Home

The Bureau of Meteorology Statistical Downscaling Model Graphical

1. 1 0 1 0 9 OSummer Autumn Winter O Spring fs OSummer GAutumn Winter O Spring piper 0 8 dean RA 2 Da i e 07 o 2 06 0 54 T 044 0 5 O Susi O 03 Gio 0 2 4 0 3 014 PAL TAL IR tel ciel at 0 2 0 0 0 1 LIL LUI Et ea 4 4 Ual SY amp 20 e 9 gt amp 20 0 0 T T T T T T T T T Tmax SEL F s SL Tmin SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC _ _ 0 8 r OSummer GAutumn Winter Spring OSummer BD Autumn Winter Spring 0 7 0 6 5 5 o 5 3 B04 5 031 5 8 03 0 2 1J IJ LJ LIL LIL 0 1 4 r T r T T T r T 0 0 Rain SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC pE SWA NUL SEA SMD SEC TAS 1 0 z ae Summer Autumn Winter OSpring 1 0 T Summer GAutumn Winter Spring I 0 9 0 8 0 8 0 7 0 7 c c 2 0 6 4 2 0 6 T p o5 3 0 5 6 0 4 4 5 044 0 3 9 0 3 4 0 2 4 0 2 4 cs oe Ce Cae sees caee wie es cats oot co Omi cei ie 0 0 0 0 DTmax SWA NUL SEA SMD SEC TAS DTmin SWA NUL SEA SMD SEC TAS Fig B4 As per Figure B3
2. 0 7 5 0 7 2 06 S 06 ul p 0 5 5 D 0 5 8 0 4 4 8 0 4 4 0 3 5 0 3 4 o t i i j zi 0 1 li 01 0 0 r 7 r r r r r T I 0 0 T i T Rain SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC pE SWA NUL SEA SMD SEC TAS 1 0 1 0 0 9 OSummer m Autumn Winter O Spring 0 9 4 OSummer E Autumn Winter Spring 0 8 5 0 8 c 07 c 07 2 0 6 0 6 9 0 5 3 0 5 8 0 4 8 0 4 0 3 0 3 0 2 0 2 0 1 0 1 0 0 0 0 i T T DTmax SWA NUL SEA SMD SEC TAS DTmin SWA NUL SEA SMD SEC TAS Figure B3 Correlation between daily observed and reconstructed series separated by season different coloured bars and region names on the x axis for the six predictands considered names at the bottom left corner Each value is the average correlation across all the stations available in a particular region For Tmin Tmax and Rain 10 regions are available only the 6 covering the southern half of the continent are available for pE dTmin and dTmax The ability of the modelled series to reproduce year to year variability is also important in a climate change context It is evaluated by computing the Pearson correlation between seasonal means of the observed and reconstructed series Fig B4 60 BoM SDM GUI documentation Authors B Timbal et al in the same way as daily variability The length of the observed record is 1958 to 2003 for all predictands except for pan evaporation for which it is 1975 to 2003
3. 4 5 6 7 Bureau of Meteorology and C S I R O 2007 Australian Regional Climate Change Projections AGO Technical report 150 pp Burton A C Kilsby H Fowler P Cowpertwait and P O Connell 2008 RainSim A spatial temporal stochastic rainfall modelling system Env Model Software in press Charles S B Bates P Whetton and J Hughes 1999 Validation of downscaling models for changed climate conditions case study of southwestern Australia Clim Res 12 1 14 Charles S B Bates and N Viney 2003 Linking atmospheric circulation to daily rainfall patterns across the Murrumbidgee River Basin Water Sci Tech 48 7 233 240 Christensen J B Hewitson A Busuioc A Chen X Gao R Jones R Kolli W T Kwon V Magafia Rueda L Mearns C Mefiendez J Raisanen A Rinke A Sarr and P Whetton 2007 In Climate Change 2007 The Physical Science Basis Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change chap Chapter 11 Regional Climate Projections Cambridge University Press Cambridge United Kingdom and New York NY USA Drosdowsky W 1993 An Analysis of Australian seasonal Rainfall anomalies 1950 1987 I Spatial Patterns Int J Climatology 13 1 30 Hessami M P Gachon T Ouarda and A St Hilaire 2008 Automated regression based statistical downscaling tool Env Model Software 23 6 813 834 44
4. Observations from 1960 to 1999 are shown as a thick black line on the 20 century diagram for the 21 century case the period 1970 to 1989 is used to provide a reference for the downscaled model simulation for 2046 to 2065 The thresholds used depend on the mean and variance of the series and differ from observations to models and from the 20 century to the 21 century as shown on the graphs For example the observed probability that hot spells above 36 7 C last up to 8 days is about 0 01 in Mildura The same probability for the downscaled model suggests a spell length between 5 and 7 days left graph in Fig G2 but the threshold used which depends on the mean and variance of the individual series matches the observed threshold Keeping these results in mind future projections of hot spell duration show that much larger thresholds are used due to the mean warming but the 85 BoM SDM GUI documentation Authors B Timbal et al probability of spell length is hardly changed Also there is an indication that longer spells could become more likely as model curves are slightly more clustered toward the observed trend than they were in the current climate case but this may not be significant as it is likely to be by less than a day apart for more extreme spells right graph in Fig G2 Of course if the threshold used for the 20 century were kept results would be very different as days above 36 7 C are projected to be far more freq
5. seo The Centre for Australian Weather and Climate Research i A partnership between CSIRO and the Bureau of Meteorology Australian Government Bureau of Meteorology CSIRO The Bureau of Meteorology Statistical Downscaling Model Graphical User Interface user manual and software documentation CAWCR Technical Report No 004 B Timbal Z Li and E Fernandez December 2008 www cawcr gov au The Bureau of Meteorology Statistical Downscaling Model Graphical User Interface User Manual amp Software Documentation Authorship The Graphical User Interface GUI was developed in the research centre of the Bureau of Meteorology by Zhihong Li under scientific guidance from Bertrand Timbal during the course of the development Bradley Murphy and Elodie Fernandez contributed to this work The documentation was compiled by B Timbal Z Li and E Fernandez The development of the Bureau of Meteorology statistical downscaling model BoM SDM has long been supported by the Department of Climate Change DCC as part of the Australian Climate Change Science Program ACCSP While this CAWCR research report summarises the development of the BoM SDM as of the date of publication the work is still on going and users are encouraged to contact the authors to find out about the latest developments Enquiries should be addressed to Bertrand Timbal Centre for Australian Weather and Climate Research A Partnership between the B
6. BoM SDM GUI documentation Authors B Timbal et al 8 Hewitson B and R Crane 2006 Consensus between GCM climate change projections with empirical downscaling Precipitation downscaling over South Africa Int J Climatol 26 1315 1337 9 Jones D W Wang and R Fawcett 2007 Climate Data for the Australian Water Availability Project Final milestone report Bureau of Meteorology 10 Jovanovic B D A Jones and D Collins 2008 A High Quality Monthly Pan Evaporation Dataset for Australia Climatic Change published on line 11 Kalnay E M Kanamitsu R Kistler W Collins D Deaven J Derber L Gandin M Iredell S Saha G White J Woollen Y Zhu M Chelliah W Ebisuzaki W Higgins J Janowiak K Mo C Ropelewski J Wang A Leetma R Reynolds R Jenne and D Joseph 1996 The NCEP NCAR 40 year reanalysis project Bull Amer Meteor Soc 77 431 411 12 Kilsby C P Jones A Burton A Ford H Fowler C Harpham P James A Smith and R Wilby 2007 A daily weather generator for use in climate change studies Env Model Software 22 12 1705 1719 13 Lavery B A Kariko and N Nicholls 1992 A historical rainfall data set for Australia Aus Met Mag 40 33 39 14 Lavery B G Joung and N Nicholls 1997 An extended high quality historical rainfall dataset for Australia Aus Met Mag 46 27 38 15 Lorenz E 1969 Atmospheric Predictability as Revealed by Naturally O
7. and long term linear trends 11 BoM SDM GUI documentation Authors B Timbal et al In each of the 192 cases determining the best statistical scheme to use i e the optimisation of the SDM was based on a subjective analysis of the various metrics listed above No attempt was made to develop an objective approach to obtain the optimal combination of predictors Several reasons can be listed to justify allowing for expert assessment e Predictors are not strictly independent and therefore their importance as a single predictor is not necessarily a good guide for their contribution in a combination of predictors due to their inter dependence e It is generally not the case that a single combination of predictors will yield superior results for all metrics used and hence a trade off exists between the importance of the various predictors and their role e The choice of predictors is also driven by the reliability of these variables in climate models For all these reasons it was decided that a subjective analysis of the optimal combination of predictors is the most effective way to optimise individual SDMs Each SDM was developed in two stages In the first stage the best combination of predictors was determined in the second stage additional parameters were optimized The BoM SDM includes a large number of choices and each could be considered as a tuneable parameter However previous studies have shown that only three paramete
8. 0 6 and 1 C at the 10 percentile 1 to 1 5 C at the 50 percentile and 2 to 2 5 C at the 90 percentile These values are comparable with the suggested warming of 1 3 t 02 2 C shown here for the A2 scenario a high emissions scenario and 0 8 to 2 C for the B1 scenario a 84 BoM SDM GUI documentation Authors B Timbal et al low emissions scenario The range is not as large as with the direct model outputs but in that case the entire CMIP3 database 23 models were used compared with only 10 models in our case study and the low and high emissions scenarios in the CSIRO and BoM 2007 projections are based on more than the single scenario B1 or A2 used here In addition to the mean local warming and the changes of the PDF it is possible to take a closer look at the details that the local warming implies For example the length and duration of hot spells number of consecutive days above a particular threshold can be looked at The downscaling of the climate models underestimates the hot spell duration Fig G2 HSD milduro HSD milduro QBS Thre 36 7 GCM 36 9 GCM 36 5 OBS Thre 36 7 GCM 38 7 GCM 37 9 Pb of duration Pb of duration Figure G2 Hot spell duration expressed as the probability of having a certain number of days exceeding a certain threshold for Tmax in summer in Mildura for the end of the 20 century left and for the middle of the 21 century using emissions scenario A2 right
9. Jovanovic et al 2008 Following on the early work on temperature and rainfall the SDM was optimised for the new HQ network Timbal et al 2008 Therefore downscaled projections for 6 surface predictands can currently be obtained through the GUI y Rainfall Tmax T min pE AT max AT min SU 0 The technique was first extended to the entire extra tropical half of the Australian continent generally south of 33 S Timbal et al 2008a and more recently to the Tropical half of the Australian continent It has then been applied to the latest round of GCM projections for the 21 century which were assembled as part of the 4m Assessment of the Intergovernmental Panel on Climate Change IPCC The continent was separated into ten distinct climate entities Fig 1 roughly following the rotated Empirical Orthogonal Functions EOFs for rainfall suggested by Drosdowsky 1993 1 The Southwest of Western Australia SWA similar to the Indian Ocean Climate Initiative IOCI triangle IOCI 1999 south of a north boundary from Geraldton to Kalgoorlie about 30 S and west of a line from Kalgoorlie to Esperance about 122 E BoM SDM GUI documentation Authors B Timbal et al 2 The Nullarbor NUL region a vast region from the SWA in the west and the Eyre Peninsula in South Australia in the east about 136 E and from the coast to 30 S This area has very few high quality observations and therefore its optimisations are likely
10. SMD dash dotted SEC solid line TAS dash dotted NWA dash dotted NMR dashed QLD dotted and MEC dash dotted Small domain per regions Small domain per seasons 80 80 70 70 60 60 50 50 40 ie 30 30 20 A 10 ae 0 10 DJF MAM JJA SON 0 Figure A4 Percentage of cases in which the smaller geographical window is the best domain to search for analogues Two sizes were tested for each SDM However the most important factor appears to be the dependence on the region Although the original two sizes tested for each region were chosen carefully based on rules derived from previous studies regarding size and position Timbal and 53 BoM SDM GUI documentation Authors B Timbal et al McAvaney 2001 there were cases such as SWA where the small domain was too small and hence seldom chosen only 8 of the time whereas in SMD the small domain was chosen 75 of the time and between 29 and 67 in the other regions The choice between monthly and seasonal anomalies reveals a marked preference for using a single seasonal mean Fig A5 left in particular for the transitional seasons autumn 92 and spring 88 of all cases This result was expected as the underlying assumption behind using monthly mean is to reduce the non stationarity across the months when calculating the anomalies hence reducing the importance of the annual
11. Storch 1999 Simplicity flexibility and robustness are essential to ensure that a single technique can be used across a range of variables and several climatic regions 39 BoM SDM GUI documentation Authors B Timbal et al This is a landmark project offering downscaled climate change projections across a large part of the Australian continent It is worth noting that currently downscaled predictands series are constructed independently from one variable to another This is possibly a limitation for impact studies that require several predictands i e rainfall and temperature b Case studies performed with this tool A complete analysis of the outputs generated by the SDM GUI is not viable users are encouraged to generate outputs relevant to their particular location and application and to document results and possible issues arising from this exercise Nevertheless some case studies have been documented in previous publications and are repeated in the appendices to illustrate the benefits and some known limitations from the downscaled projections 1 Future projections for Tmax in Mildura Victoria are shown in Appendix G this case was documented in Timbal et al 2008a and 2 Spatial heterogeneities within a typical model grid box for both temperature along the NSW coast and rainfall in the southwest of Western Australia are shown in Appendix H these cases were documented in the Australian Climate Change rep
12. a sufficient predictor In these rare occurrences with two very similar and highly correlated predictors being used the risk of over fitting the SDM exists However in these instances it was found that the gain in skill for the SDM warranted this choice Finally thermal predictors rarely matter for rainfall to reproduce current climate however their importance in a future warmer world remains possible E g it was found while comparing analogue base projections for rainfall in the South West of Western Australia with direct model outputs and another SDM that the inclusion of Tg5o was necessary to have consistent projections between the three approaches Timbal et al 2008 Moisture variables Fig A2 bottom left are also important as predictors across all predictands with the notable exception of Tmax Specific humidity is almost always picked up apart from pan evaporation for which relative humidity is more skilful Rainfall is often part of the optimised predictor s combination to downscale rainfall As for temperature there are instances about 20 of the cases where it is used in combination with another low level tropospheric moisture fields suggesting a possible over fitting Finally some measure of the air flow either the zonal u or meridional v component of the wind is often added to the optimised combination Fig A2 bottom left The fact that it is an additional predictor to the de facto combination of synoptic
13. al The Australian Bureau of Meteorology has developed a Statistical Downscaling Model SDM to downscale large scale predictors from coupled global atmospheric oceanic General Circulation Models GCMs to local surface predictands the Australian continent The method is based on the idea of analogous synoptic situations Using statistically downscaled climate projections require some understanding of the SDM its strengths and weaknesses It is highly recommended that prior to running the model user familiarised themselves with the scientific literature relevant to this tools provided in the documentation page New user should familiarise themselves with the tool by reading the User manual and software documentation also available on line Tf you have any query relevant to this GUI use the contact links available Please note this is a beta version and hence feedbacks to improve this tool are strongly encouraged Click the button below to start running the Graphical User Interface GUT and access the SDM Start the GUI Figure 4 The front page of the GUI for the Bureau of Meteorology SDM c The interactive web interface SelectM S DeleteM S Selectalisn Figure 5 The default interactive GUI as it appears to the user before any action is taken 30 BoM SDM GUI documentation Authors B Timbal et al The starting point of the GUI Fig 5 displays a large region around the Australian continent using Map
14. amp PRCP amp Vs50 Autumn MSLP amp Tmax Tsso amp Qsso MSLP amp PRCP amp Vs50 SMD Winter MSLP amp Tsso amp Tmax amp MSLP amp Ts50 amp Qg50 MSLP amp PRCP amp Vsso Spring Usso MSLP amp Tso amp Qsso MSLP amp PRCP amp Vsso Summer MSLP amp Tmax MSLP amp Tsso amp Qs50 MSLP amp Tmax amp Qg50 amp Usso Autumn MSLP amp Tmax MSLP amp Ts50 amp Qs50 MSLP amp PRCP amp Qsso amp Usso SEC Winter MSLP amp Tmax MSLP amp Tsso amp Tmin amp Usso MSLP amp PRCP amp Usso Spring MSLP amp Tsso amp Tmax amp MSLP amp Tsso amp Qsso MSLP amp PRCP amp Qs50 amp Usso Summer MSLP amp Tsso amp Tmax amp Usso MSLP amp Ts50 amp Qs50 MSLP amp Qgs0 amp Vs50 Autumn MSLP amp Tg50 amp Tmax amp Usso MSLP amp Tsso amp Qsso MSLP amp Qg50 amp Veso TAS Winter MSLP amp Teso amp Qsso amp Usso MSLP amp Teso amp Tmin amp Usso MSLP amp PRCP amp Qsso amp Usso Spring MSLP amp Tsso amp Tmax amp Usso MSLP amp Ts50 amp Tmin amp Usso MSLP amp PRCP amp Qs50 amp Usso Summer Tss0 amp Usso amp Tmax Tsso amp Qeso PRCP amp V7 amp Reso Autumn Ts50 amp V700 amp Tmax Tg50 amp Q70 PRCP NWA Winter Tsg50 amp Tmax Tsg50 amp Tmin amp Qg50 PRCP amp Qg50 Spring Tsso Tit Qro amp Tmin PRCP amp Rm Summer Tss0 amp Usso amp Tmax Tin PRCP Autumn Trax Tin PRCP amp Tmax amp V350 QLD Winter Tmax amp MSLP Tmin PRCP Spring Tsso amp Qsso
15. but for seasonal means In addition the dashed lines indicate the 95 confidence level for these correlations The percentage of observed inter annual range i e the difference between the highest and lowest seasonal totals in the observed record reproduced by the reconstructed series is also evaluated Fig B5 61 BoM SDM GUI documentation Authors B Timbal et al 100 OSummer Autumn EB Winter D Spring 100 n 90 90 OSummer Autumn Winter D Spring a 80 80 o 7 a 74 S 60 60 S 50 8 5 50 5 40 30 I 20 2 10 4 20 4 0 10 4 PO P LELLO LL 0 Tmax 5 gt amp A Aj g SD SN amp Tmin SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC O Summer Autumn w Winter Spring 100 200 90 OSummer Autumn 180 A G 80 E Winter E Spring 160 n g 140 g 70 5 120 J g 604 ai g 100 8 i a 60 30 40 20 20 10 Rain SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC pE SWA NUL SEA SMD SEC TAS 100 100 90 GSummer E Autumn 90 OSummer Autumn 80 4 mWinter Spring 80 MWin
16. cycle when choosing an analogue This result is particularly marked for temperature 95 of all SDMs for Tmax and 90 for Tmin and less for rainfall 65 as expected since the annual cycle is more pronounced for temperature than for rainfall Fig A5 right Monthly anomalies per seasons Monthly anomalies per predictands 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 4 i 0 T T T 0 L T T T T T DJF MAM JJA SON Tmax Tmin Rain pE dTmax dTmin Figure A5 Percentage of cases for which monthly anomalies are chosen instead of anomalies based on seasonal means Finally the calendar window dTcal was tested Fig A6 This is the number of days before and after the calendar dates which are used to find an analogue For example when searching for an analogue for the 1 of July 1 if dTcal 15 then analogues can only be chosen between the 16 of June and the 16 of July 2 if dTcal 30 analogues are chosen between the 1 of June and the 31 of July and 54 BoM SDM GUI documentation Authors B Timbal et al 3 if dTcal 60 analogues are chosen between the 1 of June this remains unchanged as analogues are only searched for within the same calendar season and the 30 of August Calendar window MB 15days M30days O 60 days 50 40 30 DJF MAM JJA SON Season Fi
17. data rWw r r WWW other 85407 Mar 24 23 19 intervar_tmax_l ps rw r r 1 www other 98167 Mar 24 23 19 m_pf_tmaxl ps rw Yr r 1 www other 70737 Mar 24 23 19 m_tmax_spell_cdrl ps rw r r WWW other 63885 Mar 24 23 19 m tmax_spell_csdl ps sEW r r 1 www other 71556 Mar 24 23 19 m_tmax_spell_hdrl ps rw r r 1 www other 57648 Mar 24 23 19 m_tmax_spell_hsdl ps me A e A WWW other 184164 Mar 24 23 19 marree_tmax_l data Cwr r 1 www other 10144 Mar 24 23 19 results_stm Ww r r 1 www other 184164 Mar 24 23 19 tarcoola_tmax_l data tut WWW other 184164 Mar 24 23 19 woomera_tmax_l data rw r r 1 www other 33 Mar 24 23 19 xoption conf rW r r 1 www other 39 Mar 24 23 19 xstn conf xoption conf and xstn conf are two additional files generated by the GUI to store the options and chosen stations The base directory user directories and run directories currently are granted read write and execute privilege for user group and others as GUI users are www 71 BoM SDM GUI documentation Authors B Timbal et al web server and belong to the others group This level of permissions for other is needed in order to save the files 72 BoM SDM GUI documentation Authors B Timbal et al Appendix E The Statistical Downscaling Model SDM The SDM has been designed to be completed in two parts a computing intensive part which deals with the treatments of the predictors up to the selection of the an
18. functions of maximum temperature in summer in Mildura Victoria located in SMD for the end of the 20 century top left and for the middle of the 21 century using emissions scenario BI top right Similar plots show the scenario A2 for the middle bottom left and the end of the 21 century bottom right Observations are shown as a thick black line and the downscaling of each of the 10 individual climate models from the CMIP3 database are shown as red lines Mean variance and maximum and minimum of the series are shown on the graphs as the top text line followed by the range of the same statistics across the 10 downscaled models 83 BoM SDM GUI documentation Authors B Timbal et al The downscaling of all climate models provides a good estimate of the local PDF for Tmax in summer in Mildura Fig G1 top left The mean and variance from the downscaling of the GCMs encompass the observed values observations statistics are shown on the top line in each diagram while the model range is given in the following two lines It is worth noting that the particular shape of the PDF with a broad maximum around 28 C and a secondary maximum around 35 C is also captured The transformation of the PDF under future projections is shown for 2050 for the B1 Fig G1 top right and A2 scenarios Fig G1 bottom left and for 2100 for the A2 scenario Fig G1 bottom right in this order the graphs look at the details warming as a function of
19. i Observed trends Ee Rain pE Observed trends 2 0 6 0 y 0 85x H nai y 0 34x r 0 76 1 54 y r 0 43 TRO 3 h A 3 J e e 5 1 0 ter te 204 o 3 0 5 3 3 e r r 5 2 prey E 0 l i 241 0 40 2 0 Za 104 00 1 0 20 1 5 5 fo o5 0 15 20 25 Sella A 2 0 6 0 54 o 3 e c aa i 4 0 J 0 14 De 19 y 0 46x Ba gp 4 r 0 75 owe DTmax Observed trends DTmin Observed trends Figure B6 Scatter plot of the reconstructed versus observed linear trends fitted on the length of the longest possible record for the six predictands dots as well as for the normalised reconstructed trends signs For each series there is one point per season and per regional average for all existing stations in that region The total number of points per series is 24 for dTmin dTmax and pE six regions times four seasons and 40 for Tmax Tmin and Rain ten regions times four seasons The line of best fit is shown by the full dashed line and its slope as well as the correlation between the two variables in the bottom right top left corner for the original normalised reconstructed series However since reconstructed series reproduced only part of the inter annual range it is expected that it would affect the linear trend fitted on the reconstructed series The impact of this underestimation of the inter annual variance can be shown by plotting the same graph for normalised reconstructe
20. inter annual and long term see Appendix B for details The technique was found to reproduce the main characteristics of the PDFs of local observed variables In particular the technique was found to be unbiased and it was able to reproduce the local means However like most statistical downscaling techniques the analogue approach did not fully reproduce the observed variances of the series That limitation occurred across all variables with a varying degree of severity This drawback was found to be particularly important for rainfall due to the non normal distribution of daily rainfall the reduction of variance leads to an underestimation of the mean Timbal et al 2006 It was therefore decided to introduce a correction factor for rainfall applied to the reconstructed series and based on the local observed climate It was done so as to reduce the underestimation of the variance and reproduce an unbiased mean This correction was kept to a very simple level and is unchanged across all rainfall stations and seasons in order to minimise the risk of over fitting the SDM to the current climate It was shown that the technique overall was skilful at reproducing the observed PDFs for the right reasons The daily variability is well reproduced as captured by the day to day correlation between the observed and reconstructed series In addition the technique has skill in capturing inter annual variability as well as long term observed climatic tr
21. of a statistical downscaling method is not to correct a poorly performing climate model but to translate large scale circulation simulated by the climate model to the local scale relevant for an impacts study Therefore the ability of the climate model to represent realistically the large scale circulation is relied upon and any inaccuracies in the climate model will affect the downscaled predictand series and the appropriateness of the local series for impact studies has to be evaluated This step is also required to deal with the mean biases from climate model simulations During the application of the SDM to simulations of the current climate model biases mean climatology and daily variability are removed when predictors are normalised The mean climatology of the control simulation which includes the climate model drift is kept as well as the standard deviation When the SDM is applied to simulations of future climate the normalisation of the predictors is made using means and standard deviations calculated from the simulations of the 20 century rather then the future scenarios This two stages approach is needed to factor into the daily anomalies the climate change signal i e the difference between the climatology of the simulation of the 20 century versus the future projections and to remove the model drift i e the mean value for the 20 century simulation This means that when the SDM is applied to future projectio
22. of the 21 experiments for both the A2 and B1 scenarios 29 BoM SDM GUI documentation Authors B Timbal et al Part B The Graphical User Interface 1 The software structure f Program flow chart Static HTML pages GUI main Page Manage the header Link to documentation Initialise the GUI Recdbaccpportunity Main Page f j f Dynamically generated Manage the user outputs Display results Remove past runs Download outputs Manage user s runs Manage the display of stations Run the downscaling model Display graph SDM Script running IDL and generating the outputs Fai Figure 3 Architecture of the GUI for the Bureau of Meteorology SDM 26 BoM SDM GUI documentation Authors B Timbal et al The GUI interface is constructed with four static HTML pages using JavaScript Fig 3 Additional pages and output datasets are generated dynamically using Perl in response to user inputs The last component of the GUI is an Interactive Data Language IDL script which starts the calculation of the SDM itself The entire SDM data inputs statistical calculation and visual outputs is written in IDL The SDM code can be run interactively in research mode however no access to the SDM code is given to the user through the GUI g User management In order to run the SDM GUI a temporary directory structure is created in the BOM computer system to store the results from the GUI Al
23. pm This module uses two languages to generate the GUI HTML to build the text fields drop down menus and buttons and JavaScript to generate the Google map using Google map API and allow the communication of Google map with the HTML forms This module gets called whenever the GUI page is refreshed All available stations are hard coded in this module to be displayed on the Google map Google map API requires a signed up key The key is host dependent and currently is for host gale ho bom gov au If the web interface is to be run on a different platform a new key will have to be applied for and placed in this module Details on how to get the key for Google map API can be found at http www google com apis maps signup html 67 BoM SDM GUI documentation Authors B Timbal et al x_run_gstam pl This program is executed once the button Run Downscaling is activated by the user It specifies the run directory to be used gathers all the options chosen by the user with the GUI and writes these in two files xoption conf and xstn conf stored in the run directory xoption conf contains all the options relevant to the choice of individual SDMs and hence the corresponding Change Of Date COD files for this particular run xstn conf contains the selected stations These two files are accessed by the initial module of the SDM called by xidldown_m in IDL language x_makeResultsTable pm This module generates an HTML table to show
24. some of the outputs of the SDM code 78 BoM SDM GUI documentation Authors B Timbal et al Appendix F The graphical outputs Upon completion of the downscaling calculation the user can access several graphical outputs depending on the predictand downscaled In all cases an inter annual variability plot is generated as well as a probability density function PDF plot for temperature variables or a cumulative distribution function CDF plot for rainfall In addition some figures are available to describe the probability of extreme duration of particular spells On each of these plots the thick black curve represents the observations and the red curve s the downscaled series one per model When results are plotted for the 20 century 1960 1999 the observations are the true observations for that 40 years period However for future projections 20 years period either in the middle or at the end of the 21 century the black curve is constructed using past observations for a reference 20 years period 1975 to 1994 Some statistics are provided in each graph for both the observations and the range of future projections The range is expressed across the selected GCMs i e highest and lowest value are identical if only one GCM is used see figure F3 for an illustration and will encompass a broader range of uncertainties as more GCMs are used When several statistics are shown on a plot highest and lowest values are in
25. state of the art in climate modelling with advances in sophistication in the physical parameterisations used a greater number of components included and increased resolution in both ocean and atmosphere simulations Model outputs were obtained from the IPCC Model Output website at http www pcmdi lInl gov ipcc info_for_analysts php Up to 23 GCMSs contributed to the CMIP3 dataset however only a subset of this database could be used because the SDM relies on daily outputs for the predictors Therefore GCMSs are used only if the modelling group provided daily data for both the simulation of the 20 century and for simulations of the 21 century under different emissions scenarios The models were ranked according to a measure of their sensitivity AT last column in Table 3 calculated using the global warming produced by the model under the AlB scenario when approximated by linear regression over the 21 century BOM and CSIRO 2007 This measure helps the user to select models according to their global sensitivity 23 BoM SDM GUI documentation Authors B Timbal et al de Grid size AT riginating Group Country Acronym km C CSIRO Australia CSIRO 200 2 11 NASA Goddard Institute for Space Studies U S A GISSR 400 2 12 Canadian Climate Centre Canada CCM 300 2 47 Meteorological Research Institute Japan MRI 300 2 52 Geophysical Fluid Dynamics Lab U S A GFDL2 300 2 53 Meteo France
26. the observed and reconstructed series is shown solid line as well as the slope of the relationship and the correlation between the two variables indicated in the bottom right corner of each graph In general the reconstructed series exhibit trends of a magnitude strongly related to the observed trends the correlations vary between 0 87 for rainfall and 0 46 for dT min suggesting that the SDMs have skill in reproducing the observed trends However the slope of these relationships is constantly less than one between 0 14 and 0 57 pointing to an underestimation of the observed trend which is of concern in regard to the ability of the SDM to reproduce the magnitude of the future changes 63 BoM SDM GUI documentation Authors B Timbal et al 40 y 0 97x r 0 73 3 0 i ni a 1 a Fe 3 5 Cc g pr 5 5 3 O 2 a 5 peo 9 9 tene a rs 2 05 gelo 5 10 15 Bo 25 30 3 ag y 0 57x 10 r 0 72 1 5 Tmax Observed trends Tmin Observed trends 3 0 3 0 0 78 Toer y 0 88x 204 a a r 0 69 i Eur 5 0 104 74 c g t e S i o g 4 0 3 20 9 5 P0 05 bo 0 5 1 0 1 5 2 0 2 5 3 2 ei 10 2 o e Pea E 3 fe E E 2 2 0 4 S 2 8 S 3 0 oO o tc 3 i A oc 4 0 b 4 dai aq y 0 27x 26 on r 0 73
27. the downscaling options and creates a link to the downscaling results It will open the resultsTable txt file in the user directory to get the details of the user s chosen options displayed in the table and to indicate where the downscaling results are located x_showResult pl When the Show Results button in the downscaling results table is clicked x_showResults pl is executed It generates a page in a new window that gives the downscaling results as a list of figures and data x_showFigure pl When a figure in the list is highlighted and the Show Figure button is clicked x_showFigure pl is executed which will take the diagram in PS format and convert it into a GIF image to be displayed in the browser 68 BoM SDM GUI documentation Authors B Timbal et al x_showData pl When a data file in the list is highlighted and the Show Data button is clicked x_showData pl is executed which will open the data file and display it in the browser x_download pl When the Download link is clicked all figures are converted to high resolution PNG and zipped The original PS file and data files also get zipped A new page is generated to show the link of those zipped files which can be downloaded x_deleteRun pl If the Delete this run button in the downscaling results table is clicked the results of this run will be deleted and the entry of this run in the resultsTable txt of this user is set to 1 which means that this run dire
28. the line described earlier as the northern boundary of SWA region in the north by a line from Port Headland to Yulara in NT and in the west by the WA state border about 129 E and The Northern monsoon region NMR limited in the south by the northern boundary of NWA region and by the NT state border about 26 S and in the west by the 136 E meridian The Mid East Coast MEC the coastal band east of a line from Mackay in QLD to Port Macquarie in NSW and The Queensland state QLD excluding its southeast corner which is included in the MEC region BoM SDM GUI documentation Authors B Timbal et al The number of surface predictands available for each climatic region is summarised in Table 1 Predictands TAS SWA NUL SEA SMD SEC MEC NWA NMR QLD Temperature 7 9 5 11 15 13 11 6 7 17 Rainfall 8 34 12 31 24 9 12 5 2 12 Pan evap 3 7 2 5 9 7 5 4 7 9 Dew point 1 3 3 3 2 7 4 4 4 3 Table 1 Number of stations for every predictand and in each climatic region and the sum of these stations in the ten regions represented as a percentage of the corresponding entire high quality network last column 10 BoM SDM GUI documentation Authors B Timbal et al 4 The development of the SDMs used in the GUI a General principles guiding the optimisation of the SDMs The development and validation of the SDM was performed using the highest quality data available for both surface predictands and large scale pr
29. the magnitude pf the forcings and the global warming It is worth noting that the observed PDFs are not strictly identical nor are the statistics for the observations in the case of the 20 century and for future projections This is due to the different baselines 1960 to 1999 for the 20 century and 1970 to 1989 for future projections One of the criticisms often made about downscaled climate change projections is that it is not possible to validate the underlying assumption that the statistical linkage remains valid in a warmer world Hewitson and Crane 2006 This point was partially answered during the model evaluation stage by showing that the SDMs were able to reproduce the observed trends In this case study it is possible to compare the point specific projections with direct model projections coming from the CMIP3 models for grid boxes surrounding Mildura documented in the Climate Change in Australia report BoM and CSIRO 2007 and available online at http www climatechangeinaustralia gov au Although this comparison is not a perfect match because numbers are coming from very different methodologies it provides an interesting overview of the consistency among these different projections For 2050 direct model projections suggest a range for high emissions scenarios between 1 and 1 5 C at the 10 percentile 2 and 2 1 C at the 50 percentile and 2 5 to 3 C at the 90 percentile and for low emissions scenarios between
30. 18th BMRC Modelling Workshop pp 123 126 24 Timbal B 2006b Statistical Downscaling in BMRC In Climate change research in the Bureau of Meteorology S B Power and K Pearce eds BMRC Research Report 125 68 72 25 Timbal B and J Arblaster 2006 Land covers change as an additional forcing to explain the rainfall decline in the South West of Australia Geo Res Let 33 L07717 doi 10 1029 2005GL025361 46 BoM SDM GUI documentation Authors B Timbal et al 26 Timbal B J Arblaster and S Power 2006 Attribution of the late 20th century rainfall decline in South West Australia J Climate 19 10 2046 2062 27 Timbal B and D Jones 2008 Future projections of winter rainfall in southeast Australia using a statistical downscaling technique Clim Change 86 165 187 28 Timbal B E Fernandez and Z Li 2008a Generalization of a statistical downscaling model to provide local climate change projections for Australia Environ Soft Model doi 10 1016 j envsoft 2008 07 007 29 Timbal B P Hope and S Charles 2008b Evaluating the consistency between statistically downscaled and global dynamical model climate change projections J Climate 21 22 6052 6059 30 Trewin B C 2001 Extreme temperature events in Australia PhD Thesis School of Earth Sciences University of Melbourne Australia 31 Uppala S P Kallberg A Simmons U Andrae V da Costa Bechtold M Fiorino J Gibson J Hase
31. 2046 05012046 06012046 07012046 08012046 09012046 10012046 11012046 12012046 13012046 14012046 15012046 16012046 17012046 18012046 19012046 20012046 21012046 22012046 23012046 24012046 25012046 26012046 27012046 28012046 29012046 30012046 31012046 01022046 02022046 03022046 04022046 05022046 06022046 07022046 CSIRO MIROC 31 30 37 30 38 50 39 40 40 00 34 80 29 90 38 80 28 70 30 80 34 20 30 10 36 30 29 40 35 10 28 00 35 30 33 70 36 60 40 40 35 90 26 10 33 10 31 60 35 20 30 90 34 90 37 60 43 40 37 60 43 20 40 20 43 20 31 60 43 50 35 00 35 70 41 10 34 50 39 80 38 80 38 80 46 60 25 40 35 70 27 60 41 20 30 30 46 60 37 10 39 40 28 90 30 20 28 60 26 30 32 60 25 90 35 90 32 50 28 30 30 30 26 40 26 20 31 20 35 40 36 10 32 60 37 40 30 00 38 00 30 20 41 20 30 00 40 10 31 90 35 10 MPI 28 30 28 50 30 70 Figure 9 ones Observation 22 90 17 20 22 90 24 50 24 40 28 30 32 30 36 50 41 20 33 80 29 10 26 50 23 80 Tmax data for Mildura the first column displays the dates the subsequent one column per model show the projected downscaled value s and the last one displays the actual or equivalent observations for that day 36 BoM SDM GUI documentation Authors B Timbal et al For a user to display a graph selected from the left box using the Show Figure button beneath the figure list a package able to display postscript files would have to be available on the us
32. BoM SDM GUI documentation Authors B Timbal et al Scenario Mean warming Uncertainty range B1 1 8 C 1 1 2 9 C AIT 2 4 C 1 4 3 8 C B2 2 4 C 1 4 3 8 C A1B 28 C 1 7 4 4 C A2 3 4 C 2 0 5 4 C AIFI 4 0 C 2 4 6 4 C Table 4 Mean global warming estimates for 2100 relative to 1990 for 6 emissions scenarios derived from Figure SPM 3 of the IPCC 2007 report The A2 scenario is based on a very heterogeneous world with continuously increasing population and technologically fragmented economic development leading to one of the highest emissions scenarios available In contrast in the B1 scenario the emphasis is on global solutions to economic social and environmental sustainability and B1 is one of the lowest emissions scenarios available A simulation of the 20 century was also used to complement the scenarios and provide a benchmark on the reliability of the downscaled GCMs Three experiments were therefore used for each GCM a simulation of the 20 century hereafter 20C3M and two future projections for the 21 century using the two emissions scenarios mentioned above The data available from the GCMs cover three time slices for which daily DMOs were available and used by the GUI e 40 years from 1961 to 2000 for the 20C3M experiment e 20 years from 2046 to 2065 in the middle of the 21 experiments for both the A2 and B1 scenarios and e 20 years from 2081 to 2100 at the end
33. Data powered by Google It provides an option to choose the background between physical maps satellite pictures or a combination of both using the buttons displayed at the top right The user has a series of four steps to follow in order to carry out the downscaling process Step 1 Step 2 Choose the predictand and season Select a predictand from the Predictand dropdown list Rainfall Tmax Tmin PE ATmax and ATmin Note the extension of the GUI to pE dTmax and dT nin is currently underway Note it is possible to jump to step 3 following this selection and come down to the other steps later Select one of the calendar seasons from the Season dropdown list Summer December January February Autumn March April May Winter June July August and Spring September October November Note it is not possible at this stage to select several seasons i e all year in one go Choose the model s and a scenario Note it is not possible to combine the choice of model and the choice of scenarios i e all simulations selected are based on a single choice of scenario Select the climate scenario from the Scenario dropdown list Available scenarios are a simulation of the 20 Century 20C3M providing daily data from 1961 to 2000 and future emissions scenarios A2 and BI for the 21 century providing daily data for 20 year time slices 2046 2065 for A2_50 and B1_50 and 2081 2100 for A2_100 and B1_100 Note it is not
34. France CNRM 200 2 81 Geophysical Fluid Dynamics Lab U S A GFDLI 300 2 98 Institut Pierre Simon Laplace France IPSL 300 3 19 Centre for Climate Research Japan MIROC 300 3 35 Max Planck Institute for meteorology DKRZ Germany MPI 200 3 69 Table 3 Global climate models from the CMIP3 database to which the SDM was applied The name and country of the originating group the acronym used in the GUI and the approximate size of the model horizontal grid box are shown Models are ranked according to a sensitivity measure last column from the least sensitive at the top to the most sensitive at the bottom Future greenhouse gas and aerosol emissions due to human activities such as energy generation transport agriculture land clearing industrial processes and waste are considered in the future scenarios To provide a basis for estimating future climate change the IPCC prepared a large range of greenhouse gas and sulphate aerosol emissions scenarios for the 21 century that combine a variety of assumptions about demographic economic and technological driving forces likely to influence future emissions Up to six future emissions scenarios were used by the modelling groups that contributed to the CMIP3 database Each emissions scenario leads to a different projected global warming range Table 4 In order to reduce the amount of data used only two scenarios for the 21 century have been downscaled A2 and BI 24
35. Range but at low elevation Cabramurra inland at high elevation and Wagga Wagga on the western side of the Great Dividing Range using the statistical downscaling technique described in the text Observations are shown as thick black lines the 10 individual climate models from the IPCC ARA database are shown as red lines It is also possible to estimate the chances that the warming will exceed fixed thresholds such as 1 5 C This probability varies markedly across the grid box as expected from the warming range projected at each location it is less than 5 in Jervis Bay but reaches 80 in Canberra and Wagga Wagga and it exceeds 95 in Cabramurra 88 BoM SDM GUI documentation Authors B Timbal et al b For rainfall Sub grid scale heterogeneity can also be particularly important for rainfall which is spatially highly variable This result was illustrated for the southwest of Western Australia which has suffered one of the most severe rainfall declines in recent history Ryan and Hope 2005 2006 Future projections based on the A2 scenario for winter June to August rainfall Fig H2 indicate a further rainfall reduction of the total rainfall TR on the graphs alongside a reduction of the average rainfall intensity AI a reduction of the rainfall intensity for the middle of the distribution R50 and an increase of the percentage of dry days D There is a high level of consistency among models They all suggest a decl
36. Vgso MSLP amp Q935 amp Tmin SEA Winter Timax amp Roos MSLP amp Qo25 amp Tmin MSLP amp Q925 amp Tin Spring Timax amp Roos MSLP amp Qo25 amp Tmin MSLP amp Q925 amp Tin Summer Tmax amp Reso MSLP amp Qgs0 amp Tmin amp Vgso MSLP amp Qgs0 amp Tg50 Autumn Tax amp Reso MSLP amp Qgso amp Tmin amp Vgso MSLP amp Qg50 amp Tmin amp Ugso SMD Winter Timax amp Roos Tin MSLP amp Qg50 amp Tmin amp Vsg50 Spring Trax amp Rgso MSLP amp Qg50 MSLP amp Qgs0 amp Tg50 Summer Tax amp Ros MSLP amp Qo25 amp T min MSLP amp Qo25 amp Tmin Autumn Tmax amp Roos MSLP amp Qo25 amp T min MSLP amp Qo25 amp Tmin SEC Winter MSLP amp Tmax amp Roos MSLP amp Q935 amp T min MSLP amp Qo95 amp Tin Spring Timax amp Roos amp Ugso MSLP amp Q935 amp T min MSLP amp Q925 amp Tin Summer MSLP amp Tgs0 amp Roos MSLP amp Qg50 MSLP amp Qg50 Autumn MSLP amp Tmax MSLP amp Qos MSLP amp Qo95 amp Tin TAS Winter MSLP amp Tmax amp Roos Tinin MSLP amp Qgs0 amp Tmin Spring MSLP amp Tmax amp Roos MSLP amp Qo25 amp Tmin MSLP amp Qo25 amp Tmin Table 2b Optimum combination of parameters for each calendar season and the six regions for dTmax dTmin and pE 19 BoM SDM GUI documentation Authors B Timbal et al 5 Evaluation of the SDMs used in the GUI A careful evaluation of the skill of the SDM was carried out on different timescales daily
37. able differences from one part of the grid box to another in Jervis Bay near the coast it is expected to be between 0 5 and 1 0 C while in Canberra the expected warming is more than twice as large between 1 1 and 2 3 C The warming is slightly larger further inland Wagga Wagga and even larger for higher elevation Cabramurra The increased warming away from the coast is a well known and expected feature depicted by GCMs the additional warming with elevation is less documented as it is usually not captured by GCMs which have a limited ability to reproduce the orography due to coarse model resolution The projected warming is broadly similar with direct model outputs for the area between 1 4 and 1 8 C Although direct model outputs provide a hint that coastal warming is expected to be lower than further inland the downscaled results provide a more contrasting response 87 BoM SDM GUI documentation Authors B Timbal et al CANBERRA mean 30 6 D T 1 3 to 2 5 PT Oe ee ee I oN a Oo O O N Gouina iia CABRAMURRA mean 18 3 D T 1 5 to 2 9 Goma IO meon 23 7 DT 10 iS 20 25 30 35 0 Figure H1 Projected changes of the observed probability density functions of maximum temperature in summer for 2050 using emissions scenario A2 in four neighbouring locations with contrasting climates Jervis Bay a coastal location Canberra 150 km inland in the Great Dividing
38. all Stn Remove Stn Step 4 Run Downscaling Process Run Downscaling Clear CSIRO_20C3M GFDL1_20C3M GFDL2_20C3M GISSR_20C3M Figure 6 View of the GUI once a region here Southwest of Western Australia and a predictand here Tmax have been selected Once the Run Downscaling button has been clicked the GUI transfers the user s chosen options and runs the SDM IDL code on the Bureau machine Gale Only a part of the entire SDM code is run on the fly to generate graphics and data output It relies on intermediary files pre calculated for all possible cases available via the GUI These intermediary files are labelled Change Of Date COD files and contain the optimum analogue i e a single best match for any model day chosen within the 1958 to 2003 period in the NNR database except from pan evaporation for which data is restricted to the period 1975 2003 This optimal analogue depends on all the parameters chosen in the GUI the predictand the season the region and the individual synoptic situation produced by the CMIP3 GCM The downscaling process will take between one and two minutes to be completed depending on the number of stations and the number of models being chosen 33 BoM SDM GUI documentation Authors B Timbal et al 3 Results provided by the GUI a Generating several outputs Upon completion of the downscaling calculation a table is generated and displa
39. all the climate change signals Users should also be aware of the important benefit provided by SDMs compared to using Direct Model Outputs DMOs more complete discussions on the advantages can be found in Timbal 2006b e Provide local climate change information e By pass some of the physical parameterizations sued to calculate model predictands which are also subject to fitting to observed data as do SDMs e Provide a cost effective way to describe the cascade of uncertainties relevant to local climate change projections BoM SDM GUI documentation Authors B Timbal et al And finally it is recommended that users of downscaled projections are informed about the best practices in this area E g the IPCC Wilby et al 2004 has established a guideline which is very relevant to users of the GUI described here BoM SDM GUI documentation Authors B Timbal et al 3 The predictands obtained from the GUI For surface predictands the best possible stations available for long term climate purposes are the High Quality HQ dataset assembled by the National Climate Centre NCC from the Bureau of Meteorology Variables used were daily HQ temperature extremes Trewin 2001 and daily HQ rainfall amount Lavery et al 1992 and 1997 Both datasets were extended to 2005 by the NCC More recently HQ datasets were developed for surface humidity dew point daily extremes dTmin and AT max by Lucas 2006 and for surface pan evaporation pE
40. alogues stored in the Change Of Date COD files and a second part to perform calculations for the reconstructed predictands and generate plots comparing reconstructed and observed series Fig El The details on each subroutine performing the function described above are given in Figure E2 and the subsequent text Interactive use of the entire Statistical Downscaling Model Treatment of the predictors read predictors Plot options for predictors fields Several predictors Anomalies and or EOFs 4 Calculated analogue A Change of date User driven use of COD files single COD file i GUI driven interactive use of multiple _m 5 COD files i Predictands calculated series Plot PDFs Plot spells Auto correlation of the series options Rain occurrence Calculate skill Main driver of the SDM Calculate inter annual variability and statistics Figure E1 Flow chart of the structure of the Bureau of Meteorology SDM 13 BoM SDM GUI documentation Authors B Timbal et al Interactive use of the entire Statistical Downscaling Model Treatment of the predictors p_pcplot p_coplot p_corrsta p_avplot p_dmoplot PIZZA PLAIN GUI driven interactive use of multiple _m User driven use of d COD files single COD file Predictands calculated series sp_intervar _m ee NINA Figure E2 Flow chart
41. am ACCSP and the support of the DCC The extension of the technique to moisture variables was supported in part by the ACCSP and in part by the South Eastern Australian Climate Initiative SEACI The predictand databases were obtained from the National Climate Centre NCC of the Bureau of Meteorology BoM NCEP NCAR reanalyses were accessed from the Physical Sciences Division PSD of the Earth System Research Laboratory ESRL USA and processed by M Collier CSIRO and P Hope BoM We acknowledge the international modelling groups for providing their data for analysis the Program for Climate Model Diagnosis and Intercomparison PCMDI for collecting and archiving the model data the JSC CLIVAR Working Group on Coupled Modelling WGCM and their Coupled Model Intercomparison Project CMIP and Climate Simulation Panel for organizing the model data analysis activity and the IPCC WGI TSU for technical support The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science U S Department of Energy Thanks are due to Aurel Moise and Lawson Mawson from the BoM for transferring the CMIP3 database to Australia The authors wish to thank P Wiles Bureau of Meteorology NSW regional office and A Roubicek University of Macquarie NSW for their very useful suggestions in improving earlier versions of this document 43 BoM SDM GUI documentation Authors B Timbal et al Bibliography 1 2 3
42. amp Tmax Tsso amp Qsso amp Tmin PRCP Summer Teso amp Tmax Tsg50 amp Q350 PRCP Autumn Ts50 amp Tmax Tsg50 amp Tmin amp Qg50 PRCP amp Ugso amp Tmax NMR Winter Tg50 amp Qsso amp Usso Taso amp Qsso PRCP Spring Tsso Teso amp Qsso amp Tmin PRCP Summer Tss0 amp Usso amp Tmax Tmin MSLP amp Tmax amp R70 Autumn Tss0 amp Usso amp Tmax Tmin PRCP amp Qsso amp Usso amp Vaso MEC Winter MSLP amp Tmax Tmin MSLP amp Reso amp Vsso Spring Tg50 amp Reso Taso amp Qaso Reso amp Tmax amp Ugso Table 2a Optimum combination of parameters for each calendar season and the ten regions for Tmin Tmax and Rainfall 18 BoM SDM GUI documentation Authors B Timbal et al pE ATmax dT nin Summer MSLP amp Tmax amp Roos Qgso amp Tmin Qo25 amp Tin Autumn MSLP amp Tgs0 amp Reso MSLP amp Qo25 amp Tmin Qo25 amp Tmin SWA Winter MSLP amp Tmax amp Roos MSLP amp Qgs0 amp Tmin Qo25 amp Tmin Spring MSLP amp Tmax amp Reso Qss0 amp Tsso Qozs amp Tmin Summer Tg50 amp Ro25 MSLP amp Qo25 amp T min MSLP amp Q925 amp Tmin Autumn Tgso amp Roos MSLP amp 0935 MSLP amp Qo25 NUL Winter Trax amp Roos MSLP amp Qo25 amp T min MSLP amp Q935 amp Tg50 Spring Tg50 amp Roos MSLP amp Q925 amp Tmin Q925 Summer Tmax amp Roos MSLP amp Qos MSLP amp Qo95 amp Tin Autumn Trax amp Roos MSLP amp Q935 amp Tmin amp
43. art arcana aaa 26 oji Escrmandseni ilconte 27 h Individual subr utnes ann a aa a eee 28 2 Usine th GUD ea nap e ra A air 29 a Access UR Dorka na a a a cat canted a E a s aa isa 29 by Front page layout rara 29 c The interactive web interface eee 30 iii BoM SDM GUI documentation Authors B Timbal et al gt Resulis provided by the GUI prese ble e a ees 34 a Generating several OWtpuls arie nta iero 34 b Graphics and Data Display and Download eeeeeseceseeeeeeeeaeeenneees 35 Termination amd Cosmi till lille Lee bia 37 4 Summary and additional information 39 a Summary on the existing tool role arte 39 b Case studies performed with this t00l 40 c Future developments of the 001 rear 40 d Other existing software complementary to this t001 e 42 Acknowledsemenis lariana tri lr eg risna isis 43 Biblostaphys hatbleanisoler ida a a a aee ie iii ai 44 Appendix A Analysis of the SDM optimised parameters i 49 Appendix B Analysis of the SDM skill ii 56 Appendix Perl scapito Re eek ota geet 66 Appendix D Directory structure 2 4 lied aie ccs es SG eee 70 Appendix E The Statistical Downscaling Model SDM i 13 Appendix F The sraphicalvutputs icaro aa 79 Appendix G Case study 1 projected Tmax in Mildura eeeeeeseeceeceeeeeees 83 Appendix H Case stu
44. ase and period of interest are both the entire 1958 to 2005 period but analogues are not allowed to come from the same calendar year or consecutive December January and February in the case of summer This provision is to ensure that no artificial skill is introduced in the SDM i e due to seasonal anomalies not captured by atmospheric large scale forcings such as soil moisture This decision was based on previous developments of the SDM using fully cross validated methodology Timbal and McAvaney 2001 Timbal et al 2003 Timbal 2004 14 BoM SDM GUI documentation Authors B Timbal et al c Change of Date files Whenever the SDM is run the main output is the COD file where the optimal analogue for every meteorological situation for the period of interest is kept COD files contain as many lines as there are days in the period of interest and three columns the first one contains the date for which analogues were search for the second column contains the date of the best matching analogue and the third column contains the Euclidean distance providing a measure of the closeness of the best matching analogue From this COD file local predictands are calculated by using the observed values in the desired location for the date given as the best matching analogue step 3 in Fig 2 A consequence of this methodology is that once the SDM has been optimised for a particular season region variable using HQ networks to ensure that da
45. be able to provide meaningful downscaled projections for dTmin The optimised SDMs were then applied to the CMIP3 climate models described in the next section It is worth noting that the search for analogues is performed on normalised anomalies of the predictor fields In the case of climate model simulations normalised anomalies are calculated for the simulation of the 20 century in effect removing model biases in the mean and variance reproduction of predictor fields The model mean climatology is then used as a baseline to calculate anomalies for the predictors in future projections This methodology assumes that the same model biases are affecting future projections and is commonly used in climate change science Implicitly this methodology corrects model biases both in the mean state and in daily variability therefore the traditional evaluation of GCM performance in reproducing surface predictands as used in the Climate Change in Australia report BoM and CSIRO 2007 is not relevant here However all users should carefully evaluate the suitability of downscaled models for any impact study using the local predictands of interest generated by downscaling GCM simulations of the 20 century 20C3M It is worth remembering that the selection of the optimum combination of predictors is solely based on the observed past climate The SDMs then rely on the hypothesis that the chosen predictors will capture the essence of the large scale chang
46. ccurring Analogues J Atmos Sci 26 636 646 16 Lucas C 2006 A high quality humidity database for Australia in Proceedings of the 17 Australia New Zealand Climate Forum p p35 Canberra Australia 45 BoM SDM GUI documentation Authors B Timbal et al 17 Ricketts J and C Page 2007 A web based version of OzClim for exploring climate change impacts and risks in the Australian region in MODSIMO07 proceedings pp 560 566 18 Robertson A S Kirshner and P Smyth 2004 Downscaling of daily rainfall occurrence over Northeast Brazil using a hidden Markov model J Climate 17 4407 4424 19 Solymosi N A Kern A Maroti Agots L Horvath and K Erdelyi 2008 TETYN An easy to use tool for extracting climatic parameters from Tyndall data sets Env Model Software 23 7 948 949 20 Timbal B and B J McAvaney 2001 An Analogue based method to downscale surface air temperature Application for Australia Clim Dyn 17 947 963 21 Timbal B A Dufour and B J McAvaney 2003 An estimate of climate change for Western France using a statistical downscaling technique Clim Dyn 17 947 963 22 Timbal B 2004 South West Australia past and future rainfall trends Clim Res 26 3 233 249 23 Timbal B 2006a Statistical Downscaling an important part of ACCESS In The Australian Community Climate and Earth System Simulator ACCESS challenge and opportunities BMRC Research Report 123
47. ctory is ready for storing results from another downscaling run x_deleteUser pl Need to be created run to do the final clean up exit the GUI and remove all stored data from the user after checking that it is OK to be added to the flow chart 69 BoM SDM GUI documentation Authors B Timbal et al Appendix D Directory structure The base directory on Gale is bm ghome bxt public_html cgi bin gdownGUlresults and user directories are sub directories Due to the IDL licence limitation up to three users are allowed to run the downscaling model at the same time Therefore the user directory structure is as follows gale bm ghome bxt public_html cgi bin gdownGUIresults IWXrFWXIWX 1 bxt bmrc 129 Dec 5 18 24 user txt drwxrwxrwx 7 bxt bmrc 1024 Nov 21 17 36 userl drwxrwxrwx 7 bxt bmrc 1024 Nov 21 17 36 user2 Arwxrwxrwx 7 bxt bmrc 1024 Nov 21 17 36 user3 user txt contains limited information on the user i e at what time the user started to use the GUI This time is used to determine if this user upon returning to the GUI after a while should use an existing directory or get a new one Currently after 24hrs a user directory and all its contents are deleted to make space for other users Each user directory has five run sub directories allowing users to make 5 downscaling runs and store their outputs Once 5 runs are performed the user needs to delete old simulations in order to perform new ones by clicking on th
48. curate On each graph each point corresponds to a single location for a single season with the observed mean value on the x axis and the reconstructed mean along the y axis The number of points in each graph is equal to the total number of stations captured in one of the ten climate regions only six regions for dTmin dTmax and pE times four between 1012 for rainfall and 76 for dew point temperature A perfect match would see all the points aligned with the diagonal For most predictands errors in the reproduction of the mean are relatively small and there is no clear evidence that the SDMs have a bias toward higher or lower values Furthermore there is no evidence that the SDMs have more difficulties in reproducing observations at either side of the spectrum minimum or maximum values Similarly results are shown for the reproduction of the variance Fig B2 It is clear that the technique has a tendency to underestimate the observed variance points are aligned below the diagonal for all variables This is a known problem of statistical downscaling techniques van Storch 1999 and although the analogue approach is less affected than many other techniques in particular linear techniques it remains an issue 56 BoM SDM GUI documentation Authors B Timbal et al 40 25 ss Mean Tmax Pa Mean Tmin 30 4 n
49. d series where the linear trend 64 BoM SDM GUI documentation Authors B Timbal et al is divided by the ratio of reproduced inter annual variability crosses in Fig B6 Although the correlations between series shown at the top left corners of the graphs in Fig B6 remain by and large unchanged the slope of the relationship in most cases is much closer to one which is a major improvement with the notable exception of AT min This latest finding enhances confidence in the model s ability to reproduce future climate changes at the local scale as it is able to do so for the current climate changes The result for dTmin however casts doubt on the validity of future projections for this variable All these findings should be considered by user when making use of projections data for impact studies Note The trend plot for Rainfall Fig B6 has been cropped due to the presence of two outliers in region NMR in summer slope 13 60 and autumn slope 3 45 These outliers are taken into account when computing the correlations and slopes 65 BoM SDM GUI documentation Authors B Timbal et al Appendix C Perl scripts Static HTML pages GUI main Page Initialise the GUI Main Page Manage the header Link to documentation Feedback opportunity Organise user directories To Dynamically generated HTML page Manage the user outputs Manage User s runs Ma
50. d the latitude longitude and height of the station By clicking on the yellow marker that station will be added to the text area next to the Selected Region field At the same time the yellow marker for this station will be removed from the map To select all stations click the Select all Stn button To remove a selected station highlight the station multiple highlights allowed and click the Remove Stn button Removed stations will re appear on the map as yellow markers In some cases a region is already selected but needs to be changed to another region To make this change click on the square shaped button next to the down arrow button on the map This will lead back to the original display where the 6 regions are shown on the map Step 4 Run the downscaling process Start the downscaling process by clicking on the button Run Downscaling 32 BoM SDM GUI documentation Authors B Timbal et al e Clicking on the Clear button will remove all selected items and lead back to the start of the interface A Downscaling Technique Home Others Step 1 Choose Predictand and Season Me _Map__ Satellite _Hybria_ Predictand Tmax z Season Summer n A Step 2 Choose Model and i a ep ose ecnario Pe Da Scenario 20th Cent Model All Model A A CCM_20C3M nae Select M S CNRM_20C3M Delete MS zl Step 3 Choose Region and stations from the map Selected Region WA Select
51. dependent from one another i e the highest values are not necessarily from the same GCM In the case of temperature a PDF is plotted Fig F1 It represents the distribution of the observations and the downscaled results Some statistics are available on the figure for the observations these are the mean Mean variance Var minimum Mn and maximum Mx values Below the same statistics are repeated for the downscaled results M M is the highest lowest mean among the models used V V is the highest lowest variance Mn Mn is the highest lowest minimum and Mx Mx is the highest lowest maximum 79 BoM SDM GUI documentation Authors B Timbal et al MILDURA PDF of TMAX L Obs Mean 31 5 Var 27 3 Mn 16 2 Mx 46 8 J 10 LGCM M 33 4 V 24 8 Mn 20 9 Mx 46 9 4 C M 33 1 V 23 0 Mn 20 5 Mx 46 6 J BE J 6 F J 4 2 J ol a 10 20 30 40 Degree Celsius Figure F1 Probability density functions of maximum temperature in summer in Mildura region SMD for the middle of the 21 century 2046 to 2065 using emission scenario A2 for CSIRO CCM and CNRM models For rainfall a CDF rather than a PDF is plotted Fig F2 Some statistics are provided in the graph for the observations first line and for the range of future projections the percentage of dry days D the average rainfall intensity AI and the median rainfall R50 for rainy days and the total rainfall TR A
52. dy 2 sub grid heterogeneity i 87 a For temperature ail aa 87 b Forrainfall sirsa a e Andar 89 iv BoM SDM GUI documentation Authors B Timbal et al Abstract Climate change information required for impact studies is of a much finer spatial scale than climate models can directly provide Statistical downscaling models SDMs are commonly used to fill this scale gap SDMs are based on the view that the regional climate is conditioned by two factors 1 the large scale climatic state and 2 local physiographic features A SDM based on an analogue approach has been developed within the Australian Bureau of Meteorology and applied to ten regions covering the entire Australian continent Six surface predictands daily minimum and maximum temperatures dew point minimum and maximum temperatures total rainfall and pan evaporation were modelled The skill of the SDMs was evaluated by comparing reconstructed and observed series using a range of metrics the first two moments of the series and the ability to reproduce day to day variability inter annual variability and long term trends Once optimized the SDMs were applied to a selection of global climate models which contributed to the Intergovernmental Panel on Climate Change 4 assessment report released in 2007 A user friendly graphical interface has been developed to facilitate dissemination of the SDM results and provides a range of options for users to obtain tailored in
53. e appropriate button for any of its five previous simulations gale 47 bm ghome bxt public_html cgi bin gdownGUIresults userl rw rw rw 1 zhl bmrc 35 Mar 24 12 05 resultsTable txt Arwxrwxrwx 2 zhl bmrc 2048 Mar 24 12 05 runl drwxrwxrwx 2 zhl bmrc 2048 Mar 24 11 38 run2 Arwxrwxrwx 2 zhl bmrc 2048 Mar 20 14 35 run3 drwxrwxrwx 2 zhl bmrc 2048 Mar 20 14 35 run4 drwxrwxrwx 2 zhl bmrc 2048 Mar 20 14 35 rund resultsTable txt summarises information about the runs contained in each sub directory 70 BoM SDM GUI documentation Authors B Timbal et al NUL tmax 0 CCM_20C3M CSIRO_20C3M Forrest Woomera Tarcoola Marree Ceduna e re HE HO HW a _S WU N E fan The first column is the user number 1 in this example the second column contains the run number the third column shows all the options relevant to this run region scenario stations if the run has been completed otherwise it just displays l All downscaling results i e figures text data and configuration files are kept in individual run directories and their contents look like this gale 47 bm ghome bxt public_html cgi bin gdownGUIresults userl runl twet lt r 1 www other 184164 Mar 24 23 19 ceduna_tmax_l data Ew E E www other 1000 Mar 24 23 19 excel_inter_tmax_1 rw r r WWW other 50 Mar 24 23 19 excel_temp_meanl CW r 6 1 www other 150 Mar 24 23 19 excel_trend_tmax_1 rw r r 1 www other 184164 Mar 24 23 19 forrest_tmax l
54. echnical advice To the extent permitted by law CSIRO and the Bureau of Meteorology including each of its employees and consultants excludes all liability to any person for any consequences including but not limited to all losses damages costs expenses and any other compensation arising directly or indirectly from using this publication in part or in whole and any information or material contained in it il BoM SDM GUI documentation Authors B Timbal et al Table of Contents LIGA TI 1 Part Az Scientific backeroundissa ratori rica 3 i Introduction On 8 AeA OSes ee fan sacre pride colo ei im 3 2 The Bureau of Meteorology Statistical Downscaling Model 4 3 The predictands obtained from the GUI 7 4 The development of the SDMs used in the GUI in 11 a General principles guiding the optimisation of the SDMS 11 b Defining daily meteorological situations i 13 e Change of Date files ea 15 d SDM optimisation versus application to climate model simulations 15 e Characteristics of the optimised SDMS 17 5 Evaluation of the SDMs used in the GUI eee 20 6 The Global Climate Models used in the GUI ii 23 Part B The Graphical User Interface aaa ia 26 H The soft wate Structure ce 20 acct ced avec tei de Boies te aaa 26 i Program TOW ch
55. edictors as it is important to ensure that data quality does not impact the statistical linkage being developed Global re analyses of the atmosphere were used for the large scale predictors Both the NCEP NCAR re analysis datasets NNR available from 1948 to present Kalnay et al 1996 and the European Centre ECMWF ERA40 re analysis datasets available from 1957 to 2003 Uppala et al 2005 were used to test the SDM The impact of data quality on the SDM performance was carefully assessed It was found that NNR from 1958 to 2005 i e withdrawing the earlier decade from 1948 to 1957 provided the best results Overall the SDM skill was higher when the NNR where used rather than ERA40 suggesting that NNR is more reliable above the Australian continent over the length of the record Therefore NNR from 1958 to 2005 are used as the reference database to search for analogues of the global climate models used in the GUI A total of 192 SDMs were produced six predictands in six regions and three predictands in the four tropical regions one for each four calendar season summer DJF autumn MAM winter JJA and spring SON Each individual SDM was optimised using a range of statistics These covered the ability of the SDM to reproduce mean and variance of the observed Probability Density Function PDF of the series and the skill of the model in reproducing day to day variability inter annual variability both using correlations
56. egions with complex topography coastal or island locations and in areas of highly heterogeneous land cover Therefore a gap exists between what climate models can predict about future climate change and the information relevant for environmental studies Statistical Downscaling Models SDMs are commonly used to fill this gap SDMs are based on the premise that the regional climate is conditioned by two factors the large scale climatic state and local physiographic features From this perspective regional or local climate information is derived by first determining a statistical model which relates large scale climate variables or predictors to regional and local variables or predictands Once this relationship is established the large scale output of GCM simulations can be fed into the statistical model to estimate the corresponding local and regional climate characteristics There have been numerous applications of SDMs both in Australia and elsewhere According to the latest IPCC assessment Research on SDM has shown an extensive growth in application and includes an increased availability of generic tools for the impact community pp 920 Chapter 11 IPCC 4 assessment Christensen et al 2007 In light of this ongoing trend the Bureau of Meteorology Statistical Downscaling Model has been extended to cover most of the Australian continent and in order to facilitate access to downscaling projections across a broade
57. ends For rainfall it is notable that although the analogue approach is least skilful in terms of correlation on both daily and inter annual time scale it gives the most reliable reproduction of long term trends of all predictands This suggests that day to day variability of rainfall amount is largely affected by local effects not captured by large scale variables The large scale variability and long term trends which remove a lot of this small temporal scale variability is well tied to large scale factors 20 BoM SDM GUI documentation Authors B Timbal et al By contrast the technique while extremely skilful in reproducing temperature series on daily and inter annual time scales seems to consistently underestimate by a factor of two the observed long term warming It was found that this underestimation is by and large due to the underestimation of the variance by the reconstructed series leading to an underestimation of the magnitude of the trend For the newly developed moisture variables the analogue technique performed relatively consistently with skill displayed across all statistics within the range of the two extremes provided by temperature and rainfall It was noted however that for dT min the technique does not appear to reproduce much of the observed long term trend thus casting doubt on the ability of large scale predictors to explain local trends for this variable and hence suggesting that the method may not
58. er machine e g the freeware ghostview Otherwise it is possible to download the files and print the figures on a postscript printer To download all figures and data click on the Download link at the bottom of the page A new browser window pops up Fig 10 displaying three downloading options v figures as PostScript files or v figures in PNG format or v ASCII data files Note Since files are relatively small all stations are bundled into one downloadable file Figures and data for downloading Figures in PostScript format Figures in PNG format Data in ASCII format Figure 10 The download page showing available options to download outputs from a single run c Termination and closing Upon completing all the simulations needed the user should hit the Close the GUI button and confirm that choice rather than shut down the browser This action will 37 BoM SDM GUI documentation Authors B Timbal et al delete all the data stored by the user and free space for a new user this is important due to the licensing limitations mentioned earlier 38 BoM SDM GUI documentation Authors B Timbal et al 4 Summary and additional information a Summary on the existing tool The BoM statistical downscaling model developed in earlier studies has been generalised across half of the Australian continent This method provides point specific climate change projections relevant for impact stud
59. es in a 21 BoM SDM GUI documentation Authors B Timbal et al future climate that will drive the local climate This hypothesis has been investigated using discontinuities in our past climate records in specific cases discussed in the introduction It will also be touched on during the evaluation of the skill of the model in particular the ability of the SDMs to reproduce observed trends 22 BoM SDM GUI documentation Authors B Timbal et al 6 The Global Climate Models used in the GUI Fully coupled Atmospheric and Oceanic General Circulation Models GCMs commonly referred to as global climate models are complex numerical models which solve well established laws of physics such as conservation of mass energy and momentum of the components of the climate system They include representations of the continental surface hydrology soil and vegetation and of the cryosphere land ice sea ice and Polar Regions As part of the Intergovernmental Panel on Climate Change IPCC 4 Assessment of Climate Change Science released in 2007 a new set of coupled GCM experiments has been studied This represents a major advance both for the evaluation of models and for the generation of climate projections The open nature of output availability has resulted in this set of experiments being subjected to unprecedented levels of evaluation and analysis the Coupled Model Intercomparison Project N 3 CMIP3 The models represent the current
60. formation Once the projections are calculated for the places of interest graphical outputs are displayed These can be downloaded jointly with the underlying data allowing users to manipulate the data in their own applications This document provides the user with both a description of the SDM and a user manual of the Graphical User Interface GUI The first part describes the scientific background the objectives of downscaling and how the BoM SDM was developed as well as the predictands obtained how the model was optimised and evaluated and the global climate models used for projections The second part is a comprehensive documentation of the GUI its structure usage and the results provided The appendices provide more details about the optimised parameters of the model its skill the GUI Perl scripts and directory structure the SDM structure and routines and the graphical outputs of the GUI The last appendix illustrates the benefits of the SDM via two case studies projected maximum temperature in Mildura and sub grid heterogeneity BoM SDM GUI documentation Authors B Timbal et al Part A Scientific background 1 Introduction Global climate models GCMs have resolutions of hundreds of kilometres whilst regional climate models RCMs may be as fine as tens of kilometres However impact assessment applications often require point specific climate projections in order to capture fine scale climate variations particularly in r
61. gure A6 Optimum size of the calendar window 15 30 or 60 days as a function of the season of the year The limitation to search for an analogue only during the same calendar season is arbitrary but responds to the fact that large scale meteorological forcings of the local climate change seasonally during the year as reflected in the optimum combination of predictors discussed earlier Accordingly the size of the calendar window is used to force the choice of analogue during the same part of the season However a bigger value for the calendar window would provide a potentially larger pool of analogues hence the choice of the calendar window is a trade off between these two effects As expected there is tendency to choose a smaller calendar window for the transient seasons spring and autumn when the underlying stationarity assumption is less true and a tendency to choose a larger calendar window in winter and in particular in summer when the size of the pool of analogues is most important 55 BoM SDM GUI documentation Authors B Timbal et al Appendix B Analysis of the SDM skill The skill of the SDM was evaluated using a range of metrics on different time scales First the ability of the technique to reproduce the observed Probability Density Functions PDFs was evaluated by looking at the first two moments of the PDFs the mean and the variance The reproduction of the mean values for each predictand Fig B1 is very ac
62. ies across the continent in locations where high quality observations of the current climate are available Individual SDMs were optimised using the high quality BoM network of observations for temperature daily extremes Tmax and Tmin rainfall dew point temperature daily extremes dTmax and dTmin and pan evaporation Stations were lumped together into ten regional climate entities five covering the southern part of the continent from the southwest of Western Australia to the east coast one for the island of Tasmania and four covering the tropical northern part of the continent For these last four regions the SDM has only been optimised for temperature and rainfall Each individual SDM was optimised in two steps first the optimal combination of predictors was determined and then additional parameters of the statistical model were determined size of the domain from which predictors are obtained calendar window and method to calculate daily anomalies In total 192 individual SDMs were optimised six predictands times six regions the southern half of Australia plus three predictands times four regions the tropical northern part of the continent and in all cases for four seasons The analogue approach used here is one of the simplest existing downscaling methods This allowed for the extensive work done calibrating the model Despite its simplicity this method has been shown to compare well with more advanced techniques Zorita and von
63. ighest lowest slope MILDURA F Obs Mean 37 5 Range 4 0 Slo 0 9 4 209 GOM Mie 555 RIT AS Sh 37 DO M 33 2 R 2 4 Sl 3 24 34 P 32 30 28 26 24 2050 2055 2060 2065 Figure F3 Inter annual variability of maximum temperature in summer in Mildura region SMD for the middle of the 21 century 2046 to 2065 using emission scenario A2 for CSIRO CCM and CNRM models The x axis represents the years of observations projections the y axis the average temperature in degrees Celsius Finally in order to document the ability of the technique to reproduce extreme events the probability of duration of anomalous long spells are shown e Dry and wet spell for rainfall e Hot and cold spells for temperature as well as return period for hot and cold days For rainfall a wet spell duration WSD is defined as the number of consecutive days with rainfall recorded in excess of 0 3mm This threshold can be modified within the 81 BoM SDM GUI documentation Authors B Timbal et al code and could become an option chosen by the user Similarly dry spell durations DSD are being plotted using the same threshold 0 3mm to define dry days For temperature hot and cold spell durations HSD and CSD are plotted see an example for HSD left diagram in Fig 4 as well as hot and cold day return periods HDR and CDR see an example for HDR right diagram in Fig 4 A hot spell of n da
64. imbal et al Appendix A Analysis of the SDM optimised parameters Statistics on the type of predictors chosen reveal that the best combinations are based on temperature air flow moisture and synoptic predictors The optimal number of predictors within a combination Fig Al is often three apart from pan evaporation where most frequently only two predictors are used When a different number of predictors are required for the optimum combination it is often less than three predictors apart from rainfall The need for a large number of predictors for rainfall shows that it is a difficult predictand to capture from large scale analogues Not only does rainfall require a large number of atmospheric variables but it is the predictand for which skill scores are the lowest see details in the next section There are a few cases for each predictand except for pE where no combinations of predictors were found to improve on the best performing single predictor Number of Predictors used 70 Bi 82 03 4 60 50 40 30 20 10 0 Tmax Tmin Rain pE DTmax DTmin Figure A1 Number of predictors used in percentage over all SDMs for every predictand Mean sea level pressure MSLP is the most frequently chosen predictor Fig A2 top left It is used for most individual SDMs in the case of rainfall dTmax and 49 BoM SDM GUI documen
65. ine in total rainfall in this part of Western Australia from the coast between 3 and 22 in Cape Naturaliste and King River to the inland wheat belt between 0 and 36 in Corrigin On the contrary further inland Norseman the rainfall is more likely to increase as projected by 6 models than to decrease projected by only 4 models This difference between the southwest corner and further inland is consistent with the direct model outputs projections Bom and CSIRO 2007 but with a sharper contrast As was the case for temperature the statistical downscaling product has the ability to show notable differences within small distances consistent with the broad picture provided by GCM direct outputs but with a sharper contrast In addition as the full distribution at each location provided is given a range of statistics such as those discussed earlier can be obtained to help characterise the nature of the mean rainfall changes 89 BoM SDM GUI documentation CORRIGIN 100 0 ogs x061 A 4 5 R50 9 0 TR 157 4 100 0 80 0 LECT 30 62 A 4 5 RS0 8 5 TR 157 a ua GCM D 72 Al 3 7 R50 6 5 TR 100 60 0 F J 40 0F 20 0F 0 0 CAPENATURALISTE 100 0 ogs 32 A 7 1 R50 12 5 TR 440 80 0 GCM XD 32 Al 7 2 R50 12 5 TR 428 GCM D 41 Al 6 4 R50 10 5 TR 359 60 01 40 0 20 01 0 0 10 0 f OBS Authors B Timbal et al NORSEMAN O 76 Al 3 6 R50 6 5 TR 79 Ly GCM XD 79 A
66. ing observations is limited and or the number of classifying predictors is large The BoM SDM was first developed for daily maximum and minimum temperatures Tmax and Tmin across the Murray Darling Basin MDB in Australia Timbal and McAvaney 2001 The choice of a single best analogue is based on a closest neighbour using a simple Euclidean metric The metric is applied to a single vector which comprises daily normalised anomalies of point values within an optimised geographical area for the selected predictors The choice of the optimal combination of predictors and the geographical area are two key steps in the optimisation of the analogue model The SDM was then extended to rainfall occurrences Timbal et al 2003 and amount Timbal 2004 During these applications the technique was tested on other mid latitude geographical areas in both Australia and Europe The ability of the SDM to reproduce shifts in the observed climate a proxy for its ability to reproduce the shift in current climate due to global warming has been tested for the rainfall decline in the late 1960s in the Southwest of Western Australia SWA Timbal 2004 and in the mid 1990s in the Southwest of Eastern Australia SEA Timbal and Jones 2008 It shows that the method is able to capture observed rainfall BoM SDM GUI documentation Authors B Timbal et al step changes providing that the appropriate large scale forcings are used It both cases besides the syno
67. ized anomalies rr matching analogue Smallest Euclidean Distance The reference database is the NCEP NCAR reanalyses from 1958 to 2005 Phase 1 The SDM model is set up choice of parameters during a development phase downscaling from NCEP NCAR reanalyses steps 1 to 3 are repeated until all parameters are optimized evaluation is based on a range of statistics Step 3 Associated with local predictands on the Phase 2 The SDM is applied to Rain 240 climate model simulations date of the best 1 20 century simulations Dr matching analogue Evaluate downscale current climatology on 2 37 Either Rain T dT pE Evaluate climate model biases stat 3 0 Generate local 2 Future projections for 215 century downscaled series Remove climate model biases Evaluate downscaled Generate downscaled projections vs observed series Figure 2 Schematic diagram of the methodology used to optimise phase 1 and apply phase 2 the Statistical Downscaling Model The various steps involved to obtain meteorological analogues steps 1 to 3 are described in the text 13 BoM SDM GUI documentation Authors B Timbal et al Some manipulations of the elements of the vector are needed to ensure that all elements are comparable and hence have equal weight when it comes to finding the best matching analogue Every element of the vector formed is a daily normalised ano
68. l 5 4 R50 11 5 TR 94 GCM XD 82 Al 3 9 R50 6 0 TR 61 Se eres riltasai to KINGRIVER T ee D 36 Al 6 0 R50 11 0 TR 349 OBS GCM D 39 Al 5 9 R50 12 0 TR 329 GCM XD 47 Ale 5 3 R50 9 5 TR 270 10 0 Figure H2 Projected changes of the observed cumulative probability distributions of Australian rainfall in winter for 2050 using emissions scenario A2 in four locations in the southwest of Western Australia with contrasting climate Cape Naturalist along the west coast King River along the south coast Corrigin in the heart of the wheat belt and Norseman further inland Observations based on the 1970 to 1989 period are shown as thick black lines Ten individual climate models from the IPCC AR4 database are shown as red lines The x axis in mm used a logarithmic scale while the y axis is expressed in percent Some statistics extracted from the cumulative distribution are provided for the current climate first line in each insert and for the range of future projections largest rainfall on the second line and smaller rainfall on the third line The meaning of each statistics is discussed in the text 90 The Centre for Australian Weather and Climate Research is a partnership between CSIRO and the Bureau of Meteorology
69. l it is worth noting that the reproduction of the mean is dependent on the ability of the technique to reproduce the observed variance because rainfall is not normally distributed In the case of the other variables as daily values 57 BoM SDM GUI documentation Authors B Timbal et al are more normally distributed the underestimation of the variance does not have a flow on effect on the reproduction of the mean 40 25 35 Tmax variance Tmin variance 20 30 mo xo 2 05 2 2 DI t 20 v c c Q 15 4 9 10 4 O O 10 cc 5 5 4 0 r i i i i i r 0 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 Observed Observed 300 17 e 16 o Rain variance 15 pE variance ee 13 Ke bd 124 200 5 2114 3 319 7 150 4 o g E S 7 2 S 8 100 oc cc 1 44 4 34 50 5 14 0 7 i i t 7 0 TF _ _ aS 0 50 100 150 200 250 300 012 3 4 5 6 7 8 9 10 1112 13 14 15 16 17 Observed Observed DTmax variance DTmin variance Reconstructed Serger a i c WARUONWDOO NWHUDNDWOSO Reconstructed N 3 4 5 6 7 8 9 10111213 1415 16 17 18 19 20 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 Observed Observed Figure B2 As per Figure BI but for standard deviations same units as for the means For this reason Timbal et al 2006 introduced a correction factor to adjust the recons
70. l individual user directories are treated as distinct sub directories of a common root directory on a temporary disk space gwork on the Bureau s research machine Gale more details on the user management and directory structure are provided in Appendix B Due to IDL licensing limitations IDL is a CPU intensive process with limited number of licences available no more than three users are currently allowed to use the GUI simultaneously If a fourth user tries to start a downscaling process a warning will be issued indicating that the GUI has run out of licenses The newcomer will have to wait until one of the current users quits the program Furthermore each individual user directory is subdivided in sub directories for each SDM run allowing the user to perform and store the outputs of up to five downscaling experiments Once this limit is reached the user will need to remove an existing run in order to perform a new one Limited user information is kept in each user directory including at what time the user started using the GUI When returning to the GUI after a while this time is checked to determine if the user should use an existing directory or get a new one Currently after 24 hours users loose their directory and its contents making it available for new users 27 BoM SDM GUI documentation Authors B Timbal et al h Individual subroutines The static HTML pages and their usage are described in detail in the next
71. ler A Hernandez G Kelly X Li K Onogi S Saarinen N Sokka R Allan E Andersson K Arpe M Balmaseda A Beljaars L van de Berg J Bidlot N Bormann S Caires F Chevallier A Dethof M Dragosavac M Fisher M Fuentes S Hagemann E Holm B Hoskins L Isaksen P Janssen R Jenne A McNally J F Mahfouf J J Morcrette N Rayner R Saunders P Simon A Sterl K Trenberth A Untch D Vasiljevic P Viterbo J Woollen 2005 The era 40 re analysis Quart J Roy Meteor Soc 131 2961 3012 32 Van Den Dool H 1994 Searching for analogues how long must we wait Tellus 46A 314 324 33 Von Storch H 1999 On the Use of Inflation in Statistical Downscaling J Climate 12 3505 3506 47 BoM SDM GUI documentation Authors B Timbal et al 34 Wilby R C Dawson and E Barrow 2002 SDSM a decision support tool for the assessment of regional climate change impacts Env Model Software 17 145 157 35 Wilby R L S P Charles E Zorita B Timbal P Whetton and L O Mearns 2004 Guidelines for Use of Climate Scenarios Developed from Statistical Downscaling Methods Published on line supporting material to the Intergovernmental Panel on Climate Change 27pp 36 Zorita E and H von Storch 1999 The analog method as a simple statistical downscaling technique comparison with more complicated methods J Climate 12 2474 2489 48 BoM SDM GUI documentation Authors B T
72. maly calculated from the long term climatology either using monthly or seasonal means All elements of the vector are therefore dimensionless with a mean of zero and a standard deviation of 1 over the length of the period of interest for which analogues are searched for Each vector formed one per daily meteorological situation of the period of interest is then compared to similar vectors formed for every daily meteorological situation in the reference database the reference database is the NNR from 1958 to 2005 The search for the optimal analogue is conducted using Euclidean distance between vectors The smallest Euclidean distance i e the sum of the squares of the differences between each element of the two vectors being compared defines the best matching analogue providing it is within the requirements in terms of calendar window step 2 in Fig 2 A best matching analogue is calculated for every single day of the period of interest and kept in change of date COD files The period of interest is also 1958 to 2005 for the development phase but differs when the SDM is applied to climate simulations i e different time slices during the 20 and 21 century It is worth noting that here the optimisation of the SDM i e the repetition of steps 1 to 3 until the statistics used show the highest skill for the SDM was not done in a full cross validated manner to ensure that the largest dataset could be used The reference datab
73. n T is the temperature in C U and V are the zonal and meridional wind components in m s and p SONO RO A EH Subscript numbers indicate the atmospheric level for the variable in hPa 17 BoM SDM GUI documentation Authors B Timbal et al Tmax T min Rain Summer MSLP amp Tsso amp Vs50 MSLP amp Tsso amp Qss0 amp Usso MSLP amp PRCP amp Qg50 amp Usso Autumn MSLP amp Taso amp Veso MSLP amp Ts50 amp Qsso amp Usso MSLP amp PRCP amp Qsso amp Usso SWA Winter MSLP amp Tsso amp Vs50 MSLP amp Tsso amp Qsso MSLP amp Qgs0 amp Usso Spring MSLP amp Tsso amp Vs50 MSLP amp Tsso amp Qsso MSLP amp Qgs0 amp Usso Summer MSLP amp Tsso amp Vs50 MSLP amp Tsso amp Qs50 MSLP amp PRCP amp Qg50 amp Usso Autumn MSLP amp T3590 amp Vsg50 Tgs0 amp Qeso MSLP amp Tsso amp Qg50 amp Usso NUL Winter MSLP amp Tsso amp Usso MSLP amp Tsso amp Qsso amp Usso MSLP amp Tsso amp Qeso amp Usso Spring MSLP amp Tg50 amp Vaso Tiso amp Osso MSLP amp Qsso amp Tsso Summer MSLP amp Tsso MSLP amp Ts50 MSLP amp PRCP amp Ts50 Autumn MSLP amp Tmax MSLP amp Tsso amp Qsso MSLP amp Tmax amp Qgso amp Usso SEA Winter MSLP amp Tsso amp Tmax amp MSLP amp Tss0 amp Qg50 MSLP amp PRCP amp Vg50 Spring Usso MSLP amp Tss0 amp Qsso MSLP amp PRCP Summer MSLP amp Tmax Tsso amp Qg50 MSLP
74. nage the display of stations Run the downscaling model Display results Show station data Display graph I i f I I I Remove i I I Download i SDM Script running IDL Figure C1 Flow chart of the GUI for the Bureau of Meteorology SDM x_startGUI pl This is the starting point of the web interface The first thing it does is to determine if there is any user directory available to store downscaling results This program checks the cookie from the HTTP request to find out if this user is an old user In that case the user is given a previously assigned directory Otherwise a new cookie is set for 66 BoM SDM GUI documentation Authors B Timbal et al the user and a new directory is assigned provided there is still enough disk space available for an additional user The cookie will expire after 24 hours The getUserDir routine from the module x_util pm is called from x_startGUI pl to determine the user directory When the user directory is determined x_makeGUI pm is executed to generate the web interface and x_makeResultsTable pm is executed to generate the downscaling results table empty at this stage x_util pm This is a utility module that contains routines that manage the user and the downscaling runs each user can make It determines user directory and run directory It is the program where the base directory can be modified x_makeGUI
75. ns the elements of the vector are not normal anymore mean is not equal to zero and variance is not equal to one That abnormality is due to the climate change signal affecting the predictors and is the source of the climate change signal that the SDM will generate for the downscaled projections 16 BoM SDM GUI documentation Authors B Timbal et al e Characteristics of the optimised SDMs A complete analysis of the results from the two optimization steps is provided in Appendix A here we only present a summary of the best combination of predictors The predictors considered were chosen based on experience acquired while developing the BoM SDM Timbal and McAvaney 2001 Timbal et al 2003 Timbal 2004 evidence in the literature from other studies in similar areas Charles et al 1999 2003 and the availability of variables in the climate model database In most cases the predictors are low level atmospheric fields but in some instances large scale rainfall or Tmax or Tmin from the GCMs are used as predictors The optimum combination of predictors varies across regions seasons and predictands Tables 2a for Tmax Tmin and rainfall and 2b for dTmax ATmin and pE The predictors are defined as follows v MSLP is the Mean Sea Level Pressure in hPa Tmin and Tmax are the surface min and max temperature in C PRCP is the total rainfall in mm Q is the specific humidity in g Kg R is the relative humidity i
76. o no 15 25 2 2 7 20 FETE c c 915 8 o 57 io ira 04 5 4 0 T T T T T T T 5 T T T T T 0 5 10 15 20 25 30 35 40 5 0 5 10 15 20 25 Observed Observed 600 15 14 550 Mean rainfall 13 500 12 450 4 o 11 2 400 2 10 4 S 350 3 g 7 300 7 c c 5 250 9 6 200 2 5 150 i 100 5 50 4 1 0 A Onaman Pa E SEE VS S E S E TT 0 At ao A 0 50 100 150 200 250 300 350 400 450 500 550 600 012 3 4 5 6 7 8 9 10 11 12 13 14 15 Observed Observed 20 17 19 J 16 7 18 Mean DTmax 1 Mean DTmin 16 4 13 9 15 4 912 2114 144 eq 9 13 10 2 2 9 124 v sl A 11 O c 74 o 10 O 64 O 9j is Q 54 om 84 oc 4 74 34 6 24 5 4 14 44 04 e a ot oo ot oo oo ot 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 10123 4 5 6 7 8 9 101112131415 1617 Observed Observed Figure B1 Scatter plot of the reconstructed versus observed means of the series for the six predictands On each graph there is one point per station and per season The total number of points per graph is the number of stations across the ten regions times four for Tmin Tmax and Rain the six regions times four for dTmin dTmax and Evap The line of perfect fit the diagonal is shown The colour code refers to the season blue is winter green is spring red is summer and orange is autumn Units are C for temperatures and dew point temperatures mm for rainfall and mm day for pan evaporation In the case of rainfal
77. odel predictors plotted using grid boxes instead of smooth lines c_analog pro This is the subroutine where the analogues are calculated once the predictors have been prepared according to the chosen options The calculation of predictors can be achieved using different methods Results are achieved in the COD files and several analogues can be archived p_correl pro Once analogues have been calculated this procedure can plot the spatial correlation between a day and its analogue averaged across all dates considered o_sta pro 76 BoM SDM GUI documentation Authors B Timbal et al This subroutine contains all the options relevant to the predictands It can be accessed as a standalone once the calculation of the analogue is completed to run the SDM relevant to the choice of predictors and parameters of the models o_sta_m pro is the same subroutine but set up to run through several COD files when used interactively This subroutine is accessed by the GUI and supplied with the user options g_sta pro This is the main code for the treatment of the predictands Reconstructed series based on the choice of analogue stored in the COD files are generated here Then the code calls a series of subroutines to calculate statistics and generates plots for local predictands The version that deals with multiple COD files is g_sta_m pro sp_pfplot pro This procedure plots PDFs of the reconstructed and observed series for near Gaussian
78. of the subroutines within the Bureau of Meteorology SDM The file names starting with o_ denote options files g_ denotes general subroutines c_ denotes calculation on the predictor s fields p_ denotes subroutines to plot predictor results and sp_ denotes calculations on surface predictands Files with _m suffix denote subroutines where multiple use of change of date files is possible Scripts to run the SDM are called xidl xidldown is used to run the entire model from the treatment of the predictors to the calculation of the analogues COD files and the treatment of the predictands Options exist to use this script and re run 74 BoM SDM GUI documentation Authors B Timbal et al predictands calculations while using an existing COD file xidlsta_m is used when COD files already exist and several are to be used to compare outputs from different SDM projections or the downscaling of several climate models o_down pro This subroutine contains all the options to run the SDM relevant to the choice of predictors and parameters of the models The code contains several notes detailing the choices regarding 1 the choice of model simulations 2 the available predictors 3 the existing choice of predictands 4 the different regions of interest and 5 the options for choosing a season g_down pro This is the main part of the SDM it receives from o_down pro the user defined options to run the statistical model and calls
79. olution grid Ricketts and Page 2007 or the TETYN software http sourceforge net projects tetyn which used global database of climate indices Solymosi et al 2008 Besides tool based on DMOs a few attempts have been made to enhance the relevance of the climate information provided by on line software for high resolution impact studies by downscaling climate change information The first documented attempt was the UK based SDSM www sdsm org uk this is a general downscaling methodology that one can optimise and applied anywhere once the local data needed are provided Wilby et al 2002 Hessami et al 2008 described a regional optimisation and application for Canada More similar in scope and method to the tool described here is the UK based Environment agency Rainfall and Weather Impacts Generator EARWIG however relies on a weather generator type of downscaling Kilsby et al 2008 as does RainSim Burton et al 2008 while the BoM GUI described here relies on a physically based downscaling method rather than a stochastic weather generator namely meteorological analogues 42 BoM SDM GUI documentation Authors B Timbal et al Acknowledgements The development of the Bureau of Meteorology BoM statistical downscaling technique has long been supported by the Department of Climate Change DCC The latest development of the graphical user interface was undertaken as part of the Australian Climate Change Science Progr
80. ort BoM and CSIRO 2007 c Future developments of the tool As noted earlier currently downscaled predictand series are constructed independently from one variable to another This is possibly a limitation for impact studies that require several predictands i e rainfall and temperature The possibility to generate multivariate projections and ensure full consistency between predictands will be investigated 40 BoM SDM GUI documentation Authors B Timbal et al Further on going developments are under way to maximise the benefits from this work E g it is planned to provide gridded projections The BoM has developed daily high resolution 0 05 by 0 05 degree gridded data for rainfall and temperature Jones et al 2007 This is an interesting development that could be used in the current framework to provide downscaled climate change projections on the same grid scale The adaptation of this synoptically driven technique to tropical areas has proven a challenging exercise Results presented here show that the technique is less skilful in hot climate e g summer and northern Australia in particular for rainfall It has also confirmed earlier findings that direct model outputs such as rainfall are more important in tropical areas to reproduce the local rainfall Robertson et al 2004 In the course of this development it becomes apparent that improvements of the technique are needed to remove some of the issues described in
81. possible at this stage to select several scenarios in one go Select the climate model s from the Model dropdown list The user can choose all available models or select an individual CMIP3 model Click the Select M S button a combination of model s and scenario will be added to the box below All models can be selected in one go or individual 31 BoM SDM GUI documentation Authors B Timbal et al models one by one By clicking on an individual model it can be removed from the list using the Delete M S button Step 3 Select a region and the station s Note this step can actually be activated as soon as the predictand is chosen Since the stations to be displayed depend on which HQ dataset is used it does not require the other pull down menus from step 1 and step 2 to be used Select the region of interest from the map There are 10 available regions depicted as rectangular areas with a green marker on the upper left corner When the mouse is over a green marker a tool tip appears showing the abbreviation of the region name Clicking on the green marker will display the region on the map and the Selected Region field will be filled with the abbreviation of the region Fig 6 shows an example for the Southwest of Western Australia SW A region Select the stations Available stations are displayed by yellow markers within the rectangular area When the mouse is over a yellow marker a tool tip shows the station name an
82. ptic dynamic and thermal variables the optimal combination of predictors must include a moisture related variables Furthermore downscaled projections from the BoM SDM were compared with direct model outputs and were shown to be consistent when averaged back on a scale relevant to climate model resolution Timbal et al 2008b The technique was also used to enhance the signal in the classic signal to noise problem that is the possible attribution to external causes of an observed change detected as significant In the case of the rainfall decline in SWA using the SDM it was possible to partly attribute this decline to greenhouse gases and other large scale emissions Timbal et al 2006 with an additional contribution from regional land clearance Timbal and Arblaster 2006 Although the ability of the statistical linkage to reproduce non stationary climates has been thoroughly tested in several instances it is one of several limitations often cited as possible short comings of SDMs Hewitson and Crane 2006 Christensen et al 2007 Users should be familiar with these issues and possible limitations in using downscaled outputs as discussed as discussed in Timbal 20062 e That large scale predictors relevant to the local predictands are adequately reproduced by the climate models e That the statistical relationship between predictors and predictands is stable and valid in a warmer world and e That the chosen predictors encompass
83. r user community a graphical user interface GUI that provides projections across the southern half of the Australian continent has been developed The GUI is a web based tool using a combination of static HTML HyperText Markup Language pages and dynamically generated pages It provides users with access to projections for a series of surface predictands at point specific locations BoM SDM GUI documentation Authors B Timbal et al 2 The Bureau of Meteorology Statistical Downscaling Model The Australian Bureau of Meteorology BoM has developed a SDM using the idea of a meteorological analogue Timbal and McAvaney 2001 This is one example of a more general type of SDM based on weather classification methods in which predictands are chosen by matching previous i e analogous situations to the current weather state The method was originally designed for weather forecasting applications Lorenz 1969 but was abandoned due to its limited success and lack of suitable analogues for systems with large degrees of freedom Van Den Dool 1994 The popularity of the method has recently increased with the availability of datasets covering longer time periods following the completion of several reanalysis projects and the recognition that the dimension of the search space i e the spatial domain and the number of predictors must be suitably restricted when identifying analogues The analogue method still performs poorly when the pool of train
84. rainy day is currently defined as a day with at least 0 3mm of precipitation but this threshold can be modified in the code WAKOOL 100 0 T roBs D 70 Al 4 2 R50 8 5 TR 111 GCM D 79 Al 5 0 R50 9 0 TR 93 GCM D 79 Al 5 0 R50 9 0 TR 93 60 0 F Percent o oL 1 I 1 0 10 0 mm day Figure F2 Cumulative distribution functions for rainfall in winter in Wakool region SMD for the end of the 20 century 2081 to 2100 using emission scenario A2 for the CSIRO model The x axis logarithmic scale represents rainfall in mm day and the y axis the corresponding probability in percent 80 BoM SDM GUI documentation Authors B Timbal et al An inter annual variability plot Fig F3 is also generated whenever surface predictands are chosen It depicts the year to year evolution of the seasonal mean temperatures or seasonal total rainfall across the period considered as shown in Figure Fl for summer Tmax in Mildura from 2046 to 2065 Some statistics are available on the figure for the observations the mean for temperature or total for rainfall Mean range Range i e the difference between the highest and lowest seasonal means and slope Slo i e the slope of the linear regression fitted to the observations Then the same statistics are provided for the downscaled results M M is the highest lowest mean among the models used R R is the highest lowest range and Sl SI is the h
85. recip mU Vv 100 R 100 Q 80 80 60 60 40 40 0 r T r T T 0 iN e 1 l imax Tmin Rain pE DImaoe DTmin Tmax Tmin Rain pE DTmax DTmin Figure A2 Percentage of SDMs where a particular type of variable is part of the optimal combination of predictors synoptic predictors top left moisture predictors bottom left temperature variables top fight and air flow predictors bottom right for the individual SDMs separated according to the predictands being modelled For Tmax Tmin and Rain each bars based on 40 cases ten regions times four seasons for pE DTmax and DTmin each bar is only based on 24 cases six regions times four seasons 50 BoM SDM GUI documentation Authors B Timbal et al Thermal predictors Fig A2 top right are very important especially for Tmax and Tmin In general Tgso is the most important thermal predictor although Tmin is more important for dTmin and AT max Dew point temperature in most instances has a weak diurnal cycle with maximum values in the early morning and maximal value in the afternoon Lucas 2006 Therefore while the importance of Tmin for dTmax is logical its relevance for dT min is less intuitive A total percentage in excess of 100 means that in some cases more than one thermal predictor is used in the optimised combination The additional predictor is usually Tmax for Tmax and Tmin for Tmin a clear indication that in these cases lower tropospheric temperature alone is not
86. redictand considered graphical outputs are described in detail in Appendix F In order to display the data for a particular location select the station file in the right box and click the Show Data button A pop up window will appear containing the daily values for that location Fig 9 The number of columns depends on the number of GCMs chosen one per model plus a first column containing the date and a last column containing observations either actual observations for the 20 century scenario or equivalent observations from 1975 to 1994 for the 21 century scenarios details are provided in Appendix F 35 BoM SDM GUI documentation Authors B Timbal et al Downscaling Output User 1 Run 3 Figures Data m_cpf_rain1 m_cpf_rain2 m_rain_spell_wsd1 m_rain_spell_wsd2 m_rain_spell_dsdl m_rain_spell_dsd2 intervar_rain_1 intervar_rain_2 Download all figures and data Show Figure coondambo_rain_la kondoolka_rain_1a nonning_rain_la purple_downs_rain_la yardea_rain_la farina_rain_1a marree_rain_la pennalumba_rain_la goode_rain_la lakehamilton_rain_1a elliston_rain_la nullarbor_rain_la Figure 8 The outputs page showing all available outputs figures left box and data files right box Upon a selection from either the list of figures or data files it will be displayed in a pop up window mildura_tmax_1 data Date 01012046 02012046 03012046 0401
87. represents the probability of duration on the left plot and the return period on the right plot 82 BoM SDM GUI documentation Authors B Timbal et al Appendix G Case study 1 projected Tmax in Mildura To illustrate the usefulness of the BOM SDM as a tool to obtain detailed point specific climate change projections a case study was completed using the GUI In this example maximum daily temperatures in summer were sought for Mildura in the northwest of Victoria within the SMD region which is an important regional centre in particular for agriculture Additional case studies for both temperature and rainfall can be found in the Climate Change in Australia report BoM and CSIRO 2007 MILDURA PDF of TMAX MILDURA PDF of TMAX Obs Mean 31 5 Yar 27 2 Mn 16 2 Mx 468 10 GCM M 33 5 4 26 8 Mn 21 8 Mx 46 9 M 32 3 V 22 8 Mn 16 9 Mx 46 0 Obs Meqn 31 5 Var 27 6 Mn 14 5 Mx 46 9 10 FGCM M 31 8 V 28 4 Mn 19 5 Mx 46 8 M 31 5 V 24 8 Mn 16 9 Mx 46 0 aB e T wu o 6 Cb o v a a 4 4 0 Z Di matt 10 20 30 40 10 20 30 40 Degree Celsius Degree Celsius MILDURA PDF of TMAX MILDURA PDF of TMAX Obs Mean 31 5 Var 27 4 Mn 16 2 Mx 46D 12 fObs Mean 31 5 Var 27 3 Mn 16 2 Mx 468 10 FOCM M 33 7 V 27 2 Mn 22 1 Mx 46 5 M 32 8 V 23 1 Mm 18 5 Mx 45 7 10 B E Do 6 D B a a 4 4 2 2 D c amp 0 ri 10 20 30 40 10 20 30 40 Degree Celsius Degree Celsius Figure G1 Probability density
88. roducing day to day variability that is driven by large scale synoptic changes A random choice of analogue may reproduce perfectly the observed mean and variance but may not be a skilful model The Pearson correlation between daily observed and reconstructed series was calculated separately per region per season and for each predictand Each number is an average across all observations available in each region Fig B3 The results show a contrast between predictands the SDM appears to be more successful for Tmax Tmin and dTmax than for pan evaporation and dT nin For rainfall correlations are by far the lowest although due to the very large sample considered about 4500 days all these correlations are significant at least at the 95 level based on rain occurrences only in the case of rainfall indicating some level of skill For most variables there is a marked seasonal cycle in skill consistent across all regions i e the analogue approach is particularly successful in autumn and spring for temperature predictands Tmax and Tmin nearly reaching a correlation of 0 9 for Tmax in autumn in several regions and for pan evaporation albeit with lower values For rainfall although correlations are low across all seasons they peak in winter between 0 3 and 0 4 apart from QLD region where the correlation is below 0 2 in all seasons and NMR region where the peak is in spring For dew point seasonal variations of the results are less marked and no
89. rs are important in affecting the choice of analogues Timbal 2004 and were therefore systematically explored as part of the extension of the technique across the Australian continent 1 The size of the geographical domain used for the predictors latitude and longitude Only two domain sizes chosen a priori based on previous studies were tested 2 The calendar window from which analogues are found Three periods were tested 15 30 and 60 days prior to and after the model date 3 The way the daily anomalies are calculated using either three monthly values or a single seasonal average 12 BoM SDM GUI documentation Authors B Timbal et al b Defining daily meteorological situations Daily meteorological situations are summarised by a single vector step 1 in Fig 2 containing a value for every latitude longitude point within the geographical domain chosen and for all the predictors selected i e the vector length is equal to the size of the geographical domain number of latitudes times number of longitudes multiplied by the number of predictors The mathematical need for a vector is to be able to combine different maps one per predictor chosen as part of the optimised combination of predictors Step 1 Summarize the daily synoptic situation by n Step 2 select predictors upper air and surface variables select size of the geographical domain Search for the best single vector combined maps of normal
90. section A brief summary of the Perl scripts pl and modules pm is provided in Appendix C 28 BoM SDM GUI documentation Authors B Timbal et al 2 Using the GUI a Access URL Currently the GUI for the SDM runs on gale the Bureau s research machine can only be accessed from within the Bureau s intranet and is not available on an external web site However access to the broader community on a dedicated web site supported by the Department of Climate Change DCC is being planned The URL for access is http gale ho bom gov au bxt sdm_qui html Note Currently the GUI has only been tested and optimised for the Microsoft Internet Explorer web browser problems will be encountered if users attempt to run the GUI with other browsers such as Firefox b Front page layout The front page layout Fig 4 has been kept to a minimum It gives access to the available documentation scientific literature underpinning the technical tool and the user manual this document including some technical description of the software and the IT choices made to optimise the usage The front page provides a button Contact for users to send feedback comments or queries to the development team In all subsequent pages hitting the Home button will bring the user back to this front page Hitting the Start the GUI button will lead the user to the interactive web interface itself 29 BoM SDM GUI documentation Authors B Timbal et
91. series temperature dew point evaporation There is a separate subroutine sp_pfplotprec pro to plot accumulated PDFs for rainfall Both subroutines have an _m option to plot multiple results sp_spell pro This procedure plots spell durations and inter arrival times between events An _m option exists to plot multiple results In addition there is a procedure to plot inter annual variability of the number of fixed spells sp_interspell pro sp_rocindex pro This procedure calculates statistics on the reproduction of rainfall occurrences WMO index hit rate Itis mostly useful during the validation phase of the SDM sp_verif pro T11 BoM SDM GUI documentation Authors B Timbal et al This procedure calculates a range of statistics on the ability of the SDM to reproduce observed series correlation root mean square error biases It is mostly useful during the validation phase of the SDM sp_autocorr pro This procedure calculates the auto correlation of the reconstructed series and compares it to the observed series It also calculates the probability of an event following the same event such as rain occurrences sp_intervar pro This procedure plots inter annual variability of the reconstructed series compared to the observed one calculations for that plot are done within the g_sta pro subroutine There is an _m pro version to plot multiple results Additional subroutines and utilities exist to help manage
92. t consistent across regions Finally there seems to be a high consistency in the 59 BoM SDM GUI documentation Authors B Timbal et al performance of the SDMs across the ten regions considered Overall no particular region stands out as a climatic entity where the SDM skill in reproducing day to day variability is consistently lower or higher across all variables and seasons The fact that the model was assumed to be applicable in all the extra tropical regions of the Australian continent where the climate is driven by synoptic disturbances is vindicated by these results 1 0 5 gt 1 0 0 9 J O Summer E Autumn M Winter E Spring da OSummer B Autumn Winter Spring 0 8 PA 2 07 0 8 o 0 7 5 06 5 6 io 06 4 05 amp 044 g 05 o 5 04 O 03 amp 0 0 2 5 0 3 0 1 5 0 2 0 0 i T T i i i T 0 1 o 0 0 T T T FX TS LC PECE E Tmax F D S LS S Tmin SWA NUL SEA SMD SEC TAS QLD NWA NMR MEC 19 DS B Aut m Wint D Sprin 19 mmer mn inter 0 9 5 ee ib pu 0 9 DSummer Autumn Winter OSpring 0 8 0 8
93. ta quality does not affect the choice of parameters optimisation of the SDM the COD file becomes independent of the predictand database Therefore it is applicable to any other observations provided they are within the same region and of the same meteorological variable That feature makes possible extensions of the tool to user chosen datasets as well as gridded observations future perspectives for the BoM GUI are discussed in the Summary and additional information section More details on the COD files and how they are used by the GUI are provided in Appendix E d SDM optimisation versus application to climate model simulations The same method minimisation of the Euclidean distance to find the best matching analogue is used during the optimisation of the SDM Phase 1 in Fig 2 and for its application to climate model simulations Phase 2 in Fig 2 Both phase 1 and 2 include all three steps but once the SDM has been optimised for a particular season region variable the choice of parameters predictors and additional parameters is frozen when applied to model simulations 15 BoM SDM GUI documentation Authors B Timbal et al The SDM is always applied first to simulations of time slices of the current climate i e the latter part of the 20 century so that the reliability of the downscaling for a particular climate model can be assessed Although the SDM removes some of the biases affecting climate models the objective
94. tation Authors B Timbal et al dTmin but is picked up far less often for pan evaporation This feature combined with the fact that the SDM shows overall low skill discussed in the next section for pan evaporation suggests that MSLP is a critical predictor for a synoptically driven technique such as the analogue approach when it does not perform as a useful predictor none of the other available predictors can compensate for the lack of skill contribution from MSLP To predict rainfall in the tropical half of Australia MSLP combined with Z500 geopotential height at 500hPa adds extra skill to the model Unfortunately Z500 is not available in the CMIP3 datasets so the SDM has been downgraded for these regions to less skilful models using only variables stored from the IPCC models runs Synoptic predictors Temperature predictors 120 120 MSLP mT850 mTmax OTmin 100 100 80 80 60 60 40 40 20 20 ii BEE a be E u Tmax Tmin Rain pE DTmax DTmin Tmax Tmin Rain pE DTmax DTmin Moisture predictors Air Flow predictors 120 120 E P
95. ter OSpring o 4 o 704 8 60 g 604 ta 8 50 g 50 d 0 5 404 30 o 30 20 5 20 10 4 10 4 0 r r r 7 0 DTmax SWA NUL SEA SMD SEC TAS DTmin SWA NUL SEA SMD SEC TAS Figure B5 As per Figure B3 but for the percentage of the observed inter annual variances reproduced by the reconstructed series Overall the SDMs show a slightly better correlation on an inter annual time scale than on a daily timescale but results are very consistent between the two time scales The lowest correlation values are obtained for rainfall albeit much improved compared to the daily time scale and highest for temperature The seasonal variations of the results grossly resemble what was noted on the daily time scale but if anything they are slightly less obvious either differences are smaller or consistency across regions is not as strong For rainfall the model is most able to reproduce inter annual variability in winter as was the case for daily variability and in autumn for dew point that was also the case for daily variability Although the correlations suggest 62 BoM SDM GUI documentation Authors B Timbal et al that the technique is able to capture most of the inter annual variability it appears to reproduce only a fraction of the inter annual range Fig B5 Results are particularly low for pan evaporation below 50 in most instances and AT min For Tmax the percentage of reproduced variance is low in winter 30 to 60 b
96. the subsequent procedures to analyse the predictor s fields looping over the number of predictors chosen before calling the subroutine to calculate the analogue and finally calling the procedure to treat the predictands c_dataselect pro This subroutine reads the binary files containing any model reanalyses IPCC models other fields and extracts the appropriate window as defined by the user in o_down pro If necessary the model grids are interpolated to be comparable with the reanalyses c_pca pro 75 BoM SDM GUI documentation Authors B Timbal et al This subroutine calculates normalised anomalies from the model fields according to some options chosen by the user 1 the base climatology validation control scenario 2 what type of anomalies monthly seasonal spatial and 3 whether to detrend the data or not Then the procedure will plot if required the mean and standard deviation of the predictors Finally if required the code will calculate the leading Empirical Orthogonal Functions EOFs onto which the analogues are to be calculated unless the option rawfield has been selected for the choice of analogue p_ pro These subroutines generate some plots of model predictors v p_avplot pro mean and variance of any predictor field Y p_pcplot pro the leading PCs Y p_coplot pro the loading of the leading PCs Y p_corrsta pro spatial correlation of a field with a local observation and v p_dmoplot pro m
97. thermal moisture is evident from examining Tables 2a and b It is most useful for rainfall and Tmax and least useful for dew point temperature and pan evaporation The zonal component is the most frequently used 51 BoM SDM GUI documentation Authors B Timbal et al Overall statistics on the choice of the three additional parameters optimised for each individual SDM show some patterns worth commenting The optimum size of the geographical domain used to search for analogues the small domains are shown in Fig A3a and large domains in Fig A3b is most often the larger of the two sizes tested Fig A4 There is a hint that the need to reduce the size of the geographical domain is seasonally dependent with the use of the smaller domain being more frequent during the warmer seasons 40 of al SDMs in summer and 54 in autumn compared to 40 and 38 in winter and spring Figure A3a Small Geographical domains on which the predictors are used for the six areas of interest SWA solid line NUL dashed SEA dotted SMD dash dotted SEC solid line TAS dash dotted NWA dash dotted NMR dashed OLD dotted and MEC dash dotted 52 BoM SDM GUI documentation Authors B Timbal et al G Figure A3b Large Geographical domains on which the predictors are used for the six areas of interest SWA solid line NUL dashed SEA dotted
98. this documentation Therefore the development of the BoM GUI is likely to continue is the near future to ensure that it provides state of the art downscale climate change projections Additional planned developments concern the web access to the GUI Currently the GUI is only available internally within the BoM and is used by regional climate service centres to provide tailored climate change projections upon request The model is currently being ported on a national computing facility the Australian Partnership for Advanced Computing APAC to provide a wider access for the research community Furthermore plans to provide outputs directly to the general public as part of a larger online climate projections site a central portal providing access to observed climate data and future projections generated by different methods OzClim or the BoM SDM are being investigated 41 BoM SDM GUI documentation Authors B Timbal et al d Other existing software complementary to this tool The GUI has been developed specifically to be used within Australia and is particularly directed at impact studies e g agriculture health ecology and economy Currently most available tools rely on DMOs rather than downscaled projections e g the Ozclim software http www csiro au ozclim home do developed by the CSIRO to provide regional projections across the Australian continent where various climate models information are interpolated on a high res
99. to be the least successful Fig 1 Location of the high quality networks across Australia for rainfall R temperature T dew point temperature D and pan evaporation E The boundaries of the ten areas of interest are overlaid Tasmania TAS Southwest of Western Australia SWA Nullarbor Plain NUL the Southwest of Eastern Australia SEA the Southern part of the Murray Darling basin SMD the South East Coast SEC the Mid East Coast MEC Queensland QLD the Northern Monsoon Region NMR and the Northwest of Western Australia NWA 3 The Southwest of Eastern Australia SEA southwest of a line from Melbourne 38 S and 145 E to Port Augusta 33 S and 138 E this area was identified as the centre of the early winter rainfall decline and having strong similarities with the SWA Timbal and Jones 2008 4 The Southern half of the Murray Darling Basin SMD limited to about 33 S in the north limited in the west by the line described earlier separating the region from the SEA and limited in the east by the Great Dividing Range GDR BoM SDM GUI documentation Authors B Timbal et al 10 The South East Coast SEC the coastal band east of the GDR from Wilson Promontory in Victoria in the south all the way along the NSW coast up to the Queensland border The island of Tasmania TAS including all Bass Strait islands The Northwest of Western Australia NWA limited in the south by
100. trasted with what is happening for record low temperatures shifting from 16 2 C to between 19 5 and 22 7 C a massive increase of up to 6 5 C by 2100 with the A2 scenario compared to the lowest value observed between 1970 and 1989 Similar shifts might be a reasonable assumption for record hot temperatures but it is not possible to infer this from the current set of results Therefore in the case of warm extremes only the frequency of occurrences below the local record temperature would be meaningful for future projections 86 BoM SDM GUI documentation Authors B Timbal et al Appendix H Case study 2 sub grid heterogeneity a For temperature Climate models provide grid box averages Skelly and Henderson Sellers 1996 Due to the coarse resolution of most climate models uniform information is given for a large area and therefore sub grid heterogeneity is not captured Hence future projections do not provide these local details which are highly important to understanding the impact of projected climate change on the natural and man made environment The importance of local differences across a geographical region represented by a grid box of a global climate model is illustrated in Figure H1 It shows probability density functions of temperature change for four locations in southeast New South Wales that could easily fit within a single grid box This area encompasses part of the Great Dividing Range The results show not
101. tructed rainfall series enhance the variance and improve the reproduction of the mean The rationale for the correction applied is that the reconstructed rainfall is affected by the size of the pool of analogues to choose from therefore in the case of large rainfall events which are rare the error is larger The size of the pool depends on the ratio of rain days over dry days It was decided that a very simple factor should be applied to limit some of the danger linked to artificially enhanced variances 58 BoM SDM GUI documentation Authors B Timbal et al in downscaling techniques van Storch 1999 Therefore a single factor was used which only depends on the availability of dry and wet days to find a suitable analogue across all stations and all seasons 140 1x 22 dC lt 15 C factor 2 ra an factor wet Where Nary and Nwet are the number of dry and wet gt 0 3mm days observed for the season at an individual location Note that all rainfall results presented so far include this factor showing the ability of this simple correction to return unbiased mean rainfall estimates Fig B1 and very low rainfall variance biases compared to the other predictands for which no attempt has been made to enhance the daily variability Fig B2 Besides the ability of the technique to reproduce the observed shape of the PDFs as defined by the first two moments of the series it is important to ensure that the technique is skilful in rep
102. uent as implied by the PDFs in Fig G1 Looking at hot spell duration using a threshold fitted to the new distribution implies that the population and the environment would have adapted to the mean warming However as users can access ASCII files with the daily values used to construct the PDFs they can perform additional analyses on extremes and decide on which thresholds to use The graphs generated by the GUI Figs G1 and G2 serve to illustrate the main characteristics of the downscaled results and the possibility of generating many analyses as the locally constructed daily series in the 20 centur climate are highly realistic y y gnly When dealing with extremes and using downscaled projections it is important to keep in mind that the statistical linkage established using the data is more likely to be robust for the middle of the distribution for which there are more observations than for the extremes which are not so extensively sampled For example since the SDM uses an analogue approach it is not possible to get daily projected values that have not been observed before e g new record temperatures In the case study the maximum value is tied to the past record in summer observed at Mildura during the 1958 to 2003 period 46 9 C That explains why despite the mean warming e g for A2_100 bottom right in Fig G1 the upper tail of the distribution does not shift right with the rest of the distribution This situation can be con
103. ureau of Meteorology and CSIRO GPO Box 1289 Melbourne VIC 3001 AUSTRALIA b timbal bom gov au Phone 61 3 9669 4697 Fax 61 3 9669 4660 ISSN 1836 019X National Library of Australia Cataloguing in Publication Timbal Bertrand 1970 The Bureau of Meteorology statistical downscaling model graphical user interface electronic resource user manual and software documentation B Timbal Z Li E Fernandez ISBN 978 1 921424 77 9 Series Technical report Centre for Australian Weather and Climate Research no 4 Notes Bibliography Also issued in print 1 Climatic changes Mathematical models 2 Statistical weather forecasting Australia Other Authors Contributors Li Zhihong Fernandez Elodie 551 60994 Copyright and Disclaimer 2008 CSIRO and the Bureau of Meteorology To the extent permitted by law all rights are reserved and no part of this publication covered by copyright may be reproduced or copied in any form or by any means except with the written permission of CSIRO and the Bureau of Meteorology CSIRO and the Bureau of Meteorology advise that the information contained in this publication comprises general statements based on scientific research The reader is advised and needs to be aware that such information may be incomplete or unable to be used in any specific situation No reliance or actions must therefore be made on that information without seeking prior expert professional scientific and t
104. ut mostly above 50 in other seasons These seasonal differences are less pronounced for Tmin For rainfall note the different y axis scale compared to the other variables the percentage of observed variance reproduced is much higher mostly between 60 and 100 but exceeds 100 in several instances Again this is a flow on effect of the correction factors applied to the reconstructed series as earlier results showed that without the correction the reproduced inter annual variability was underestimated for rainfall as well Timbal 2004 Finally an important validation step is to analyse the ability of the SDMs to capture observed long term trends This provides confirmation that the large scale predictors capture the forcing that explains local trends and hence gives confidence in the ability of the SDM to reproduce realistic local changes driven by large scale changes in future projections Linear trends were fitted to all stations and regional averages are compared with similar trends from reconstructed series calculated for the length of the series 1958 to 2003 for all predictands apart from pan evaporation 1975 to 2003 Fig B6 Each point on these graphs is a regional average of the available HQ stations thus 24 points four seasons and the six regions are shown for dTmin dTmax and pE and 40 points four seasons and the ten regions for Tmax Tmin and Rain On each graph the line of best fit constrained to intercept with 0 between
105. ws the assigned user number and the run number The second column reminds the user of the downscaling options predictand season and region selected for each individual run The third column shows the model s scenario and station s selected The fourth column allows two actions One is to Show Results discussed in the next section the other is Delete This Run This button allows the user to delete all outputs from a previous run and free space for additional runs Upon completion of the downscaling calculation the GUI has been refreshed to the original i e blank state and is ready for an additional run repeating steps 1 to 4 Results from follow on simulations are shown in the Downscaling Output table alongside previous runs but with a different run number as in Fig 7 after two runs Note There is a limit of five runs kept on the machine at any time Once five runs have been completed the user will have to delete one of the earlier runs before being allowed additional runs b Graphics and Data Display and Download Clicking on the Show Results button in the Downscaling Output table opens a new browser window Fig 8 listing all figures and data files generated by this run The number of outputs depends on the number of stations chosen There is one data file right box per station while graphical outputs left box are presented with eight individual stations per page The number of graphical outputs vary with the p
106. yed underneath the user interface Fig 7 A Downscaling Technique Bone tes Capeleeuwin Cunderdin gt Remove Stn Step 4 Run Downscaling Process Run Downscaling Clear Downscaling Output User No 1 Downscaling User Run Options r Model Scenarios and Stations Results CCM_20C3M CNRM_20C3M CSIRO_20C3M Regon SEA GFDLI_20C3M GFDL2_20C3M GISSR_20C3M Show Results IPSL_20C3M MIROC_20C3M MPI_20C3M w SE MRI_20C3M Portlincoln Snowtown Adelaide Delete This Run cason mummon Nuriootpa Mountgambier Robe Nhill Melbourne Laverton Capeotway CCM_20C3M CNRM_20C3M CSIRO_20C3M GFDL1_20C3M GFDL2_20C3M GISSR_20C3M TPSL_20C3M MIROC_20C3M MPI_20C3M MRI_20C3M Kokardine Manarra Mingenew Shaw Results Regon SWA Perangery Willigulli Boyanup Capenaturaliste Caperiche 1 2 Predictand rain Hopetoun Kendenup Kingriver Pardelup gt Season Summer Peppermintgrove Wilgarrup Cowcowing Cuttening __Delete This Run Doodardingwell Ejanding Meckenng Merredin Nungarin Thegranites Trayning Walk_walkin Codg_codgen Arthurriver Broomehill Corrigin Cranbrook Nyenilup Theoaks Lakecarmody Menzies Norseman Figure 7 The GUI once two downscaling calculations have been performed Both runs are summarised underneath the original interface which has been refreshed and is ready for a new simulation 34 BoM SDM GUI documentation Authors B Timbal et al The first column of the table sho
107. ys is defined as a series of n consecutive days with temperature exceeding a threshold a cold spell is similarly defined as a series of n days with temperatures below a defined threshold The threshold is defined separately for each series observations or downscaled models it is currently defined as the mean plus the variance for a hot spell the mean minus the variance for a cold spell This threshold can be modified in the code depending on user needs The values for the thresholds are shown on the plots for the observed series as well as the lowest GCM and highest GCM thresholds used for the models Return period plots represent the probability of the duration in number of days between two events two hot days for hot day return or two cold days for cold day return Similar thresholds to those for duration are used for return period HSD mildura HDR mildura 1 00 v 1 00 OBS Thre 36 7 OBS Thre 36 7 6 CCM 38 2 GCM 38 2 ns _ 2 e eee GCM 38 1 oso GCM 38 1 gei Pal ni oO o Ta Qa o 0 01 F g 0 01 a f 2 3 4 5 6 Fi 8 10 20 30 40 50 60 Figure F4 Probability of hot spell duration left and hot day return right for maximum temperature in summer in Mildura region SMD for the middle of the 21 century 2046 to 2065 using emission scenario A2 for the CSIRO CCM and CNRM models The x axis represents the number of days the y axis logarithmic scale

The Bureau of Meteorology Statistical Downscaling Model Graphical

Contents

Download Pdf Manuals

Related Search

Related Contents