greater than long term percentile Number of events greater than long term percentile the User can enter their own thresholds and percentile values Miscellaneous improvements e Default file directory established in settings to ensure that every screen searches in the same directory for files each time e Improved interface so that it is now easier to move between stages of the process with bigger screens and improved colour schemes Soft reset when error occurs so that User settings are not reset if a problem occurs Splash screen changed can now be removed Advanced Settings enables fixed or stochastic threshold for conditional processes Error trapping and efficiency improved throughout Help files and User manual updated accordingly APPENDIX 2 FREQUENTLY ASKED QUESTIONS The following generic and specific questions are arranged in the order in which they might typically be encountered during a downscaling procedure Q Do I need to perform any re gridding or normalisation of the predictor variables No These tasks have already been performed for the UKSDSM data set released with the software and available to non profit organis
76. ible combinations of predictors Analyses models using either AIC or BIC criteria which user can select in advanced settings Optimisation Algorithm e In addition to the dual simplex algorithm of SDSM 3 1 an ordinary least squares algorithm has been implemented This is much quicker and efficient It can be selected in Advanced Settings Screen Variables e The User can now apply an autoregression component alongside other predictors Calibrate Model e An autoregressive term can now be included in the model e Residual analysis has been added so that following calibration SDSM allows the user to plot residuals of the model either as a scatter diagram or a histogram both of which can be amended through additional settings e The Chow test has been added so the user can also now assess the calibrated model for stationarity Weather Generator e Additional information is captured within the PAR file i e SDSM version auto regression and process Scenario Generator e Additional information is provided on the model before generation begins Summary Statistics replaces Analyse Data screen e A raft of new statistics have been added Extreme Range Minimum Range Maximum N day Total Mean Wet Day Persistence Mean Dry Day Persistence Correlation for Spell Lengths Median Wet Spell Length Median Dry Spell Length Time Series Analysis e Includes a raft of additional STARDEX indices for analysis Mean dry spell Mean wet spell Med
77. ibliography are widely used in the hydrological and agricultural research communities but do not directly employ GCM output in the scenario construction processes Following a brief overview of downscaling techniques we describe the structure and operation of SDSM with respect to seven tasks 1 quality control and data transformation 2 screening of potential downscaling predictor variables 3 model calibration 4 generation of ensembles of present weather data using observed predictor variables 5 statistical analysis of observed data and climate change scenarios 6 graphing model output 7 generation of ensembles of future weather data using GCM derived predictor variables The key functions of SDSM will be illustrated using observed and climate model data for a hypothetical station Blogsville comparing downscaled daily precipitation and temperature series for 1961 1990 with 2070 2099 1 1 Downscaling techniques The general theory limitations and practice of downscaling have been discussed in detail elsewhere see bibliography Reviews typically group downscaling methodologies into four main types a dynamical climate modelling b synoptic weather typing c stochastic weather generation or d transfer function approaches Each family of techniques is briefly described below 1 1 1 Dynamical Dynamical downscaling involves the nesting of a higher resolution Regional Climate Model RCM within a coarser resolution GCM The R
78. ics Frequency Analyses Oo Scenario Generator 5 Compare Results o Time Series Analysis Reset Calibrate Settings Select Output PAR File Output File C gemx1960 89 C gemx2080 99 C ncep1961 90 C observed1961 90 Figure 6 1 The Calibrate Model screen 6 1 File handling To begin model building click on the Select Predictand File button in the top left hand corner An Open file window appears browse through until the correct directory and file are reached then click on the appropriate file name for example the maximum daily temperature at Blogsville TMAX DAT The name of the file will then appear beneath the button Follow a similar procedure to locate and select the desired predictor variables by choosing the correct drive from the pull down window in the centre of the screen The directories available on this drive will then appear in the window directly above the drive window For example locate the C SDSM Blogsville NCEP1961 90 directory All DAT files in this directory are then listed in the window above To select a predictor simply click on the file name it will be highlighted in blue To deselect a file click on it again and it will no longer be highlighted The number of predictor variables chosen is shown beneath this window The Data Period menu on the left hand side of the Calibrate Model screen allows the start and end dates of the analysis period to be changed The default
79. iles for the Scenario Generator need not be the same length as those used to obtain the regression weights during calibration To access this facility select Scenario Generator from any of the main screens The following screen appears Scenario Generator Eile Edit Help Home oO Quality Control oO Transform Data Oo Screen Variables oO Calibrate Modell o Weather Generator 5 Summary Statistics Frequency Analyses Oo Scenario Generator Oo Compare Results Oo Time Series Analysis 0 H X Reset Generate Settings Select Input File Data Select Output File View Details Save To OUT File GCM Directory re VAIO 9 gemx1 960 89 C gemx2080 99 C neep1 961 90 C observed 961 90 Figure 10 1 The Scenario Generator screen 10 1 Check settings Before starting scenario generation it may be necessary to change some of the options in the Settings menu Click on the Settings button at the top of the screen and check the appropriate Year Length box Also amend the Standard Start End Date in line with the GCM data time slices For example HadCM2 and HadCM3 have year lengths of 360 days and for the Blogsville example the period 1961 1990 was used to represent present climate forcing 10800 values Once necessary changes have been made to the Settings click on Back to return to the Scenario Generator screen 10 2 Setup The first step in scenario generation is the selection of the appropriate downsc
80. imates a normal distribution about the mean with small probabilities for very high or low temperatures Radiative forcing The change in net vertical irradiance expressed as Watts per square metre at the tropopause due to an internal change or a change in the external forcing of the climate system such as for example a change in the concentration of carbon dioxide or the output of the Sun Random See stochastic Re gridding A statistical technique used to project one co ordinate system onto another and typically involving the interpolation of climate variables A necessary pre requisite to most statistical downscaling because observed and climate model data are seldom archived using the same grid system Regional Climate Model RCM A three dimensional mathematical model that simulates regional scale climate features of 20 50 km resolution given time varying atmospheric properties modelled by a General Circulation Model The RCM domain is typically nested within the three dimensional grid used by a GCM to simulate large scale fields e g surface pressure wind temperature and vapour Regression A statistical technique for constructing empirical relationships between a dependent predictand and set of independent predictor variables See also black box transfer function Relative humidity A relative measure of the amount of moisture in the air to the amount needed to saturate the air at the same temperature expresse
81. in turn depends on regional scale predictors such as atmospheric humidity and pressure Deterministic A process physical law or model that returns the same predictable outcome from repeat experiments when presented with the same initial and boundary conditions in contrast to stochastic processes Domain A fixed region of the Earth s surface and over lying atmosphere represented by a Regional Climate Model Also denotes the grid box es used for statistical downscaling In both cases the downscaling is accomplished using pressure wind temperature or vapour information supplied by a host GCM Divergence If a constant volume of fluid has its horizontal dimensions increased it experiences divergence and by conservation of mass its vertical dimension must decrease Downscaling The development of climate data for a point or small area from regional climate information The regional climate data may originate either from a climate model or from observations Downscaling models may relate processes operating across different time and or space scales Dynamical See Regional Climate Model Emission scenario A plausible representation of the future development of emissions of substances that are potentially radiatively active e g greenhouse gases aerosols based on a coherent and internally consistent set of assumptions about driving forces and their key relationships Ensemble member A set of simulations members in which a determinis
82. known as the stratosphere The tropopause is typically located 10km above the Earth s surface Uncertainty An expression of the degree to which a value e g the future state of the climate system is unknown Uncertainty can result from a lack of information or from disagreement about what is known or knowable It can also arise from poorly resolved climate model parameters or boundary conditions Unconditional process A mechanism involving direct physical or statistical link s between a set of predictors and the predictand For example local wind speeds may be a function of regional airflow strength and vorticity Vorticity Twice the angular velocity of a fluid particle about a local axis through the particle In other words a measure of rotation of an air mass Weather generator A model whose stochastic random behaviour statistically resembles daily weather data at single or multiple sites Unlike deterministic weather forecasting models weather generators are not expected to duplicate a particular weather sequence at a given time in either the past or the future Most weather generators assume a link between the precipitation process and secondary weather variables such as temperature solar radiation and humidity Weather pattern An objectively or subjectively classified distribution of surface and or upper atmosphere meteorological variables typically daily mean sea level pressure Each atmospheric circulation pattern should have
83. le of a Stretched Exponential plot using the same data as in Figure 9 6 Frequency Analysis Plot File Edit Help 0 8 amp 4 9 Back Reset Settings Copy Print Help Period Annual Fit Gumbel gt TMAXDAT TMAXNCEP61 30 0UT 40 50 60 Return period years Figure 9 9 Gumbel plot of maximum and downscaled temperature for Blogsville for the period 1961 1990 9 Frequency Analysis Plot File Edit Help e a aj o Reset Settings Back Copy Print Help Period Annual Fit Stretched Exponential gt TMAX DAT TMAXNCEP61 30 0UT 40 50 60 Return period years Figure 9 10 Stretched Exponential plot of maximum and downscaled temperature for Blogsville for the period 1961 1990 10 SCENARIO GENERATION The Scenario Generator operation produces ensembles of synthetic daily weather series given daily atmospheric predictor variables supplied by a GCM either under present or future greenhouse gas forcing The GCM predictor variables must be normalised with respect to a reference period or control run and available for all variables used in model calibration see Section 2 2 The procedure is identical to that of the Weather Generator operation in all respects except that it may be necessary to specify a different convention for model dates and source directory for predictor variables As in the case of the Weather Generator see Section 7 input f
84. lect an appropriate directory and file name Click on the Save Summary File As button on the right hand side An Open file window appears browse through until the correct directory is reached then enter a suitable file name for example TMAXOBS76 90 TXT The name of the file will then appear beneath the button The final step is to select the required diagnostics Click on the Statistics button at the top of the menu The following screen will appear Statistics Selection File Edit Help KA bed g E a E rv m rv E g E E E E E E nge E Delta Periods ait Ea E Period 6 Stat f eriod C Start EEC Baie 31 12 1990 ie ee aa 31 12 2040 CTE hse 31 12 2070 ie Rem ate 31 12 2099 Figure 8 2 The Statistics Selection screen The screen is divided into three areas The first headed Generic Tests lists statistics that may be applied to any variable mean maximum minimum sum variance median count peaks over below threshold percentile inter quartile range autocorrelation skewness and maximum N day total etc The second headed Conditional Tests lists statistics that are only applicable to daily conditional series percentage wet mean dry wet spell length maximum dry wet spell length standard deviation of dry wet spell peaks over threshold as a percentile peaks over threshold as a percentage of total rainfall etc Note that the definition of a wet day can be adjusted using the Event Threshold under Setting
85. lyses may take longer to perform and or may exhaust available memory when large data sets are processed 1 INTRODUCTION Even if global climate models in the future are run at high resolution there will remain the need to downscale the results from such models to individual sites or localities for impact studies DOE 1996 p34 General Circulation Models GCMs indicate that rising concentrations of greenhouse gases will have significant implications for climate at global and regional scales Unfortunately GCMs are restricted in their usefulness for local impact studies by their coarse spatial resolution typically of the order 50 000 km and inability to resolve important sub grid scale features such as clouds and topography As a consequence two sets of techniques have emerged as a means of deriving local scale surface weather from regional scale atmospheric predictor variables Figure 1 1 Firstly statistical downscaling is analogous to the model output statistics MOS and perfect prog approaches used for short range numerical weather prediction Secondly Regional Climate Models RCMs simulate sub GCM grid scale climate features dynamically using time varying atmospheric conditions supplied by a GCM bounding a specified domain Both approaches will continue to play a significant role in the assessment of potential climate change impacts arising from future increases in greenhouse gas concentrations Climate Model
86. mble members are considered equally plausible local climate scenarios realised by a common set of regional scale predictors The extent to which ensemble members differ depends on the relative significance of the deterministic and stochastic components of the regression models used for downscaling For example local temperatures are largely determined by regional forcing whereas precipitation series display more noise arising from local factors The former will yield similar looking ensemble members the latter larger differences between individual members Once the above selections have been completed click the Synthesize button at the top of the menu After a few seconds the follow dialogue box will appear e Synthesis completed Figure 7 2 The synthesis completed dialogue box Click on OK to return to the Weather Generator screen 7 3 Blogsville example Selections for Blogsville are illustrated in Figure 7 1 In this example the Weather Generator synthesized 20 runs of 15 years daily maximum temperature using five regional scale predictors The data were synthesized using independent predictors withheld from model calibration i e for the period 1976 1990 5479 values Figure 7 3 shows the first few values of 12 ensemble members held in TMAXNCEP76 90 0UT Figure 7 4 shows the corresponding TMAXNCEP76 90 SIM file which contains meta data associated with the synthesis In both cases the files have been opened in
87. nd Date 31 12 1990 Ensemble Member s ALL Month Mean Maximum Minimum Variance Sum Dry spell January 3 045 29 296 0 001 15 446 49 013 2 024 February 3 158 34 653 0 001 19 230 48 551 2 160 March 2 925 28 769 0 001 14 439 46 836 2 218 April 3 521 37 187 0 001 22 219 61 999 2 151 May 3 979 48 352 0 001 30 314 67 189 2 345 June 4 329 47 977 0 001 34 955 66 436 2 393 July 4 414 59 685 0 001 45 807 60 934 2 411 August 3 940 52 272 0 001 37 452 50 678 2 638 September 3 577 39 864 0 001 28 145 32 383 3 898 October 3 162 29 850 0 001 18 152 38 398 2 885 November 3 282 37 050 0 001 21 071 47 933 2 385 December 3 384 42 538 0 001 23 388 47 016 2 528 Winter 3 186 47 774 0 001 19 205 139 917 2 290 Spring 3 485 49 481 0 001 22 674 176 023 2 276 Summer 4 237 70 525 0 001 39 379 178 048 2 560 Autumn 3 314 44 072 0 001 21 917 118 714 3 146 Annual 3 554 74 419 0 001 25 835 617 365 2 598 Standard Deviations of Results January 0 246 6 897 0 000 3 147 4 050 0 068 February 0 237 9 065 0 000 4 885 3 908 0 089 March 0 221 4 844 0 001 2 342 3 618 0 087 April 0 218 9 504 0 000 5 166 3 799 0 091 May 0 217 13 578 0 000 6 207 3 717 0 139 June 0 272 9 371 0 000 5 698 4 250 0 096 July 0 289 21 614 0 000 10 241 3 521 0 123 August 0 368 22 797 0 000 14 034 5 202 0 108 September 0 300 11 797 0 001 9 426 2 784 0 175 October 0 250 4 568 0 001 3 684 3 046 0 150 November 0 207 9 421 0 001 4 176 3 543 0 136 December 0 205 17 239 0 000 6 435 2 797 0 115 Winter 0 109 14 608 0 000
88. nd file B TMAX61 75 PAR WordPad File Edit View Insert Format Help Del 644 gt amp 5 12 366 01 01 1961 10957 01 01 1961 5478 F ALSE 1 1 False TMAX DAT ncepp__uxx ncepp__zxx ncepp500xx ncepvlagxx ncepzlagxx 7 571 8 564 11 13 15 16 17 17 15 13 10 645 724 418 426 513 203 508 155 106 7 670 C SDSM Blogsville observed1961 90 TMAX DA 1 322 652 2 1 248 136 2 1 163 616 3 0 681 858 3 0 138 962 4 0 326 878 4 0 104 636 3 0 457 138 4 0 181 227 3 0 316 472 3 0 683 245 2 1 437 1 546 2 T For Help press F1 Figure 6 4 The PAR file produced by the Calibrate Model screen 7 WEATHER GENERATOR The Weather Generator operation produces ensembles of synthetic daily weather series given observed or NCEP re analysis atmospheric predictor variables and regression model weights produced by the Calibrate Model operation see Section 6 The Weather Generator enables the verification of calibrated models assuming the availability of independent data as well as the synthesis of artificial time series representative of present climate conditions The Weather Generator can also be used to reconstruct predictands or to infill missing data To access this facility click the Weather Generator button from any of the main screens The following screen will appear Weather Generator File Edit Help A Home 5 Quality Control Transform Data o
89. ngth SD wet spell length Peaks over threshold POT as of total Mean dry day persistence Mean wet day persistence Correlation for spell lengths Median dry spell length Median wet spell length Average length of spells with amounts less than the wet day threshold Average length of spells with amounts greater than or equal to the wet day threshold Longest spell with amounts less than the wet day threshold Longest spell with amounts greater than or equal to the wet day threshold Standard deviation of spells with amounts less than the wet day threshold Standard deviation of spells with amounts greater than or equal to the wet day threshold Count of peaks over User specified threshold defined as a percentile of all data Ratio of the sum of all values over the User specified threshold defined as a percentile of all values to the sum of all values Total number of consecutive dry days divided by total number of dry days Total number of consecutive wet days divided by total number of wet days Overall persistent of spells both wet and dry Median length of spells with amounts less than the wet day threshold Median length of spells with amounts greater than or equal to the wet day threshold 9 FREQUENCY ANALYSIS The Frequency Analysis option allows the User to plot various distribution diagnostics for both modelled ensemble members and observed data To access this facility select Frequency Analysis from a
90. ny of the main screens The following screen appears 3 Frequency Analyses File Edit Help A Home Quality Control Transform Data Screen Variables Calibrate Model Weather Generator 5 Summary Statistics Frequency Analyses 5 Scenario Generator 5 Compare Results 3 Time Series Analysis Reset Q Q Plot PDF Plot Line Plot FA Graphical FA Tabular Settings Modelled Data Select Modelled Data File Not selected File Not selected s Period date Gumbel anaidaie Stretched Exponential Threshold Save Results To All Mean Ensemble s File Not selected Figure 9 1 The Frequency Analysis screen 9 1 Setup The first stage in the process is to select an observed data file and or modelled data file to analyse By clicking on the appropriate selection button for example Select Observed Data an input dialogue window appears in which the desired files can be selected The second stage of the process is to enter the analysis period by entering the appropriate start and end dates in the Analysis Period box In the Data Period box the User can select the time period for the analysis all the data individual months or seasons If modelled ensemble data are being analysed the User can select which part of the ensemble to include in the analysis by selecting either All Members Ensemble Mean Ensemble Member or All Ensemble Member in the Ensemble box Ifthe Apply thres
91. o address this limitation This arises because of the generally low predictability of daily precipitation amounts at local scales by regional forcing factors The unexplained behaviour is currently modelled stochastically within SDSM by artificially inflating the variance of the downscaled series to accord better with daily observations Even so the model can produce unrealistic behaviour if the stochastic component is not properly handled This again underlines the importance of independent testing of all model parameters against data withheld from model calibration Ultimately however the plausibility of all SDSM scenarios depends on the realism of the climate model forcing Systematic biases in the mean and variance of GCM predictors can be reduced through normalisation with respect to a control period as in the case of all pre prepared SDSM predictors Biases in large scale patterns of atmospheric circulation in GCMs e g shifts in the dominant storm track relative to observed data or unrealistic inter variable relationships are much harder to accommodate Where possible Users should not therefore restrict themselves to the use of a single GCM or emission scenario for downscaling By applying multiple forcing scenarios via different GCMs ensemble members time slices or emission pathways better insight may be gained into the magnitude of these uncertainties Finally the authors welcome constructive suggestions about the design or
92. ocal knowledge To investigate potentially useful predictor predictand relationships click on the Screen Variables button from any of the main screens The following screen will appear Screen Variables File Edit Help D Home oO Quality Control 5 Transform Data 5 Screen Variables oO Calibrate Model S Weather Generator 5 Summary Statistics 3 Frequency Analysis 5 Scenario Generator 5 Compare Results Time Series Analysis E as Reset Analyse Correlation Scatter Settings Data Fit Start 01 01 1961 meee 31 12 1990 E Biogsvile Sel eriod Annual No of predic lected 0 Figure 5 1 Illustration of the Screen Variables screen using daily maximum temperatures for Blogsville 1961 1990 5 1 Setup The first step in the Screen Variables operation is the selection of the predictand and predictor files The predictand file e g observed daily maximum temperature daily precipitation totals etc must be supplied by the User in SDSM format see Section 2 4 Click on the Select Predictand File button An Open file window will appear browse through until the appropriate directory has been located Click on the predictand data file for example the maximum daily temperature at Blogsville TMAX DAT located in C SDSM Blogsville observed 1961 90 Follow a similar procedure locate and select the desired Predictor Variables by choosing the correct drive from the pull down window in the
93. on D Hulme M and Lal M 2003 Guidelines for use of climate scenarios developed from Regional Climate Model experiments IPCC Task Group on Scenarios for Climate Impact Assessment TGCIA Prudhomme C Reynard N and Crooks S 2002 Downscaling of global climate models for flood frequency analysis Where are we now Hydrological Processes 16 1137 1150 Wilby R L and Wigley T M L 1997 Downscaling general circulation model output a review of methods and limitations Progress in Physical Geography 21 530 548 Wilby R L and Wigley T M L 2000 Precipitation predictors for downscaling observed and General Circulation Model relationships International Journal of Climatology 20 641 661 Wilby R L Charles S Mearns L O Whetton P Zorito E and Timbal B 2004 Guidelines for Use of Climate Scenarios Developed from Statistical Downscaling Methods IPCC Task Group on Scenarios for Climate Impact Assessment TGCIA Wilks D S and Wilby R L 1999 The weather generation game a review of stochastic weather models Progress in Physical Geography 23 329 357 Xu C Y 1999 From GCMs to river flow a review of downscaling methods and hydrologic modelling approaches Progress in Physical Geography 23 229 249 NCEP re analysis and Hadley Centre experiments Gordon C Cooper C Senior C Banks H Gregory J Johns T Mitchell J and Wood R 2000 The simulation of SST sea ice extents and ocean hea
94. ons FAQs section in Appendix 2 Finally definitions of commonly used technical terms related to statistical downscaling are provided in a Glossary 2 OVERVIEW OF SDSM STRUCTURE AND UKSDSM ARCHIVE Downscaling is justified whenever GCM or RCM simulations of variable s used for impacts modelling are unrealistic at the temporal and spatial scales of interest either because the impact scales are below the climate model s resolution or because of model deficiencies Downscaling may also be used to generate scenarios for exotic variables such as urban heat island intensity that can not be obtained directly from GCMs and RCMs However the host GCM must have demonstrable skill for large scale variables that are strongly correlated with local processes In practice the choice of downscaling technique is also governed by the availability of archived observational and GCM data because both are needed to produce future climate scenarios The SDSM software reduces the task of statistically downscaling daily weather series into seven discrete steps 1 quality control and data transformation 2 screening of predictor variables 3 model calibration 4 weather generation using observed predictors 5 statistical analyses 6 graphing model output 7 scenario generation using climate model predictors Select Quality predictand control Station data Select Screen y_ __ predictors variables Scatter plot
95. or a given time period Largest n day total calculates the n day total for all possible windows in the time period and presents the largest value Prec gt annual ile Percentage of total precipitation above the specified annual percentile All precip from events gt long term ile Percentage of all precipitation from events that are greater than the specified long term percentile No of events gt long term ile Count of the total number of events in the time period that are greater than the specified long term percentile Plot By clicking the Plot button the selected statistics are displayed as a time series graph as in Figure 12 2 In this case the raw data from PRCP DAT are plotted as a line chart covering the period 1961 1970 the maximum number of years that can be plotted as a line chart is 10 years Time Series Plot SDSM Time Series Chart PRCP DAT i Data points Figure 12 2 Time series plot of raw data from PRCP DAT file 12 2 Adjusting chart appearance The appearance of the time series chart can be adjusted in several ways For example by clicking on the Settings button the User is presented with a settings form that allows various adjustments to be made Figure 12 3 Time Series Chart Settings SDSM Time Series Chart Data points PRCP DAT Po Te cer Ts tow Lewd er cr at Figure 12 3 An example of the Time Series Chart Settings form Wilby amp Dawson 20
96. orrelation measures the strength of association between events separated by a fixed interval or lag The autocorrelation coefficient varies between 1 and 1 with unrelated instances having a value of zero For example temperatures on successive days tend to be positively autocorrelated Bayesian Information Criterion BIC A measure used to distinguish between two competing statistical models that takes into account the goodness of fit of the model whilst penalising models with larger numbers of parameters The BIC also depends on the number of data points and tends to favour simpler models compared with the AIC Black box Describes a system or model for which the inputs and outputs are known but intermediate processes are either unknown or unprescribed See regression Climate The average weather described in terms of the mean and variability of relevant quantities over a period of time ranging from months to thousands or millions of years The classical period is 30 years as defined by the World Meteorological Organisation WMO Climate change Statistically significant variation in either the mean state of the climate or in its variability persisting for an extended period typically decades or longer Climate change may be due to natural internal processes or to external forcings or to persistent anthropogenic changes in the composition of the atmosphere or in land use Climate model A numerical representation of the climate
97. ot Note that two File Selection windows are provided to allow the User to select files from different directories Only a maximum of five files from the two File Selection windows can be selected in total Data Allows the User to specify the required time period Note that if an attempt is made to plot a period longer than the available data set as defined in the global Settings an error message will appear Save Results To The User can choose to save a summary of the calculated results to a data file The option will not work if plotting Raw Data as no summary statistics are calculated in this case The default format text file is comma separated CSV so data can be opened in a spreadsheet for further analysis Clicking on the Clear button deselects the selected file Time Period Allows the User to select from Raw Data Month Season Annual or Water Year Raw Data simply graphs the data from the chosen file s as a daily time series plot for the selected period set by the User under Data Start and Data End No statistics are derived for Raw Data Note the Water Year runs from October to September and is referred to by the year in which it starts Season is referred to by the year in which it ends Winter December January February is referred to by the year in which the January February fall When selecting a Month Season Annual or Water Year Time Period SDSM calculates the chosen metric from the list below Select Statistic
98. predictand and regional scale atmospheric predictor variables The parameters of the regression model are written to a standard format file with the extension PAR along with meta data recording details of the calibration period model type predictors used etc see Section 6 4 SDSM optimises the model using either dual simplex or ordinary least squares optimisation see Advanced Settings in Section 3 2 The User specifies the model structure whether monthly seasonal or annual sub models are required whether the process is unconditional or conditional In unconditional models a direct link is assumed between the predictors and predictand e g local wind speeds may be a function of regional airflow indices In conditional models there is an intermediate process between regional forcing and local weather e g local precipitation amounts depend on wet dry day occurrence which in turn depend on regional scale predictors such as humidity and atmospheric pressure Furthermore it is possible to apply standard transformations to the predictand in conditional models see Section 3 2 and or to specific predictors see Section 4 2 To access the model building facility click the Calibrate Model button at the top of any screen The following screen will appear Calibrate Model File Edit Help A Home Quality controll GJ Transform Data Screen variables Calibrate Model weather Generator gt Summary Statist
99. proaches give comparable results Ordinary Least Squares is much faster The User can also select a Stepwise Regression model by ticking the appropriate box Stepwise regression works by progressively including more variables and selecting the most parsimonious model of the predictand according to one of two metrics either Akaike s Information Criterion AIC or the Bayesian Information Criterion BIC Settings File Locates standard and advanced settings held in a User defined reference file and directory A new or updated settings file is created whenever the Save button is clicked at the top of the screen The C SDSM INI settings file is automatically loaded whenever SDSM starts up Advanced Settings File Help 0 0 u o Back Reset Load Save Help formation Select File Figure 3 4 The Advanced Settings screen Press Reset at any time to reload the original settings or Back to return to the Settings screen followed by Back again to return to the last open screen 4 QUALITY CONTROL AND DATA TRANSFORMATION Few meteorological stations have complete and or fully accurate data sets Handling of missing and imperfect data is necessary for most practical situations In some cases it may also be necessary to transform data prior to model calibration SDSM enables both quality control and data transformation 4 1 Quality control To check an input file for missing data and or suspect values click on the Quality Cont
100. r present and future regional climate forcing Version 4 2 performs the tasks required to statistically downscale climate model output namely quality control of input data screening of candidate predictor variables model calibration synthesis of present weather data generation of future climate scenarios basic statistical and time series analyses and graphing results SDSM provides a robust and parsimonious technique of scenario construction that complements other methods e g direct use of climate model output dynamical downscaling sensitivity analysis etc Prospective Users should however consider the relative strengths and weaknesses of each category of downscaling to determine whether SDSM is most appropriate for the task in hand The authors strongly caution that the software should not be used uncritically as a black box This is a very real danger when employing regression based modelling techniques Rather the downscaling should be based upon physically sensible linkages between large scale forcing and local meteorological response Therefore good practice demands rigorous evaluation of candidate predictor predictand relationships using independent data Furthermore the local knowledge base is an invaluable source of information when determining sensible combinations of predictors Daily precipitation amounts at individual stations continue to be the most problematic variable to downscale and research is ongoing t
101. redictand and predictors as in Section 5 2 but for Select analysis period choose February from the drop down list Then click on the Correlation button at the top of the Screen Variables menu The results are shown in Figure 5 3 Results File Help CORRELATION MATRIX Analysis Period 01 01 1961 31 12 1990 Annual Missing values 0 Missing rows 0 1 2 3 4 5 TMAX DAT 1 0 083 0 074 0 034 0 679 neepp__uxx dat 0 083 1 0 001 0 043 0 083 ncepp_vss dat 0 074 0 001 1 0 023 0 194 neepp_zxx dat 0 034 0 043 0 023 1 0 562 neepp500xx dat 0 679 0 083 0 194 0 562 1 ncepulagss dat 0 055 0 759 0 057 0 037 0 037 ncepvlagss dat 0 041 0 037 0 519 0 236 0 055 neepzlagxx dat 0 068 0 002 0 202 0561 0 501 1 2 3 4 5 6 7 8 PARTIAL CORRELATIONS WITH TMAX DAT Partial r P value neepp__uxx dat 0 033 0 0015 necepp__vxx dat 0 071 0 0000 neepp__zxx dat 0 526 0 0000 ncepp500xx dat 0 823 0 0000 ncepulagsx dat 0 010 0 3421 ncepvlagxx dat 0 157 0 0000 ncepzlagxx dat 0 244 0 0000 Figure 5 3 The Results screen for the Blogsville example Partial correlations indicate that p500 and p_z have the strongest association with TMAX once the influence of all other predictors has been removed 5 4 Scatterplot The Scatter button is used for visual inspections of inter variable behaviour for specified sub periods annual seasonal or monthly The resultant scatterplot s indicate the nature of the association linear non linear etc
102. regional forcing Figure 5 2 so Monthly is checked in the Model Type box The Unconditional option is checked in the Process box because a direct link is assumed to exist between the regional scale predictors and local temperature The date range in the Data menu is set at 1961 to 1975 ensuring that the second half of the data i e 1976 to 1990 is retained for model validation Select Calculate to derive the Chow statistics for the model and in Advanced Settings the optimisation algorithm is set to Ordinary Least Squares Save the output results to TMAX61 75 PAR Once the appropriate selections have been made click on the Calibrate button The process may take several seconds and on completion a summary screen will appear Calibration Results see Figure 6 2 reporting the percentage of explained variance R squared value the Standard Error for the model the Chow statistic and Durbin Watson statistic for each month Calibration Results File Help Incepp__zxx dat ncepp500xx dat ncepylagxx dat ncepzlagzx dat Unconditional Statistics RSquared Durbin Watson 568 1 161 Figure 6 2 The Calibration Results screen Click on the Back button and a Scatter Plot will be displayed The example in Figure 6 3 shows an even spread of residuals across all values of the modelled predictand that is desirable 3 Scatter Plot File Edit Help Back b w 4 9 Reset Settings Copy Print Help Residual Plot g A A
103. ressure fields surface 500 and 850 hPa 2 4 SDSM file protocols For convenience the SDSM file protocol is described in two parts Firstly the file name system and file structure of the UKSDSM archive Secondly the meta data and output files produced by SDSM more generally 2 4 1 UKSDSM file structure and nomenclature Figure 2 3 shows how the directory structure of the UKSDSM data archive relates to ancillary file systems in SDSM The UKSDSM archive is organised into three levels At the highest level are the data sources presently NCEP HadCM2 HadCM3 CSIRO or CGCM2 At the second level are the nine cells shown in Figure 2 2 At the third level are files containing individual predictor variables Scenarios Calibration Results Present climate Station Blogsville TMAX DAT tmax sim TMAX PAR nceprhumee dat tmax out nceptempee dat nae tmax txt Figure 2 3 SDSM file structure with example file names see Table 2 2 for definitions of file name extension Each file in the archive complies with a generic nomenclature of the form source variable grid box dat The source is denoted by characters 1 4 the variable name by characters 5 8 and the grid box by characters 9 10 All files have the extension dat for example the file name nceprhumee dat indicates that the source is NCEP ncep the variable is near surface relative humidity rhum and the grid bo
104. rol button from any of the main screens The following screen will appear Quality Control File Edt Help D Home Qualty Control Transform Datal screen variables Calibrate model weather Generator summary Statistics Frequency Analysis Scenario Generator Compare Results Time Series Analysis v x Reset _ CheckFile Settings Select File Select File ecte Figure 4 1 The Quality Control screen Click on the Select File button An Open file window will appear browse through until you have located the directory and file to be checked in this example the Blogsville maximum daily temperature TMAX DAT Click on the required data file then on Open To activate the quality control procedure click on the Check File button at the top of the screen The following confirmation will appear Quality check complete x I 10957 values processed in TMAX DAT Figure 4 2 The Quality check complete dialogue box Click on the OK button to view the quality control information In this example there are 10957 values with no missing data i e no missing value codes of 999 were detected The data range from 6 7 to 34 8 C and have a mean of 13 1871 C see Figure 4 3 Click on the Reset button to clear the screen entries or select another file to perform a new quality check 3 Quality Control Ele Edt Help ba Home Qualty Control _ Transform Da
105. s The default is zero i e wet days are defined as all days with non zero precipitation totals Note the Sum is averaged by the number of years in the data set providing the monthly seasonal annual mean sum See section 8 3 for an explanation of the statistics available The third headed Delta Periods is where the Delta time periods are entered and the type of Delta Statistic is selected see below By checking the appropriate boxes the User can select up to eight statistics for analysis The defaults are the mean maximum minimum sum and variance Click on Back to return to the Summary Statistics screen Once all the above selections have been completed click on the Analyse button at the top of the menu After a few seconds the Results screen will appear Results File Help SUMMARY STATISTICS FOR TMAX DAT Analysis Start Date 01 01 1976 Analysis End Date 31 12 1990 Ensemble Member s Not applicable Month Mean Maximum Minimum Variance Sum January 6 443 13 800 6 700 13 899 199 733 February 6 418 17 900 2 900 15 259 181 427 March 9 516 20 100 1 600 9 384 295 007 April 11 865 22 900 2 100 11 937 355 940 May 15 782 26 100 7 400 12 711 489 247 June 18 542 31 800 9 900 14 887 556 273 July 21 148 32 200 13 600 13 415 655 580 August 20 619 34 800 0 000 12 011 639 180 September 17 924 26 400 10 700 8 109 537 713 October 14 141 27 100 7 700 7 949 438 360 November 9 627 17 100 0 600 10 930 288 800 December 7 493 15 900 3 700
106. s and scatterplots Ultimately however the User must decide whether or not the identified relationships are physically sensible for the site s in question Q How can I determine if I have chosen the correct predictor variables for the predictands that I require The correlation statistics and P values indicate the strength of the association between two variables Higher correlation values imply a higher degree of association Smaller P values indicates that this association is less likely to have occurred by chance A P value lt 0 05 is routinely used as the cut off so a P value of 0 37 would indicate that the predictor predictand correlation is likely to be due to chance However even if P lt 0 05 the result can be statistically significant but not be of practical significance there s a difference Even if a high correlation and low P value is returned the Scatterplot indicates whether this result is due to a few outliers or is a potentially useful downscaling relationship The Scatterplot may also reveal that one or both of the variables should by modified using the Transform operation to linearise the relationship Q What does the Event Threshold parameter in Settings do The Event Threshold parameter specifies the boundary between the two states in a Conditional process model For example if the Conditional process is precipitation changing the Event Threshold from 0 to 0 3 will result in more dry days and fewer
107. s for the specified period and plots them on a line graph For example if the User chooses a Time Period of January and selects Sum SDSM will plot the annual series of January Sums for the selected fit period i e the sum for January 1961 sum for January 1962 and so on as a line chart Select Statistics The User selects the summary statistics to be plotted by clicking the appropriate check button in this section default is Sum SDSM calculates the chosen statistic for the selected Time Period repeated across the range of the fit period and plots these as a time series chart A number of these statistics are based on the widely used STARDEX lt indices see Goodess et al 2007 Sum Mean Maximum are self explanatory measures for the selected time period Winter Summer ratio is calculated as the sum of the winter data December January February divided by the sum of the following summer data June July August The metric is referenced to the year in which the summer period falls Maximum dry wet spell the maximum dry wet spell length in days for the given time period Dry wet day persistence the total number of consecutive dry wet days for a period divided by the total number of dry wet days in that period Mean dry wet spell mean dry wet spell length for the period Median dry wet spell median dry wet spell length for the period SD dry wet spell standard deviation of dry wet spell length for
108. s is in turn used to resample observations at dependant locations using events occurring on the same date see Wilby et al 2003 for more details Preliminary tests of inter variable correlations produced by SDSM e g between downscaled precipitation and temperature series indicate that inter annual variations in the strength of relationships are preserved but there can be differences between the model and observations in individual months Once again it is suspected that inter variable relationships are implicitly preserved by virtue of commonality in the predictor variables used to downscale each predictand However if required it is relatively straightforward to explicitly condition one predictand on another e g daily precipitation occurrence might be used to condition maximum temperatures In this case the conditioning variable precipitation occurrence would be entered as a predictor during model calibration Q I ve calibrated my model How do I now produce values of PRCP TMAX or TMIN using GCM data Provided you have produced a PAR file via Calibrate Model the software will automatically know what predictors are needed Of course you may need to transform some of the GCM files if this was done for calibration For example if Z DAT was transformed to ZSQUARED DAT and then used to train the model the same transformation should be applied to the equivalent GCM file i e Z GCM to ZAQUARED GCM In which case be sur
109. s of SDSM As noted previously SDSM performs seven key functions The following paragraphs outline the purpose of each Further technical explanation and User guidance are provided in Sections 3 to 12 2 1 1 Quality control and data transformation Few meteorological stations have 100 complete and or fully accurate data sets Handling of missing and imperfect data is necessary for most practical situations Simple Quality Control checks in SDSM enable the identification of gross data errors specification of missing data codes and outliers prior to model calibration In many instances it may be appropriate to transform predictors and or the predictand prior to model calibration The Transform facility takes chosen data files and applies selected transformations e g logarithm power inverse lag binomial etc 2 1 2 Screening of downscaling predictor variables Identifying empirical relationships between gridded predictors such as mean sea level pressure and single site predictands such as station precipitation is central to all statistical downscaling methods The main purpose of the Screen Variables operation is to assist the user in the selection of appropriate downscaling predictor variables This is one of the most challenging stages in the development of any statistical downscaling model since the choice of predictors largely determines the character of the downscaled climate scenario The decision process is also complicated by
110. sville example The strongest correlation in each month is shown in red indicating that the relationship between maximum temperature and p500 and p__u are most important Blanks represent insignificant relationships at the chosen Significance Level For the Blogsville example select maximum daily temperatures as the predictand TMAX and the following predictor files p500 p__u p_ v and p_z see Table 2 1 In addition use the Transform facility Section 4 2 to create lagged values one day i e lag the data by 1 for the surface airflow indices The predictand does not depend on an intermediate occurrence process so Unconditional is checked under the Process option Use the default dates for the Data option i e 1961 1990 and choose Annual under Select Analysis Period Use the default Significance Level i e 0 05 then click on the Analyse button at the top of the Screen Variables menu The results in Figure 5 2 would suggest that p500 is a potentially useful predictor for April through October maximum temperature and p__u for December through March 5 3 Correlation matrix The Correlation button is used to investigate inter variable correlations for specified sub periods annual seasonal or monthly SDSM also reports partial correlations between the selected predictors and predictand These statistics help to identify the amount of explanatory power that is unique to each predictor For the Blogsville example use the same p
111. system based on the physical chemical and biological properties of its components their interactions and feedback processes and accounting for all or some its known properties Climate prediction An attempt to produce a most likely description or estimate of the actual evolution of the climate in the future e g at seasonal inter annual or long term time scales Climate projection A projection of the response of the climate system to emission or concentration scenarios of greenhouse gases and aerosols or radiative forcing scenarios often based on simulations by climate models As such climate projections are based on assumptions concerning future socio economic and technological developments Climate scenario A plausible and often simplified representation of the future climate based on an internally consistent set of climatological relationships that has been constructed for explicit use in investigating the potential consequences of anthropogenic climate change Climate variability Variations in the mean state and other statistics such as standard deviations the occurrence of extremes etc of the climate on all temporal and spatial scales beyond that of individual weather events Conditional process A mechanism in which an intermediate state variable governs the relationship between regional forcing and local weather For example local precipitation amounts are conditional on wet day occurrence the state variable which
112. t climate downscaled using observed NCEP predictors 1961 1990 and GCM HadCM3 predictors 1961 1990 By using the Compare Results operation again it is possible to compare the frequency of hot days at Blogsville under present 1961 1990 and future 2080 2099 climate forcing This was achieved by comparing POT statistics for TMAXCCF61 90 TXT with TMAXFCF70 99 TXT For example Figure 10 6 shows a significant increase in the frequency of hot days in summer by the end of the 21 century The downscaling also indicates that hot days could begin to appear as early as May by the end of the 21 century File Edit Help Reset Settings Copy Frequency of hot days 1961 90 2070 99 o E a gt a a Jan Feb Mar Apr May l Jun I Jul aug l Sep l oct Nov Dec Figure 10 6 Monthly frequency of hot days gt 25 C at Blogsville downscaled using HadCM3 predictors under present 1961 1990 and future 2070 2099 forcing 10 4 Blogsville example precipitation Precipitation downscaling is necessarily more problematic than temperature because daily precipitation amounts at individual sites are relatively poorly resolved by regional scale predictors and because precipitation is a conditional process i e both the occurrence and amount processes must be specified Figure 10 7 shows the PAR file PRCP61 90 PAR generated when a precipitation model was generated using
113. t four predictors were employed line 1 to simulate 12 months line 2 using calendar years line 3 beginning in 01 01 1961 line 4 and lasting 10957 days line 5 The model was conditional TRUE line 6 had 20 ensemble members line 7 variance inflation line 8 a fourth root transformation of the predictand line 9 and bias correction of 0 8 line 10 The predictand file was PRCP DAT and the four predictors were p_v p_z p500 and shum lines 12 onwards With the above specifications the Weather Generator was used to downscale observed NCEP predictors and Scenario Generator to downscale GCM HadCM3 predictors representing the present climate saved as PRCPNCEP61 90 0UT and PRCPCCF61 90 OUT respectively Note that a Year Length of 366 days should be checked in Settings when working with NCEP and 360 when using HadCM3 predictors Downscaled scenarios were evaluated firstly using the Summary Statistics and then Compare Results Figure 10 9 shows the summary statistics for the downscaling using NCEP predictors and Figure 10 10 shows the equivalent results for the downscaling using GCM predictors Results File Help SUMMARY STATISTICS FOR PRCPNCEP61 90 0UT Analysis Start Date 01 01 1961 Analysis End Date 31 12 1990 Ensemble Member s ALL Month Mean Maximum Minimum Variance Sum January 2 958 30 887 0 001 15 054 49 684 February 3 041 33 451 0 001 18 197 43 055 March 2 654 27 855 0 001 12 647 42 008 April 3 03
114. t transport in a version of the Hadley Centre coupled model without flux adjustments Climate Dynamics 16 147 168 Johns T C Carnell R E Crossley J F Gregory J M Mitchell J F B Senior C A Tett S F B and Wood R A 1997 The Second Hadley Centre coupled ocean atmosphere GCM Model description spinup and validation Climate Dynamics 13 103 134 Kalnay E Kanamitsu M Kistler R et al 1996 The NCEP NCAR 40 year reanalysis project Bulletin of the American Meteorological Society 77 437 471 Mitchell J F B and Johns T C 1997 On modification of global warming by sulphate aerosols Journal of Climate 10 245 267 Mitchell J F B Johns T C Gregory J M and Tett S 1995 Climate response to increasing levels of greenhouse gases and sulphate aerosols Nature 376 501 504 Tett S F B Johns T C and Mitchell J F B 1997 Global and regional variability in a coupled AOGCM Climate Dynamics 13 303 323 Statistical downscaling methods Bardossy A Bogardi I and Matyasovszky I 2005 Fuzzy rule based downscaling of precipitation Theoretical and Applied Climatology 82 119 129 Bardossy A and Plate E J 1992 Space time model for daily rainfall using atmospheric circulation patterns Water Resources Research 28 1247 1259 Beersma J J and Buishand T A 2003 Multi site simulation of daily precipitation and temperature conditional on the atmospheric circulation Climate Research 2
115. ta _ Screen Variables Summary Statistics Frequency Analysis Scenario Generator GJ _ Compare Results Calibrate Modell Weather Generator QO Time Series Analysis O v X Reset _ CheckFile Settings Minimum Figure 4 3 Results of the Quality Control check for TMAX DAT 4 2 Data transformation To transform data click on the Transform Data button from any of the main screens The following screen will appear Transform Data File Ele Edt Help A Home quality control Summary Statistics Frequency Analyses o m X Reset_ Transform Settings Transform Data Screen variables Caltbrate Modell E weather Generator _ Compare Results Time Series Analysis Scenario Generator t Input File Select File Output File File Not entered Figure 4 4 The Transform Data File screen Click on the Select Input File button An Open file window will appear Browse through until you have located the directory and file to be transformed for example the surface vorticity over Eastern England ncepp__zee dat Click on the required file If there is more than one column of input data as in the case of an ensemble simulation produced by the Weather Generator or Generate Scenario functions see Sections 7 and 10 enter the appropriate number in the Columns in Input File box To enable transformed data with multiple columns to be handled by
116. the fact that the explanatory power of individual predictor variables varies both spatially and temporally Screen Variables facilitates the examination of seasonal variations in predictor skill 2 1 3 Model calibration The Calibrate Model operation takes a User specified predictand along with a set of predictor variables and computes the parameters of multiple regression equations via an optimisation algorithm either dual simplex of ordinary least squares The User specifies the model structure whether monthly seasonal or annual sub models are required whether the process is unconditional or conditional In unconditional models a direct link is assumed between the predictors and predictand e g local wind speeds may be a function of regional airflow indices In conditional models there is an intermediate process between regional forcing and local weather e g local precipitation amounts depend on the occurrence of wet days which in turn depend on regional scale predictors such as humidity and atmospheric pressure 2 1 4 Weather generator The Weather Generator operation generates ensembles of synthetic daily weather series given observed or NCEP re analysis atmospheric predictor variables The procedure enables the verification of calibrated models using independent data and the synthesis of artificial time series for present climate conditions The User selects a calibrated model and SDSM automatically links all necessary
117. the following parameters 4 predictors ncepp_vxx dat ncepp__zxx dat ncepp500xx dat and ncepshumxx dat monthly model using NCEP data from 1961 1990 with a fourth root transformation of the predictand E PRCP61 90 PAR WordPad Ble EGR wew port Forma tiep Osc SNR A e amp 12 01 01 1961 10957 01 01 1961 0 052 2 0 000 0 035 157 2S5 0 000 0 075 0 27 0 000 0 049 2 2 0 000 0 082 e 2 9 900 0 067 x 45 217 0 000 0 109 0 05 7 2 0 000 0 056 a o 3 0 000 0 087 x o x 0 000 0 060 007 147 0 000 0 071 0 02 s 2 0 000 0 058 0 001 277 0 000 1 000 0 054 100 0 064 1 000 0 026 2 0 165 1 000 0 068 0 025 0 095 1 000 0 064 0 017 1 0 141 1 000 0 043 2 12 0 169 1 000 0 078 2 22 0 169 1 000 0 063 s et 0 152 1 000 0 042 069 0 153 1 000 0 051 o 3 0 115 1 000 a a 0 107 1 000 016 x at 0 159 1 000 075 0 027 237 0 144 C SDSM Blogavi i ie observedi961 90 PRCP DAT Figure 10 7 shows a PAR file for a precipitation model Figure 10 8 shows a SIM file used to downscale daily precipitation from observed NCEP predictors using the PRCP61 90 PAR model PRCPNCEP61 90 SIM W DBR File Edit View Insert Format Help Dae 66 igen 4 12 366 01 01 1961 10957 TRUE 20 18 2 0 8 PRCP DAT neepp vxx dat neepp 2xx dat ncepp500xx dat ncepshumxx dat For Help press F1 Figure 10 8 The SIM file for downscaling precipitation at Blogsville 1961 1990 Figure 10 8 shows tha
118. the period Spell length correlation a measure of the combined persistence of wet and dry spells for the period Partial Duration Series is calculated as the sum of data values less than or equal to the chosen threshold for the selected Time Period The default is the threshold value held in the main Settings screen This value can be adjusted by entering the required threshold in the text box this will not affect the global threshold value set in the main Settings screen and applied elsewhere Percentile calculates the specified percentile for the chosen Time Period The default is 90 but this can be adjusted by entering the required value in the text box Standard Precipitation Index SPI This is calculated for monthly time series only so the Time Period selection is ignored when SPI is chosen The SPI is derived by first calculating the monthly sums of the data then calculating a moving average of these monthly sums smoothing across the time period entered by the User in the adjacent text box The default moving average period is 3 months The smoothed data are then normalised by subtracting the mean of all the data in the fit range and dividing by the standard deviation of the smoothed data for each month Peaks Over Threshold POT This counts the number of events greater than the user specified threshold for the chosen time period Nth largest determines the nth largest value when the data are sorted into descending order f
119. tic climate model is run for multiple climate projections each with minor differences in the initial or boundary conditions Conversely weather generator ensemble members differ by virtue of random outcomes of successive model simulations In either case ensemble solutions can be grouped and then compared with the ensemble mean to provide a guide to the uncertainty associated with specific aspects of the simulation External forcing A set of factors that influence the evolution of the climate system in time and excluding natural internal dynamics of the system Examples of external forcing include volcanic eruptions solar variations and human induced forcings such as changing the composition of the atmosphere and land use change Extreme weather event An event that is rare within its statistical reference distribution at a particular place Definitions of rare vary from place to place and from time to time but an extreme event would normally be as rare or rarer than the 10th or 90th percentile General Circulation Model GCM A three dimensional representation of the Earth s atmosphere using four primary equations describing the flow of energy first law of thermodynamics and momentum Newton s second law of motion along with the conservation of mass continuity equation and water vapour ideal gas law Each equation is solved at discrete points on the Earth s surface at fixed time intervals typically 10 30 minutes for
120. up to a maximum of 100 and enter the appropriate integer in the Ensemble Size box on the right hand side of the screen the default is 20 Finally to save the scenario data to a results file it is necessary to select an appropriate directory and file name Click on the Select Output File button in the top right hand corner An Open file window appears browse through until the correct directory is reached then enter a suitable file name for example TMAXCCF61 90 0UT maximum temperature present climate forcing 1961 1990 The name of the file will then appear beneath the button Once the above selections have been completed click on the Generate button at the top of the screen After a short while a dialogue box will appear Figure 10 2 Click on OK to return to the Generate Scenario screen Results 1 Scenario Generated Figure 10 2 The Scenario Generated dialogue box 10 3 Blogsville example temperature For the Blogsville example the Scenario Generator operation was applied twice First predictors from the HadCM3 experiment for the period 1961 1990 were used to downscale present climate forcing Figure 10 3 shows the Results screen for this scenario using the Summary Statistics operation see Section 8 Results File Help SUMMARY STATISTICS FOR TMAXCCF61 30 0UT Analysis Start Date 01 01 1961 Analysis End Date 31 12 1990 Ensemble Member s ALL Month Mean Maximum Minimum Variance POT January 5 955 18 1
121. versus observed or present versus future climate scenarios The Time Series Analysis screen allows the User to produce time series plots for up to a maximum of five variables The data can be analysed as monthly seasonal annual or water year periods for statistics such as Sum Mean Maximum Winter Summer ratios Partial Duration Series Percentiles and Standardised Precipitation Index 2 1 7 Scenario generation Finally the Scenario Generator operation produces ensembles of synthetic daily weather series given atmospheric predictor variables supplied by a climate model either for present or future climate experiments rather than observed predictors This function is identical to that of the Weather Generator operation in all respects except that it may be necessary to specify a different convention for model dates and source directory for predictor variables The input files for both the Weather Generator and Scenario Generator options need not be the same length as those used to obtain the model weights during the calibration phase 2 2 UKSDSM data archive As Figure 2 1 indicates the SDSM procedure begins with the preparation of coincident predictor and predictand data sets Although the predictand is typically an individual daily weather series obtained from meteorological observations at single stations e g daily precipitation maximum or minimum temperature hours of sunshine wind speed etc the methodology is applicable to oth
122. well as multi site applications However weather typing schemes ca be parochial a poor basis for downscaling rare events and entirely dependent on stationary circulation to surface climate relationships Potentially the most serious limitation is that precipitation changes produced by changes in the frequency of weather patterns are seldom consistent with the changes produced by the host GCM unless additional predictors such as atmospheric humidity are employed 1 1 3 Stochastic weather generators Stochastic downscaling approaches typically involve modifying the parameters of conventional weather generators such as WGEN LARS WG or EARWIG The WGEN model simulates precipitation occurrence using two state first order Markov chains precipitation amounts on wet days using a gamma distribution temperature and radiation components using first order trivariate autoregression that is conditional on precipitation occurrence Climate change scenarios are generated stochastically using revised parameter sets scaled in line with the outputs from a host GCM The main advantage of the technique is that it can exactly reproduce many observed climate statistics and has been widely used particularly for agricultural impact assessment Furthermore stochastic weather generators enable the efficient production of large ensembles of scenarios for risk analysis The key disadvantages relate to the low skill at reproducing inter annual to decadal climate vari
123. x is Eastern England ee Similarly the file name h3b2p8_zsw dat indicates that the source is HadCM3 SRES scenario B2 h3b2 the variable is vorticity computed at the 850 hPa geopotential height pS_z and the grid box is Southwest England sw Alternatively the file name h2ggp_thsb dat indicates that the source is HadCM2 greenhouse gas only experiment A2gg the variable is surface wind direction p_th and the grid box is Scottish Boarders sb 2 4 2 SDSM file name protocol With the above prerequisites in mind Table 2 2 lists the file name extensions employed by SDSM and Figure 2 3 shows the associated directory structures All input and output files are text only format Individual predictor and predictand files one variable to each file time series data only are denoted by the extension dat The PAR file records meta data associated with the model calibration model weights and measures of goodness of fit percentage explained variance and standard error of the model The SIM file records meta data associated with every downscaled scenario e g number of predictor variables ensemble size period etc and the OUT file contains an array of daily downscaled values one column for each ensemble member and one row for each day of the scenario Finally TXT files are created whenever statistical analyses are undertaken by SDSM These files record summary statistics for individual ensemble members or for t
124. y to show a line chart click on the Line button at the top of the screen Eile Edit Help Back Reset Settings Copy SDSM Line Chart 4 TMAXOBS61 90 TXT Mean ww N TMAXNCEPG1 90 TXT Mean IsTalglglalg Y axis label Figure 11 2 Example of the Line chart using observed TMAXOBS61 90 and NCEP downscaled TMAXNCEP61 90 monthly mean maximum daily temperatures at Blogsville 1976 1990 11 2 Bar chart Alternatively having selected the required files and statistics from each list as in Section 11 1 click on the Bar button at the top of the Compare Results screen to produce a bar chart File Edit Help Back Reset Settings Copy SDSM Bar Chart 70 E TMAXOBS61 90 TXT Maximum E TMAXNCEP61 90 TXT Maximum 2 s 2 x oi gt Figure 11 3 Example of the Bar chart using observed TMAXOBS61 90 and downscaled TMAXNCEP61 90 monthly absolute maximum daily temperatures at Blogsville 1976 1990 11 3 Customizing charts To change or remove tick marks y axis labels chart titles or y axis maximum minimum in either the Line or Bar chart click on the Settings button at the top of the screen The following screen will appear Chart Settings File Help 0 9 Close Reset Help Enter number of tick points 0 Chart title S5DSM Bar Chart Enter new Y axis maximum 70 Y axis label v axis label Enter new Y axis minimum 0 Legend 1 title Twax0BS61 90 TXT M
125. y used for dynamical and statistical downscaling of the present climate Normalisation A statistical procedure involving the standardisation of a data set by subtraction of the mean and division by the standard deviation with respect to a predefined control period The technique is widely used in statistical downscaling to reduce systematic biases in the mean and variance of climate model output Parameter A numerical value representing a process or attribute in a model Some parameters are readily measurable climate properties others are known to vary but are not specifically related to measurable features Parameters are also used in climate models to represent processes that poorly understood or resolved Partial Duration Series Events above a defined threshold that are recorded as a time series or as a frequency distribution Essentially a peaks over threshold approach to describing the occurrence of extreme events Predictand A variable that may be inferred through knowledge of the behaviour of one or more predictor variables Predictor A variable that is assumed to have predictive skill for another variable of interest the predictand For example day to day variations in atmospheric pressure may be a useful predictor of daily rainfall occurrence Probability Density Function PDF A distribution describing the probability of an outcome for a given value for a variable For example the PDF of daily temperatures often approx

Download Pdf Manuals

Related Search

Related Contents

QTP 22 - QTP 24  120403 User manual Checkweighers BCW-4000  NeOn Toolkit User's Guide  MICHELIN X One Truck Tire Service Manual  Manuel d`utilisation  Zotac ZT-70602-10M NVIDIA GeForce GTX 750 Ti 2GB graphics card  SITOP PSU400M 24 V/20 A 6EP1536  KVT-819DVD KVT-839DVD  Dell OpenManage Server Administrator Version 5.4 SNMP Reference Guide  afman34-129  

Copyright © All rights reserved. Failed to retrieve file