Effects of different climate downscaling methods on the assessment of climate change impacts on wheat cropping systems

Transcription

1 Supplementary material for Effects of different climate downscaling methods on the assessment of climate change impacts on wheat cropping systems De Li Liu 1 Garry J. O Leary 2 Brendan Christy 3 Ian Macadam 4,5 Bin Wang 1 Muhuddin R Anwar 1 Anna Weeks 3 Corresponding Author: De Li Liu NSW Department of Primary Industries, Wagga Wagga Agricultural Institute, NSW 2650, Australia T: M: F: de.li.liu@dpi.nsw.gov.au S1 Brief description of the four downscaling methods The four downscaling methods used in the study are briefly descripted below. RCS: Raw change scaling method (RCS) is a simple CF scaling method for rapid impact assessment (Diaz-Nieto and Wilby, 2005). We used this to generate daily climate data for Simulated changes in the annual mean climate between a baseline period ( ) and the future period obtained for the climate model gridpoint nearest the site of interest and were used to modify the daily values in the observed historical data series of the site (Arnell and Reynard, 1996). Changes in annual mean daily minimum and maximum temperature are calculated as TG ( C) and the change in annual rainfall is calculated as the percentage change over the baseline period as RG (%). The TG is added to the daily observed daily minimum and maximum temperature, while RG is applied to the daily observed rainfall sequence for DTS: De-trend scaling method (DTS) generates future climate data using a pattern-scaling approach. Following (Suppiah et al., 2007), for each climate variable and GCM, changes in monthly mean values per o C of global mean temperature increase are obtained for each site and scaled by mid-range global warming values for the SRES A2 scenario from the IPCC. Daily data are generated from the monthly means using a method described by Anwar et al. (2007) and modified by (Weeks et al., 2010). A detailed description of the approach is described in S1.1. NWAI-WG: NWAI-WG, developed at NSW DPI s Wagga Wagga Agricultural Institute, is an SDSM based on a modified Richardson-type WGEN weather generator (Liu and Zuo, 2012). In this method, monthly mean GCM data are first spatial interpolated to observing station sites, and then bias corrected against station observations. The resulting monthly data are then disaggregated to daily data using modified WGEN. Daily climate data for for each site were generated by WGEN simulations with parameters derived from the site observations and the bias-corrected monthly GCM data. More detailed descriptions can be found in Liu and Zuo (2012) and S1.2. LARS-WG: LARS-WG is an SDSM based on a series weather generator (Racsko et al., 1991). LARS-WG generates synthetic time series of daily weather including maximum, minimum temperature, rainfall and solar radiation based on semi-empirical distributions of daily climate variables. To generate data for a future period, distributions of daily climate variables are derived by applying simulated changes in the distribution of daily climate data to observed distributions. The distributions are assumed to be stationary over the future period. Hence this study used LARS-WG data for three 20-year periods within the 21 st century. The detailed description of LARS-WG was given in Semenov et al. (1998) and S1.3. S1.1 Description of de-trend scaling method The de-trend scaling method is a pattern-scaling approach (Mitchell, 2003; Watterson, 2005) that scales GCM-simulated changes in climate variables per C of increase in global mean temperature by projected global warming values. Suppiah et al. (2007) used this approach to generate multi-gcm climate projections for Australia, scaling spatial patterns of change for the CMIP3 GCMs Global Warming Factors (GWF) from the IPCC. In this study patterns-of-change per degree of global warming were sourced for each of the GCMs at each for each study site from CSIRO s OzClim website ( The patterns were scaled by the mid-range IPCC GWFs for the SRES A2 emissions scenario (Bernstein et al., 2008). A downscaling technique developed by CSIRO and described Anwar et al. (2007) was then applied. For this analysis fifty years of historical climate data ( ) were detrended then scaled to represent the year The sequence, when run through a crop model, 1

2 generates a distribution of crop response for the year 2050 that accounts for the historical climate variability at that site. The change pattern for a particular GCM for a particular climate variable was prepared by taking a climate change simulation of the GCM and regressing values of the variable at the GCM gridpoints against the global mean temperatures from the same simulation. In this study, we have extracted per C changes in monthly mean climate variables for each study site for each GCM from change patterns sourced from CSIRO s OzClim website and have scaled them by mid-range IPCC GWFs for the SRES A2 emissions scenario. We have then applied a daily downscaling technique developed by CSIRO and described Anwar et al. (2007). To deal with stochastic options that give a realistic analysis representative of any particular future year, e.g. 2050, this method was changed to create a sequence of years, which represents that future year (Weeks et al., 2010). We achieved this by applying the detrended historical data referenced to the base year to the particular year of interest by applying the global warming factor for that year to the whole data sequence so that it is shifted. This means that for particular future year we have a sequence of weather data having the same number of years as the historical data. This method allows the climate change effect to be analysed without confounding between individual years. The details of this approach are described in Weeks et al., 2010 but summarised here for clarity. Daily reference data r t for minimum and maximum temperature, rainfall and solar radiation were defined as the historical climate sequences from 1935 ( y 1) to 1990 ( y 2 ). Daily-monthly reference data r m, t were extracted for a given calendar month m. A linear regression-line T a MA rm, t b was fitted to the mean-annual (MA) daily-monthly reference data versus the projection year. The daily-monthly data were detrended, s r a y y y y (S1) t 1 1 t 2 m, t m, t y where r m, t were the daily-monthly reference data, a, the gradient of the linear regression line T and y t, the reference year. The detrended data were centred around zero, u m, t sm, t sm, t (S2) with s m, t, the mean of the detrended sequence s m, t. A baseline value ( B m ) was calculated for the year 1990 for each calendar month to anchor the projections to the 1990 reference year of the IPCC GWFs. ( y2 y1 B ) m sm, t a (S3) 2 The future maximum and minimum temperature projections were then calculated as a value shifted from the baseline year. xm, t um, t y y Bm Patm GWFy t (S4) 1 2 where um, t y1 y were the daily-monthly detrended data between y 2 1 and y 2, B m, the baseline value calculated for each calendar month, Pat m, the pattern of change value defined at the co-ordinates of the reference data for month m and GWF y t, the global warming factor predicted by the IPCC for the future year y t. Rainfall and radiation were scaled from the baseline year xm, t u m, t y y Bm 1 Patm GWFy t (S5) 1 2 The monthly-daily data were then recombined to form the full daily climate changed sequence where one global warming factor for year y x was applied to the entire detrended trace. The future maximum and minimum temperature projections were calculated as xm, t um, t y y Bm Patm GWFy x, (S6) 1 2 and the rainfall and radiation projections were calculated as xm, t u m, t y y Bm 1 Patm GWFy x. (S7) 1 2 As this method applies the projected pattern of change to a monthly-detrended climate data series, the resultant change at a grid cell predicted by a GCM would differ from the following three methods that applied a pattern of change variable averaged on an annual basis. S1.2. Weather generator based statistical downscaling, NWAI-WG (NWG) NWAI-WG involves spatial and temporal downscaling procedures (Liu and Zuo, 2012). In the spatial downscaling procedure, an inverse distance weighted (IDW) interpolation method is used to downscale the monthly values from the four nearby grid points data to the site. Compared to use of the nearby grid cell method, the obvious advantage of this approach is to remove the repaid changes between grid cells so that it yields more regular isotherms. This method includes a bias correction of the monthly raw GCMs data, where the observed and raw GCM projected monthly values of the historical period are used to establish the relationship using a qqplot technique for adjusting GCMs distribution to match with the observed distribution. The same relationship is applied to adjust the GCM projected future data. Figure 1 shows an example for comparison of GCM projection before bias-correction and after bias correction. 2

3 In the temporal downscaling procedure, the bias-corrected monthly GCM data are disaggregated into daily values by a modified WGEN (Richardson and Wright, 1984) which used serial-correlation matrix A and crosscorrelation matrix B: X ( j) AX 1( j) B ( j), j=1, 2, 3 (S8) i i i where X i (j) is a 3 1 matrix containing the three climate variables of maximum temperature, minimum temperature and radiation for day i. i is a 3 1 vector of independent random components. A and B are 3 3 matrices that are defined by A M M BB 1 T 1 T M0 M1M0 M (S9) where the elements of M0 are the correlation coefficients between these three variables on the same day and those of M1 are the lag-1-day correlation coefficients. Liu and Zuo (2012) derived an analytical solution to calculate the elements of B so that the site- and month-specific parameters can be derived for the downscaling. In this approach, the 120 year observations ( ) are used to sort out into 6 groups (20 years each) according the ranked annual rainfall amounts or annual mean temperature in an ascending order. The six groups of 20 years represent the rainfall condition of the site for driest to wettest climates or coldest to hottest climates so that the site- and monthspecific parameters are used for downscaling the GCM projected future climate to match with the observed climates. S1.3 Weather generator based statistical downscaling, LARS-WG (NWG) LARS-WG uses available observed daily weather for a given site to compute a set of parameters for probability distributions of weather variables as well as correlations between them (Semenov and Stratonovitch, 2010). The occurrence and amount of rainfall are modelled for each day using semi-empirical distributions for the succession and transition of dry and wet days. Other weather variables such as temperature and radiation are conditioned on rainfall distribution (wet and dry day) due to their relation with the amount of cloud cover. In the next step, the baseline distributions of climatic variables are adjusted by changes derived from global or regional climate models (i.e. relative differences for rainfall and radiation, and absolute differences for temperature between future and baseline scenarios) to generate synthetic weather time series of arbitrary length using a pseudo-random number generator for the future. No adjustments are added to the distributions of dry and wet series and to variability of temperature (Semenov and Stratonovitch, 2010). LARS-WG has been demonstrated to perform well in diverse climates for different regions like North America, Europe, Asia and New Zealand and was able to reproduce most of the characteristics of the observed weather data well at each site (Qian et al., 2004; Qian et al., 2005; Semenov, 2007, 2008, 2009; Semenov and Doblas-Reyes, 2007; Zarghami et al., 2011). LARS-WG downscales three 20-year periods including the period , representing 2055, denoted as temperature as T 2055 and rainfall or radiation as R The available LARS-WG data are 5 years later than our proposed period We adjusted this period data for the period 2050: T T2055 (2ay b), y 2051 P SDW,2050 P2050 P2055 (S10) PSDW,2055 where y is the calendar year and a and b are the regression coefficients of the quadratic equation T SDW =ay 2 +by+c is yearly mean minimum or maximum temperature and R SDW is the 20-year mean annual rainfall or mean annual radiation after bias-correction by NWG. The mean corrections over the 7 GCMs for minimum temperature were from 0.22 ± 0.06 C at Moree to 0.15 ± 0.03 C at Walpeup and Birchip and that for maximum temperature were from 0.24 ± 0.05 C at Moree to 0.21 ± 0.05 C at Walpeup and Birchip. The ratio for correction of radiation was 1.00 ± 0.01 across all sites, but that for rainfall was ranged from ± at Moree to ± at Condobolin. Table S1 List of 7 General Climate Models (GCMs) under SRES A2 CO2 emission scenario used in this study GCM name GCM Abbreviation Country Centre code CNRM-CM3 CN France CNRM ECHAM5 EC Germany MPI-M GFDLCM2.1 G2 USA GFDL UKMO-HadCM3 HA UK UKMO UKMO-HadGEM1 HG UK UKMO INM-CM3.0 IN Russia INM IPSL-CM4 IP France IPSL 3

4 Table S2 Comparison of the observed annual mean temperature (AMT, C) in with mean bias and projected change ( C) in the period over the baseline period Coefficient of variation (CV %) for GCM projected AMT in periods and is in brackets next to Bias and, respectively. The GCM projected AMT in the periods and are tested against the observed AMT in the period by two population mean t-test GCM Moree Condobolin Wagga Walpeup Birchip Hamilton Observed annual temperature ( C) Mean (CV) 19.2(2.9) 17.3(3.7) 15.9(3.3) 16.6(2.7) 15.7(2.7) 13.0(4.0) GCM projection CN Bias o C -0.7 *** (5.2) 0.1 (3.3) -0.2 (3.1) 0.1 (2.9) 1.1 *** (2.8) 1.7 *** (2.6) o C 2.3 *** (4.3) 2.1 *** (2.4) 2.0 *** (1.9) 1.6 *** (1.5) 1.7 *** (1.4) 1.4 *** (1.4) EC Bias o C 0.6 *** (4.9) 0.7 *** (4.8) -0.4 * (2.8) 1.4 *** (3.5) 0.6 *** (3.8) 1.4 *** (3.6) o C 2.0 *** (5.3) 2.1 *** (4.5) 1.9 *** (4.2) 1.4 *** (2.8) 1.3 *** (3.2) 1.0 *** (3.7) G2 Bias o C -1.1 *** (5.1) -1.7 *** (6.2) -1.8 *** (6.3) -0.1 (4.8) -0.2 (5.2) 1.7 *** (3.4) o C 2.0 *** (5.1) 1.8 (5.6) 1.7 (5.2) 1.6 *** (3.9) 1.6 *** (4.2) 1.2 *** (2.8) HA Bias o C -3.0 *** (5.0) -0.1 (4.3) 0.0 (4.0) 0.4 *** (3.4) 0.6 *** (3.3) 1.0 *** (3.0) o C 2.0 *** (5.2) 2.0 *** (3.9) 1.9 *** (3.3) 1.9 *** (2.5) 1.9 *** (2.5) 1.8 *** (2.4) HG Bias o C -2.6 *** (3.4) -1.4 *** (3.9) -1.9 *** (3.9) 0.1 (3.0) -0.1 (3.1) 1.0 *** (3.2) o C 1.8 *** (3.5) 1.6 (2.9) 1.6 (2.8) 1.2 *** (1.9) 1.2 *** (2.0) 1.1 *** (2.7) IN Bias o C -3.4 *** (4.2) -3.9 *** (4.6) -4.1 *** (4.6) -4.7 *** (4.5) -4.3 *** (5.0) -1.7 *** (3.5) o C 2.0 *** (2.9) 1.9 *** (3.0) 1.7 *** (3.2) 1.6 *** (3.0) 1.6 *** (3.3) 1.3 *** (2.5) IP Bias o C 0.8 *** (2.9) -1.0 *** (3.2) -1.4 *** (3.4) -1.5 *** (2.9) -2.5 *** (3.1) -0.5 *** (2.7) o C 2.2 *** (2.3) 2.2 *** (3.2) 2.1 *** (3.8) 2.0 *** (3.4) 1.8 *** (4.0) 1.5 *** (3.6) Mean bias( o C) -1.3± ± ± ± ± ±1.3 Mean CV in baseline ( o C) 4.4± ± ± ± ± ±0.4 ±s.d.(%) 2.0± ± ± ± ± ±0.3 Mean CV in 2050 (%) 4.1± ± ± ± ± ±0.8 The GCM Bias and followed by *, **, *** indicate that the population mean of the observed annual mean temperature in the period is statistically different at P<0.05, P<0.01 and P<0.001 level from the population mean of the GCM projected annual mean temperature in the periods and , respectively; otherwise, they are not significant difference (P 0.05). GCM abbreviations and names under italics are the 7 GCM data available from LARS-WG. S2. Bias and GCM projected temperature and rainfall change Seven GCMs (Table S1) applied in this study showed considerable inter-regional differences among various GCMs in the climate values while simulating observed annual mean temperature (AMT) and rainfall (Table S2). A comparison of mean biases and projection change of AMT simulated by GCMs ( and ) with observed values ( ) was shown across the six regions. The GCM bias composed of 20 out of 42 cases showing significant (P<0.05) negative biases (-0.5 to -4.7 C) and 12 out of 42 cases showing significant positive biases (+0.4 to +1.7 C) of AMT in combination GCMs across the regions. However, at Hamilton (a comparatively cooler region), only INM-CM3.0 and IPSL-CM4 models simulated negative biases (-1.7 and -0.5 C), rest all GCM simulated positive AMT biases (+1.0 to +1.7 C) and this illustrate anomalous behaviour of different GCM in a region. The INM-CM3.0 GCM simulated the highest negative biases of observed AMT across all regions and at Walpeup (comparatively warmer region) negative bias reached maximum of -4.7 C, while the UKMO-HadCM3 GCM replicated best (0.0 C bias) observed AMT at Wagga (Table S2 and Fig. 1A&B). The projection change of AMT could increase between 1.0 C and 2.3 C by 2050, although the increasing magnitude of AMT varied among the seven GCM and across the regions (Table S2). UKMO-HadCM3 GCM simulated 1.9 C and 1.8 C AMT change at Birchip and Hamilton, respectively. Although individual GCMs has bias ranged from -4.7 to +1.7 C, the ensemble mean bias of AMT is within -1.4 C and +0.7 C across the regions, highlighting differences in variance and direction of change at locations predicted the GCMs. Although the GCMs ensemble mean biases of annual rainfall (AR) relative to observations is quite small (-59 to +19 mm), however, across all regions, the general pattern of AR biases could be highly variable (-406 mm to +255 mm biases) among GCMs (Table S3 and Fig 1). As an example, the ECHAM50M climate model predicted fairly good agreement (bias = 2 mm) of observed AR at Walpeup, whereas IPSL-CM4 GCM at Moree tends to underestimate (negative bias of -406 mm) AR (Table S3). Projected future (2050) changes of AR also show large variability among GCMs (Table S3). However, ensemble averages suggest an approximately up to 5 ~ 11 % decrease in AR across the regions. The GCMs that simulated with the largest decrease in AR by 2050 are IPSL- CM4 at Walpeup and Birchip (-24 %), at Hamilton (-23 %) and at Wagga (-18 %), and ECHAM50M at Condobolin (-18 %). In contrast, IPSL-CM4 simulated 10 % increase in AR at Moree (Table S3). 4

5 Table S3 Comparison of the observed annual rainfall (AR, mm) in with mean bias (Bias, mm) and projected change ( %) in the period over the baseline period Coefficient of variation (CV %, in bracket) for GCM projected AR in periods and are next to Bias and, respectively. The GCM projected AR in the periods and are tested against the observed AR in the period by two population mean t-test Moree Condobolin Wagga Walpeup Birchip Hamilton Observed annual rainfall (mm) Mean (CV) 592(23) 455(32) 545(29) 336(32) 369(31) 677(18) GCM projection CN Bias (mm) 64 (35) -28 (36) -131 *** (32) -112 *** (27) -120 *** (27) -294 *** (15) % (24) (26) *** (28) *** (29) *** (27) *** (11) EC Bias (mm) -82 (32) 47 (32) 34 (25) 1 (27) 92 *** (20) -107 *** (16) % * (50) (53) (39) -2.4 (43) -6.5 (35) -5.6 *** (24) G2 Bias (mm) -18 (36) 57 (37) 6 (34) -25 (42) 32 (37) -235 *** (26) % (34) -5.6 (37) -5.4 (31) (39) -8.6 (36) *** (29) HA Bias (mm) 243 *** (20) 193 *** (23) 93 * (22) 91 *** (28) 98 *** (25) -116 *** (22) % -3.4 *** (23) -2.1 *** (28) -2.4 (24) -4.9 (29) -5.9 (27) -8.1 *** (24) HG Bias (mm) 255 *** (21) 224 *** (22) 249 *** (21) 89 *** (24) 116 *** (21) -87 *** (19) % -0.4 *** (27) -8.3 *** (27) *** (24) -8.5 (30) -6.6 (27) -2.9 *** (18) IN Bias (mm) 130 *** (20) 224 *** (20) 137 *** (18) 106 *** (19) 138 *** (19) -127 *** (13) % -7.0 (18) -4.6 *** (17) -5.2 (19) (19) -8.2 *** (21) -9.7 *** (11) IP Bias (mm) -406 *** (36) -237 *** (30) -220 *** (22) -43 (22) 139 *** (16) -4 (11) % 10.2 *** (48) *** (29) *** (23) *** (26) (22) *** (17) Mean bias(mm) -59±25-35±14-33±13-7±3 19±9-0±1 Mean CV in baseline (%) 29±8 28±7 25±6 27±7 24±7 17±5 ±s.d.(%) -4.7± ± ± ± ± ±6.8 Mean CV in 2050 (%) 32±13 31±11 27±7 31±8 28±6 19±7 The GCM Bias and follow by *, **, *** indicate that the population mean in the observed annual rainfall in the period is statistically different at P<0.05, P<0.01 and P<0.001 from the population mean in the GCM projected annual rainfall in the periods and , respectively, otherwise, they are not significant difference (P 0.05). GCM abbreviations under italics are the 7 GCM data available from LARS-WG 5

6 Fig. S1A Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual mean temperature (AMT) at Moree and Condobolin. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 6

7 Fig. S1B Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual mean temperature (AMT) at Wagga Wagga and Walpeup. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 7

8 Fig. S1C Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual mean temperature (AMT) at Birchip and Hamilton. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 8

9 Fig. S2A Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual rainfall (AR) at Moree and Condobolin. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 9

10 Fig. S2B Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual rainfall (AR) at Wagga Wagga and Walpeup. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 10

11 Fig. S2C Comparisons between observed (OBS), raw GCM (RAW), NWAI-WG (NWG) and LARS-WG (LWG) datasets in terms of annual rainfall (AR) at Birchip and Hamilton. Raw GCM results are for data for the nearest GCM gridpoint to the sites. The values are Mean ± SDinter_annual ± SDintra_seasonal for the period of A: baseline ( ), B: , C: scaling historical period of for 2050, D: Intra-seasonal standard deviation is calculated using the mean daily values for all days that fall into the cropping growing season (1 May - 30 November) 11

12 Fig. S3 Comparison with the multi-gcms projected change in mature date (MD), crop duration (CD), runoff (RO), deep drainage (DD), potential evapotranspiration (PET) and biomass at flowering (BMF) by the four downscaling methods. The horizontal bars are the range of the 7 GCMs projected change with vertical line showing the multi-gcms projected mean change (positive in green, negative in red) 12

13 13

14 14

15 Fig. S4 The scatter plots and Pearson Correlation Coefficients (PCC) of two paired the changes of climate change impacts on wheat cropping system (SD, sowing date; FD, flowering date; MD, maturity date; CD, crop duration; RO, runoff; DD, deep drainage; PET, potential transpiration; ES, soil water evaporation; EP, plant transpiration; WUE, water use efficiency; BMF, biomass at flowering; CY, crop yield). The Dif is paired difference between two pairings with significant test by paired t-tests showing *, **, and *** for the significant at P<0.05, P<0.01 and P<0.001, respectively. The PCC followed by C, B and A indicates the significant at P<0.05, P<0.01 and P<0.001, respectively 15

16 Table S4 Significant paired t-tests for the differences between the multi-gcm mean changes ( ) in flowering date (FD), crop yield (CY) and water use efficiency (WUE) projected by each two methods of DTS, RCS, NWG and LWG. Spearman rho ( ) tests for the correlation of the changes ( ) between each two methods are shown in the bracket Methods Moree Condobolin Wagga Walpeup Birchip Hamilton Mature date (MD, days) DTS- RCS 3.3 ** (0.47) -1.2(-0.33) -0.5(0.34) -0.3(0.07) 1.6 *** (0.45) 0.5(-0.09) DTS- NWG -3.6(-0.29) -6.8 *** (0.52) -0.7(-0.01) -2.4 *** (0.39) -3.8 *** (0.28) 0.7(0.23) DTS- LWG 11.0 *** (-0.38) 0.1(0.65) -1.0(0.65) 1.7 *** (0.62) 2.6 *** (0.69C) 0.5(0.00) RCS- NWG -6.9 *** (0.33) -5.6 *** (0.16) -0.2(-0.02) -2.0 *** (0.54) -5.4 *** (0.64) 0.1(0.68 C ) RCS- LWG 7.7 *** (-0.25) 1.3(-0.50) -0.5(0.45) 2.0 ** (0.08) 1.0(0.35) 0.0(0.81 C ) NWG- LWG 14.6 *** (0.47) 6.9 *** (0.26) -0.3(0.48) 4.0 *** (0.58) 6.4 *** (-0.04) -0.1(0.27) Crop duration (CD, days) DTS- RCS 0.3(0.47) -3.6 *** (0.61) -2.1(-0.42) -2.6(-0.17) 1.0(0.31) 5.0 *** (0.70 C ) DTS- NWG -5.2 *** (0.88 B ) -3.1(0.93 B ) -5.6 * (0.35) -1.9(0.22) -3.5(0.12) 1.7(0.52) DTS- LWG 2.1 * (0.68 C ) *** (0.28) -7.9 *** (-0.27) -7.0(0.12) -2.1(0.70 C ) -1.0(0.68 C ) RCS- NWG -5.5 *** (0.74 C ) 0.5(0.38) -3.5(0.38) 0.7(0.22) -4.5(0.06) -3.3(0.66) RCS- LWG 1.8(0.62) *** (-0.40) -5.8 *** (0.09) -4.3(0.47) -3.1(0.37) -6.1***(0.89 B ) NWG- LWG 7.3 *** (0.63) *** (0.32) -2.3(0.53) -5.0(0.68 C ) 1.4(0.57) -2.8(0.79 C ) Runoff (RO, mm) DTS- RCS -2.6(-0.03) -0.7(0.18) -1.4(0.10) -0.6(0.25) -0.4(0.39) 0.6(0.63) DTS- NWG 5.0(0.64) 0.9 ** (0.63) 3.3 *** (0.23) -0.1(0.31) 0.2(0.51) -2.2(0.26) DTS- LWG -3.3 *** (0.60) -0.4(-0.19) -2.6(0.00) 0.6 * (0.46) -2.1 *** (0.45) -2.5(0.42) RCS- NWG 7.6 * (-0.10) 1.5 * (0.47) 4.7 *** (0.95 A ) 0.5(0.65) 0.6(-0.02) -2.8(0.06) RCS- LWG -0.7(-0.15) 0.3(-0.76 C ) -1.2(-0.52) 1.2 * (0.29) -1.6(-0.11) -3.1(-0.25) NWG- LWG -8.4 *** (0.41) -1.2(-0.24) -5.9 *** (-0.35) 0.6(0.61) -2.3 *** (0.50) -0.3(0.69 C ) Deep drainage (DD, mm) DTS- RCS 0.0(0.00) -0.7(0.06) -11.4(-0.10) -0.6(-0.27) -1.8(0.18) 0.7(0.34) DTS- NWG 1.0(0.00) -0.9(0.58) -7.5 *** (0.74 C ) -0.5(-0.04) -2.6(0.93 B ) -2.8(0.73 C ) DTS- LWG 0.0(0.00) 0.3 * (0.00) -0.3(0.09) 0.0(0.78 C ) 0.6(0.27) 2.2(0.70 C ) RCS- NWG 1.0(0.00) -0.2(0.05) 3.9(0.38) 0.1(0.36) -0.7(0.10) -3.6(0.03) RCS- LWG 0.0(0.00) 1.0 * (0.00) 11.1(-0.59) 0.6(-0.27) 2.4(-0.57) 1.5(-0.13) NWG- LWG -1.0(0.00) 1.2(0.00) 7.2(-0.25) 0.5(-0.02) 3.1(0.29) 5.0(0.88 B ) Potential evapotranspiration (mm) DTS- RCS 32.4 *** (0.81 C ) 22.4 *** (-0.02) 25.4 *** (-0.91 B ) 5.4(-0.42) 18.7 *** (-0.40) 26.4 *** (-0.14) DTS- NWG *** ( C ) 10.6(-0.42) 0.2(-0.67 C ) -6.0(0.36) 3.0(0.03) -9.6(-0.36) DTS- LWG 21.8 *** (0.28) -6.4(0.38) -0.6(0.41) 9.1(-0.13) 3.3(0.23) 16.1 *** (0.81 C ) RCS- NWG *** (-0.32) *** (-0.75 C ) *** (-0.04) -15.0(-0.26) *** (0.38) *** (0.61) RCS- LWG -10.6(0.38) *** (-0.34) *** (-0.43) 3.8(-0.29) *** (0.05) -10.3(0.08) NWG- LWG 21.7 ** (0.08) -0.4(0.27) -3.6(0.78 C ) 18.8 *** (0.47) 39.7 *** (0.11) 5.4(-0.29) Biomass at flowering (BMF, %) DTS- RCS -1.4(-0.29) * (0.05) -4.8(0.69 C ) ** (0.16) 3.2(0.08) 3.1 * (-0.23) DTS- NWG -8.0(0.25) -8.3(0.45) 9.4(0.21) 0.8(0.17) -4.4(0.49) 5.6 * (-0.44) DTS- LWG 2.1(0.06) * (0.43) -8.3(-0.50) *** (-0.06) -3.5(0.26) 4.4(-0.47) RCS- NWG -6.6(-0.04) 8.0(-0.36) 14.3(-0.02) 17.0 *** (0.97 A ) -7.6(-0.24) 2.5(0.57) RCS- LWG 3.5(0.22) 4.3(-0.79 C ) -3.5(-0.31) -16 * (0.42) -6.7(-0.13) 1.3(0.00) NWG- LWG 10.2(0.35) -3.6(0.42) -17.8(-0.78 C ) *** (0.56) 0.8(-0.06) -1.2(0.11) The differences followed by *, ** and *** are significant at the level of P<0.05, P<0.01 and P<0.001, respectively; otherwise, not significant (P 0.05). # The Spearman rho followed by C, B and A are significant at the level of P<0.05, P<0.01 and P<0.001, respectively; otherwise, not significant (P 0.05) Supplementary references Anwar MR, O leary G, McNeil D, Hossain H, Nelson R (2007) Climate change impact on rainfed wheat in south-eastern Australia. Field Crops Research 104: Arnell N, Reynard N (1996) The effects of climate change due to global warming on river flows in Great Britain. Journal of hydrology 183: Bernstein L, Bosch P, Canziani O, Chen Z, Christ R, Riahi K (2008) IPCC, 2007: climate change 2007: synthesis report. IPCC 16

17 Diaz-Nieto J, Wilby RL (2005) A comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom. Climatic Change 69: Liu DL, Zuo H (2012) Statistical downscaling of daily climate variables for climate change impact assessment over New South Wales, Australia. Climatic change 115: Mitchell TD (2003) Pattern scaling: an examination of the accuracy of the technique for describing future climates. Climatic Change 60: Qian B, Gameda S, Hayhoe H, De Jong R, Bootsma A (2004) Comparison of LARS-WG and AAFC-WG stochastic weather generators for diverse Canadian climates. Climate Research 26: Qian B, Hayhoe H, Gameda S (2005) Evaluation of the stochastic weather generators LARS-WG and AAFC-WG for climate change impact studies. Climate Research 29:3-21 Racsko P, Szeidl L, Semenov M (1991) A serial approach to local stochastic weather models. Ecological modelling 57:27-41 Richardson CW, Wright DA (1984) WGEN: A model for generating daily weather variables. ARS (USA) Semenov MA (2007) Development of high-resolution UKCIP02-based climate change scenarios in the UK. Agricultural and Forest Meteorology 144: Semenov MA (2008) Simulation of extreme weather events by a stochastic weather generator. Climate Research 35: Semenov MA (2009) Impacts of climate change on wheat in England and Wales. Journal of the Royal Society Interface 6: Semenov MA, Brooks RJ, Barrow EM, Richardson CW (1998) Comparison of the WGEN and LARS-WG stochastic weather generators for diverse climates. Climate research 10: Semenov MA, Doblas-Reyes FJ (2007) Utility of dynamical seasonal forecasts in predicting crop yield. Climate Research 34:71-81 Semenov MA, Stratonovitch P (2010) Use of multi-model ensembles from global climate models for assessment of climate change impacts. Climate research (Open Access for articles 4 years old and older) 41:1 Suppiah R, Hennessy K, Whetton P, McInnes K, Macadam I, Bathols J, Ricketts J, Page C (2007) Australian climate change projections derived from simulations performed for the IPCC 4th Assessment Report. Australian Meteorological Magazine 56: Watterson I (2005) Simulated changes due to global warming in the variability of precipitation, and their interpretation using a gamma-distributed stochastic model. Advances in Water Resources 28: Weeks A, Christy B, O Leary G Generating daily future climate scenarios for crop simulation. in "Food Security form Sustainable Agriculture". Proceedings of the 15th ASA Conference, November 2010, Lincoln, New Zealand. Australian Society of Agronomy ( ( Zarghami M, Abdi A, Babaeian I, Hassanzadeh Y, Kanani R (2011) Impacts of climate change on runoffs in East Azerbaijan, Iran. Global and Planetary Change 78: