HYDROLOGICAL PROCESSES Hydrol. Process. 26, 3510–3523 (2012) Published online 24 January 2012 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/hyp.8427 Statistical downscaling of extreme daily precipitation, evaporation, and temperature and construction of future scenarios Tao Yang,1*,† Huihui Li,1 Weiguang Wang,1 Chong-Yu Xu2 and Zhongbo Yu3 1 State Key Laboratory of Hydrology- Water Resources and Hydraulics Engineering, Hohai University, Nanjing 210098, China 2 Department of Geosciences, University of Oslo, P.O. Box 1047, Blindern, 0316 Oslo, Norway 3 Department of Geoscience, University of Nevada Las Vegas, Las Vegas, NV89154, USA Abstract: Generally, the statistical downscaling approaches work less perfectly in reproducing precipitation than temperatures, particularly for the extreme precipitation. This article aimed to testify the capability in downscaling the extreme temperature, evaporation, and precipitation in South China using the statistical downscaling method. Meanwhile, the linkages between the underlying driving forces and the incompetent skills in downscaling precipitation extremes over South China need to be extensively addressed. Toward this end, a statistical downscaling model (SDSM) was built up to construct future scenarios of extreme daily temperature, pan evaporation, and precipitation. The model was thereafter applied to project climate extremes in the Dongjiang River basin in the 21st century from the HadCM3 (Hadley Centre Coupled Model version 3) model under A2 and B2 emission scenarios. The results showed that: (1) The SDSM generally performed fairly well in reproducing the extreme temperature. For the extreme precipitation, the performance of the model was less satisfactory than temperature and evaporation. (2) Both A2 and B2 scenarios projected increases in temperature extremes in all seasons; however, the projections of change in precipitation and evaporation extremes were not consistent with temperature extremes. (3) Skills of SDSM to reproduce the extreme precipitation were very limited. This was partly due to the high randomicity and nonlinearity dominated in extreme precipitation process over the Dongjiang River basin. In pre-flood seasons (April to June), the mixing of the dry and cold air originated from northern China and the moist warm air releases excessive rainstorms to this basin, while in post-flood seasons (July to October), the intensive rainstorms are triggered by the tropical system dominated in South China. These unique characteristics collectively account for the incompetent skills of SDSM in reproducing precipitation extremes in South China. Copyright © 2011 John Wiley & Sons, Ltd. KEY WORDS climate extremes; statistical downscaling; climate change; projection; scenarios Received 16 August 2011; Accepted 10 November 2011 INTRODUCTION The frequent occurrence of extreme weather events such as heat waves and intense and persistent precipitation associated with subsequent flooding have raised concerns that human activity might have caused an alteration of the climate system (Yang et al., 2008), which is believed to be the culprit behind the severity of such events. There is also a widespread belief that the climate system will continue to change under the prevailing human activity and that humanity will be faced with more of these extreme events (Hundecha and Bardossy, 2008; Yang et al., 2011). This leads to the growing concerns and studies on changes in frequency, intensity, and/or magnitude of such events in the past and for estimating climate that will occur in the future. *Correspondence to: Dr. Tao Yang, Professor, State Key Laboratory of Hydrology-Water Resources and Hydraulics Engineering, Hohai University, Nanjing 210098, The People’s Republic of China. E-mail: yang.tao@ms.xjb.ac.cn † Present address: State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi, China. Copyright © 2011 John Wiley & Sons, Ltd. General circulation models (GCMs) and large-scale circulation predictors are the most important and effective tools and indicators for the climate impact study. These numerical coupled models represent various earth systems including the atmosphere, oceans, land surface, and seaice and offer considerable potential for the study of climate change and variability. Over the past decade, the sophistication of such models has increased, and their ability to simulate present and past global and continental scale climates has substantially improved. However, the resolution of GCMs remains relatively coarse and does not provide a direct estimation of hydrological responses to climate change. For example, the Hadley Centre’s Hadcm3 model is resolved at a spatial resolution of 2.5 latitude by 3.75 longitude, whereas a spatial resolution of 0.125 latitude and longitude is required by hydrologic simulations of monthly flow in mountainous catchment (Wilby et al., 2004). In other words, GCMs provide output at nodes of grid-boxes, which are tens of thousands of square kilometers in size, whereas the scale of interest to hydrologists is of the order of a few hundred square kilometers. Bridging the gap between the resolution of climate models and regional- and local-scale processes STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES represents a considerable problem for the climate change studies including the application of climate change scenarios to hydrological models. Thus, considerable effort in the climate community has focused on the development of techniques to bridge the gap, known as ‘downscaling’. More recently, downscaling has found wide application in hydroclimatology for scenario construction and simulation of (1) regional precipitation (Kim et al., 2004; Wang et al., 2011); (2) low-frequency rainfall events (Wilby, 1998) (3) mean, minimum, and maximum air temperature (Kettle and Thompson, 2004); (4) soil moisture (Jasper et al., 2004); (5) runoff (Arnell et al., 2003) and streamflows (Cannon and Whitfield, 2002); (6) ground water levels (Bouraoui et al., 1999); (7) transpiration (Misson et al., 2002), wind speed (Faucher et al., 1999), and potential evaporation rates (Weisse and Oestreicher, 2001); (8) soil erosion and crop yield (Zhang et al., 2004); (9) landslide occurrence (Buma and Dehn, 2000), and (10) water quality (Hassan et al., 1998). Downscaling methods could be broadly classified into two categories (Xu, 1999): dynamic downscaling and statistical downscaling. Both techniques have their strengths and weaknesses. In dynamic downscaling, the GCM outputs are used as boundary conditions to drive a Regional Climate Model (RCM) or Limited Area Model and produce regional-scale information up to 5–50 km. This method has superior capability in complex terrain or with changed land cover. However, this method entails higher computation cost and relies strongly on the boundary conditions provided by GCMs with considerable uncertainties. In contrast, statistical downscaling gains local or station-scale meteorological time series (predictands) by appropriate statistical or empirical relationships with surface or troposphere atmospheric features. Generally, statistical downscaling methods can deliver ensembles of daily climate that evolve in line with the large-scale, transient changes of the host GCM. Moreover, given the advantages of being computationally inexpensive, statistical downscaling method can access finer scales than dynamical methods and relatively easily applied to different GCMs, parameters and regions (Wilby et al., 2004). Therefore, it has been widely employed in climate impact studies. However, statistical downscaling approaches need much longer historical time series to build the appropriate statistical relationship. In addition, one of the assumptions of statistical downscaling is still valid in the future. This assumption cannot be testified at present. The conclusion from the most recent studies is achieved in the statistical and regional dynamical downscaling of extremes project (STARDEX, http://www.cru.uea.ac.uk/projects/stardex) that both statistical and dynamical downscaling techniques are comparable for simulating current climate (Haylock et al., 2006; Schmidli et al., 2006). The statistical downscaling has been widely employed in climate change impact assessments (Wilby et al., 1999; Huth, 2002; Tripathi et al., 2006; Ghosh and Mujumdar, 2008), due to its low expenditure on usage and the equivalent power as dynamic downscaling. Copyright © 2011 John Wiley & Sons, Ltd. 3511 In Wilby and Wigley’s study (2000), statistical downscaling techniques are described as three categories, namely: regression methods (e.g. Kim et al., 1984; Wigley et al., 1990; Storch et al., 1993); weather patternbased approaches (e.g. Lamb, 1972; Hay et al., 1991; Bardossy and Plate, 1992); and stochastic weather generators (Katz, 1996). No matter whether the method is simple or complex, it is always based on some kind of a regression relationship. The statistical downscaling model (SDSM) is best described as a hybrid of stochastic weather generator and regression-based methods (Wilby et al., 2002). Many comparative studies (Wilby et al., 1998; Dibike and Coulibaly, 2005) have shown that it has superior capability to capture local-scale climate variability and is, therefore, widely applied (Wilby and Harris, 2006). General practices in downscaling of monthly outputs from a full range of GCMs were presented as above in past years. However, research in constructing reliable scenarios of future climate extremes is still a challenge and inadequate so far (e.g. Wilby and Harris, 2006). Moreover, SDSM normally works worse in subtropical and tropical regions than in inland regions for that precipitation in subtropical and tropical regions always presents more than one flood season due to the effect of tropical cyclones, which are difficult to capture. Therefore, the main objective of the present study is to testify the capability of SDSM in downscaling extreme events in temperature, evaporation, and precipitation in the subtropical region in southern China and, if it is successful, to project their future patterns for the study region. This study strives to downscale extremes of temperature, evaporation, and precipitation in the study region, more importantly to identify the possible links between the underlying driving forces and skills in downscaling precipitation extremes in subtropical regions. It will contribute to promote current downscaling knowledge in similar subtropical regions of the world. STUDY AREA AND DATA Study area Dongjiang River is located between 114.0 ~ 116.5 E and 22.5 ~ 25.5 N (Figure 1). It has a 562 km long mainstream to the Boluo station with a drainage area of 25,555 km2. The Dongjiang River is important not only for the local region but also for Hong Kong because about 80% of Hong Kong’s water supply comes from Dongjiang River through cross-basin water transfer. Three major reservoirs (i.e. Xingfengjiang Reservoir since 1959, Fengshuba Reservoir since 1973, and Baipenzhu Reservoir since 1984) were built in the basin. Annual average air temperature is about 20.4 C. The precipitation of Dongjiang River demonstrates strong seasonality due to a subtropical monsoon climate. Owing to the influence of typhoons, precipitation exhibits strong variability in both spatial and temporal perspective. The annual precipitation varies between 1500 mm and 2400 mm. Hydrol. Process. 26, 3510–3523 (2012) 3512 Xun wu Riv er 25° N T. YANG ET AL. ng Xingfengjiang Reservior R. 24° N gR fen ia gj n Do Heyuan R. ng xia Qiu hu gz en ior p i Ba serv Re . R i Xizh Boluo Streamflow gauges Reservior 23° N Xin latitude° (N) Fengshuba ive Reservior r Li R Shenzhen 114° E 115° E 116° E longtitude° (E) Figure 1. Map of Dongjiang river basin More than 80% of the total annual precipitation falls in the flood seasons from April to September. Data Observed data sets . Measured daily maximum temperature, minimum temperature, pan evaporation, and precipitation were provided by China Meteorological Administration for 41-year period 1961–2001 at five weather stations (Table I). The areal weights of five stations were calculated using the Thiessen polygons method (Figure 2). Reanalysis predictor sets used in calibration. Twentysix different large-scale atmospheric variables derived from the daily reanalysis dataset of NCEP/NCAR in the period of 1961–2001 were used to calibrate and validate the SDSM model, which were downloaded freely from the internet sites at a scale of 3.75 2.5 (http://www.cics. uvic.ca/scenarios/sdsm/select.cgi). The geographical extent (112.5–116.25 N, 22.5–25 E) was chosen to cover the whole area with noticeable influence on the circulation patterns that govern the weather pattern observed over the Dongjiang River basin. Table I. Basic information of the five meteorological stations in the study region No 1 2 3 4 5 ID 59096 59102 59293 59298 59493 Station Lianping Xunwu Heyuan Huiyang Shenzhen Latitude(N) Longitude(E) Areal weight 24 22’ 24 57’ 23 48’ 23 05’ 22 32’ 114 29’ 115 39’ 114 44’ 114 25’ 114 00’ Copyright © 2011 John Wiley & Sons, Ltd. 0.215 0.201 0.332 0.047 0.205 Figure 2. The study area divided by the method of Thiessen polygons GCM predictor sets used in hindcast and projection. The validated SDSM was used to downscale the large-scale predictor variables derived from A2 and B2 scenarios of HadCM3 (Hadley Centre Coupled Model version 3) in the period of 1961–2099. Both scenarios are characterized by a continuously increasing global population with a consequent increase in the emission of greenhouse gas and with a higher rate in A2 than in B2. Maximum temperature, minimum temperature, pan evaporation, and precipitation were simulated during the following periods: the current (1961–2001), 2020s (2010–2039), 2050s (2040–2069), and 2080s (2070–2099). METHODOLOGY Downscaling method The SDSM, developed by Wilby et al. (2002), is employed in this study to build statistical relationships between GCM predictors and local climate variables. The software tool for SDSM is available from the internet site: http://www.cisc.uvic.ca/scenarios/index.cgi?More_ Info-Downscaling-Tool. The regional climate variables conditioned by the large-scale state may be written as: R ¼ F ðLÞ (1) in which R is the predictand (a local climate variable), L is the predictor (a set of large-scale climate variables), and F a deterministic/stochastic function conditioned by L and has to be estimated empirically from historical observations. Three implicit assumptions are made in order to use this kind of downscaling methods for assessing regional climate change: (1) the predictors are variables of relevance and are realistically simulated by the GCM; (2) the predictors employed fully represent the climate change signal; and (3) the relationship is valid also under altered climate condition. Predictor selecting. The climate system is influenced by the combined action of multiple atmospheric variables in the wide tempo-spatial space. Therefore, any single circulation Hydrol. Process. 26, 3510–3523 (2012) 3513 STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES predictor and/or small tempo-spatial space are unlikely to be sufficient, as they fail to capture key precipitation mechanisms based on thermodynamics and vapor content (Wilby, 1998). Wilby and Wigley (2000) found that in many cases, maximum correlations between precipitation and the circulation predictors occurred away from the location of the grid-box of the downscaled station and suggested that selection of predictor domain was a critical factor affecting the realisation and stability of downscaling model. The climate in many zones of China is strongly controlled by the East Asian monsoon, where the atmospheric circulation feature is quite different between winter and summer, and the scale of circulation pattern is large. Thus, it is a big challenge to choose predictors in the wide tempospatial space (Samel et al., 1999). The procedure adopted in the study for selecting suitable predictors for each predictand is as follows: Table II First, all of the 26 atmospheric variables in each one of four grid-boxes (covering the whole study area and surrounding) were taken as potential predictors. Second, these variables were then screened by SDSM to determine what amount explained variance is when the predictand and predictor(s) were statistically compared. The user was required to select predictors that produce the highest explained variance (E) and lowest standard error (SE). Finally, the predictors identified in this study were summarized in Table III. It was shown that different atmospheric predictors control different local variables: the maximum and minimum temperature are more sensitive to mean temperature at 2 m, and 850-hPa geopotential height, mean sea level pressure, and 500-hPa geopotential height are more sensitive predictors for the pan evaporation. For the daily precipitation, the relative humidity at 500 hPa and surface relative humidity are the most sensitive factors. Calibration and validation of SDSM. Before downscaling of future climate with GCM predictors, the relationship between the selected predictors and precipitation in Table III. Selected predictor variables for Dongjiang river basin downscaling Predictands Predictors Tmax Tmin Pcpn Eva 1. Mslp 2. p__u 3. p__v 4. p__z 5. p500 6. p850 7. temp 8. p5zh 9. p5th 10. rhum 11. shum 12. r500 13. r850 14. p5_v 15. p5_z √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ Where: mslp = mean sea level pressure; p__u = zonal velocity component @ surface; p__v = meridional velocity component @ surface; p__z = vorticity @ surfacep500 = 500 hPa geopotential height p850 = 850 hPa geopotential height p5th = 500 hPa wind direction; rhum = surface relative humidity; shum = surface specific humidity; r500 = relative humidity at 500 hPa; r850 = relative humidity at 850 hPa; p5_v = 500 hPa zonal wind; p5_z = Vorticity at 500 hPa; Pcpn= daily precipitation; Eva = daily evaporation; Tmax-= daily maximumoftemperature; Tmin- = daily minimum of temperature. all stations need to be calibrated by using NCEP/NCAR predictors. From the 41 years of data representing present-day climate (1961–2001), the first 30 years (1961–1990) are used for calibrating the regression model, while the rest 11 years of data (1991–2001) are used to validate the model. Measures of performance assessment Four different measures were used to evaluate the performance of the model: the coefficient of efficiency (Ens), coefficient of determination (R2), ratio of simulated and observed standard deviation (RS), and model biases. Table II. Extreme indices for temperature, pan evaporation, and precipitation Precipitation-related indices Pav Pnl90 Px1d Px5d Pxcdd Pq90 Temperature-related indices Txx Txn Txq90 Tnx Tnn Tnq10 Pan evaporation-related indices Ex1d Ex3d Ex5d Ex7d Copyright © 2011 John Wiley & Sons, Ltd. Mean of daily precipitation on all days [mm/day] Number of events > long-term 90th percentile The maximum of daily precipitation in given period [mm] Maximum total precipitation from any consecutive 5 days [mm] Maximum number of consecutive dry days [day] Empirical 90% quantile of precipitation [mm] The maximum of daily maximum temperature [ C] The minimum of daily maximum temperature [ C] Empirical 90% quantile of the daily maximum temperature [ C] The maximum of daily minimum temperature [ C] The minimum of daily minimum temperature [ C] Empirical 10% quantile of the daily minimum temperature [ C] The maximum of daily pan evaporation [mm] Maximum total evaporation from any consecutive 3 days [mm] Maximum total evaporation from any consecutive 5 days [mm] Maximum total evaporation from any consecutive 7 days [mm] Hydrol. Process. 26, 3510–3523 (2012) 3514 T. YANG ET AL. The coefficient of efficiency (Ens) describes how well the volume and timing of the calibrated predictand compares to the observed predictand and is defined by Pn ðOi Si Þ2 Ens ¼ 1 Pi¼1 (2) n 2 i¼1 ðOi OÞ in which ¼1 O n Xn O i-1 i (3) Where n is the number of time steps, Oi is the observed predictand at time step i, and Si is the simulated predictand at time step i. Coefficient of determination R2 measures the amount of variation of a dependent variable that is explained by variation in the independent. The closer the values of Ens and R2 equal to 1, the more successful the model calibration/validation is. The ratio of standard deviation of the modelled and observed indices describes the degree of dispersion of variables (Hundecha and Bardossy, 2008): RS ¼ Table IV. Performance assessment for predictands in calibration and validation Items Daily maximum temperature Daily minimum temperature Daily pan evaporation Daily precipitation Periods Ens R2 bias RS Calibration Validation Calibration Validation Calibration Validation Calibration Validation 0.90 0.90 0.94 0.94 0.65 0.61 0.50 0.48 0.90 0.90 0.98 0.94 0.65 0.65 0.50 0.48 0 1.1 0 0.12 0 0.42 0.39 0.30 0.93 0.93 0.97 0.98 0.77 0.83 0.67 0.66 bias ¼ 32 28 1 Xn ð S Oi Þ i¼1 i n (5) obs ncep A2 B2 35 30 25 Txn( C) Txx( C) 36 (4) Where Ssim is the standard deviation of the modeled indices and Sobs is the standard deviation of the observed indices. Model bias describes the amount of system deviation, which is defined by obs ncep A2 B2 40 Ssim Sobs 20 15 10 24 20 5 0 2 4 6 8 10 0 12 0 2 4 month obs ncep A2 B2 30 8 10 12 20 obs ncep A2 B2 25 20 Tnn( C) 25 Tnx( C) 6 month 15 10 5 15 0 10 0 2 4 6 8 10 -5 12 0 2 4 month obs ncep A2 B2 40 8 10 12 30 obs ncep A2 B2 30 25 Tnq10( C) 35 Txq90( C) 6 month 20 15 10 25 5 20 0 2 4 6 month 8 10 12 0 0 2 4 6 8 10 12 month Figure 3. Comparison of the indices of extreme temperature from observed data and simulated by SDSM driven by NCEP and H3A2 and H3B2 scenarios in validation period Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 26, 3510–3523 (2012) 3515 STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES Indices of extreme climate predictands Changes in extremes of climate events have received increased attention in the last years (IPCC, 2007). Since the early 1990s, it has been known that the largest changes in the climate under enhanced greenhouse conditions were likely to be seen in changes of extremes (Gordon et al., 1992). Kunkel et al. (1999) reported that potential changes in extreme events can generate greater impact on human activities and natural environment than mean climate changes. The select implementation of indices to describe extreme climate events should have several characteristics: relevant, easy to interpret, understandable for policy makers, and covering both frequency and intensity description of extreme processes comprehensively. The core indices of climate extremes recommended by STARDEX Project funded by the European Commission under the Fifth Framework Programme (FP5) (STARDEX, 2001) were used in this study. These core indices were shown in Table II. It should be noted that index of mean precipitation is also included in the list. The indices were used to examine the skills of the downscaling method in constructing scenarios for both climate extremes and means. RESULTS Model calibration and validation The calibration (1961–1990) and validation results (1991–2001) were shown in Table IV. It could be seen that both the simulated maximum and minimum temperatures were closely consistent with observations. R2, Ens, and RS between simulated and observed temperature exceeded or equaled to 0.9 in calibration and validation. The simulation of daily pan evaporation was less satisfactory (Ens and R2 were between 0.61 and 0.65). As for daily precipitation, Ens and R2 values for the downscaled precipitation were about 0.5, much lower than daily temperature and pan evaporation. The biases for the maximum temperature, minimum temperature, pan evaporation, and precipitation were 1.1 C, 0.12 C, 0.42 mm/ day, and 0.39 mm/day in validation. In summary, those biases were acceptable for practical uses. The statistical model built using SDSM is capable of reproducing daily climate variables. Inter-comparison of extreme indices of downscaling for the calibration and validation period Temperature. Generally, the performance of a downscaling model in constructing temperature indices is better than the performance of precipitation indices. It was shown (Figure 3) that the pattern of seasonal variations of temperature was well downscaled with all three datasets (NCEP/NCAR, H3A2, H3B2). In simulating the maximum of daily maximum temperature (Txx) and empirical 90% quantile of the daily maximum temperature (Txq90), the results from NCEP/NCAR were systematically lower than observations in all seasons, while the simuCopyright © 2011 John Wiley & Sons, Ltd. lations from the H3A2 and H3B2 were closer to observations. For the other four indices (the minimum of daily maximum temperature, Txn; maximum of daily minimum temperature, Tnx; minimum of daily minimum temperature, Tnn; and empirical 10% quantile of the daily minimum temperature, Tnq10, Table II), the results from NCEP/NCAR were relatively satisfactory. Tnx was underestimated in summer and winter; instead, the minimum of daily maximum temperature (Txn) from the H3A2 and H3B2 were 6 C overestimated in summer. As for Tnq10, the results from all three datasets were consistent with the observations, while H3B2 provided a worst performance for Tnn. Table V summarized the coefficient of efficiency (Ens), coefficient of determination (R2), ratio of standard deviation (RS), and biases between the 16 downscaled and observed indices. Pan evaporation. The performance for pan evaporation downscaling was less satisfactory than daily temperature. The results for daily pan evaporation are provided by Figure 4. It can be seen that in simulating these four indices (Ex1d, Ex3d, Ex5d and Ex7d, 1 Table II), all the simulated results were lower than observations in September. In general, the seasonal patterns were well simulated, while the simulated magnitude was less satisfactory. Table V. Comparison of the extreme indices between observed and simulated results during calibration (1961–1990) and validation (1991–2001) periods based on NCEP predictors Indices Periods Ens R2 1. Txx Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation Calibration Validation 0.81 0.82 0.88 0.92 0.91 0.87 0.86 0.85 0.96 0.97 0.97 0.97 0.40 0.69 0.57 0.76 0.66 0.79 0.73 0.81 0.82 0.81 0.49 0.55 0.2 0.06 0.62 0.57 0.35 0.12 0.62 0.67 0.93 0.93 0.94 0.95 0.96 0.95 0.95 0.93 0.97 0.98 0.98 0.98 0.67 0.74 0.77 0.77 0.80 0.79 0.83 0.81 0.83 0.82 0.71 0.71 0.47 0.40 0.67 0.69 0.73 0.61 0.67 0.75 2. Txn 3. txq90 4. Tnx 5. Tnn 6. tnq10 7. Ex1d 8. Ex3d 9. Ex5d 10. Ex7d 11. Pav 12. pnl90 13. px1d 14. px5d 15. Pxcdd 16. pq90 bias 1.25 2.40 1.70 1.17 0.98 2.12 1.03 1.04 0.77 0.45 0.64 0.52 0.97 0.23 2.08 0.29 2.86 0.18 3.35 0.24 0.39 0.29 0.03 0.03 13.45 14.87 12.85 16.6 4.29 3.54 2.28 2.51 RS 1.08 1.09 1.03 1.01 1.04 1.08 1.11 1.08 1.02 1.07 0.99 1.00 0.98 1.00 0.91 0.95 0.88 0.93 0.86 0.92 0.98 0.89 1.32 1.24 0.61 0.56 0.76 0.69 0.87 1.03 0.68 0.68 Hydrol. Process. 26, 3510–3523 (2012) 3516 T. YANG ET AL. obs ncep A2 B2 12 45 Ex5d(mm) Ex1d(mm) 10 8 obs ncep A2 B2 50 40 35 30 6 25 4 0 2 4 6 8 10 20 12 0 2 4 month obs ncep A2 B2 35 Ex7d(mm) Ex3d(mm) 30 25 20 15 10 0 2 4 6 6 8 10 12 month 8 10 12 65 60 55 50 45 40 35 30 25 20 obs ncep A2 B2 0 2 4 6 8 10 12 month month Figure 4. Comparison of the indices of extreme pan evaporation from observed data and simulated by SDSM driven by NCEP and H3A2 and H3B2 scenarios in validation period obs ncep A2 B2 14 10 100 pnl90(day) pav(mm/day) 12 8 6 4 80 60 40 20 2 0 obs ncep A2 B2 120 0 0 2 4 6 8 10 12 0 2 4 obs ncep A2 B2 150 8 10 12 100 50 obs ncep A2 B2 250 200 px5d(mm) px1d(mm) 6 month month 150 100 50 0 0 2 4 6 8 10 0 12 0 2 4 obs ncep A2 B2 35 30 8 10 12 25 20 15 10 obs ncep A2 B2 30 25 p90(mm) pxcdd(day) 6 month month 20 15 10 5 5 0 0 2 4 6 month 8 10 12 0 0 2 4 6 8 10 12 month Figure 5. Comparison of the indices of extreme precipitation from observed data and simulated by SDSM driven by NCEP and H3A2 and H3B2 scenarios in validation period Copyright © 2011 John Wiley & Sons, Ltd. Hydrol. Process. 26, 3510–3523 (2012) STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES A2 scenario 7 6 5 4 3 2 1 0 -1 -2 2050s Winter Spring 2020s 2020s Summer 2050s Autumn 2080s 5 5 4 4 3 2 2050s Spring 2080s Summer 2050s Autumn 2080s 3 2 1 0 0 -1 Winter 2020s 6 1 Winter Spring 2020s Summer Autumn 2050s -1 2080s 5 4 4 3 2 1 Winter 2020s 5 txq90 (°C) txq90 (°C) 7 6 5 4 3 2 1 0 -1 -2 6 TXn (°C) TXn (°C) B2 scenario 2080s TXx (°C) TXx (°C) 2020s 3517 Spring Summer 2050s Autumn 2080s 3 2 1 0 Winter Spring Summer Autumn 0 Winter Spring Summer Autumn Figure 6. Changes (%) in extreme temperature between the period (1961–1990) and the period (2011–2099) under the H3A2 and H3B2 scenarios Precipitation. Among the six indices in simulating precipitation extremes, four of them are associated with extreme wet events: 90th percentile (pq90), maximum of daily precipitation (px1d), maximum 5-day total (px5d), and number of heavy events (pnl90). The maximum number of consecutive dry days (pxcdd) describes very dry events, and mean of daily precipitation on all days (pav) describes changes of mean daily precipitation. The threshold of 1 mm was used for a wet day (Hennessy et al., 1999). A dry day was defined as having less than 1-mm precipitation. The calibration and validation results from NCEP/ NCAR were shown in Table V. It indicated that the indices were not equally well modeled. Pav has the highest performance (Ens > 0.8), while px1d (Ens< 0.3) and pxcdd (Ens < 0.4) were the worst reproduced indices, implying that the model still cannot fully capture the true persistence of the precipitation occurrence process. Monthly precipitation can be better downscaled by SDSM than the extreme precipitation. In general, the model could simulate most indices well, but the capability in simulating heavy rainfall under abnormal climate and the persistence of the precipitation occurrence was still limited. The inter-comparison between the simulated and observed six indices in the validation period was shown in Figure 5. As for p90, px5d, and px1d, the simulations were generally underestimated, and the underestimation was rather obvious in summer under H3A2 and H3B2 Copyright © 2011 John Wiley & Sons, Ltd. scenarios. Underestimation of extremes to some extent can be attributed to the short validation period which is heavily influenced by some extreme events with very high return period. For instance, the underestimation of px1d and px5d in April was because Huiyang, Heyuan, and Shenzhen stations had recorded rain as high as 146.7, 133.6, and 344 mm/day on 14 April 2000. The return period of the rainfall total in April in Shenzhen was estimated to be 100 years approximately. Since the validation period only had 10 years, the simulation could not accurately capture some abnormal and extreme storms. Although the pxcdd was underestimated using the NCEP/NCAR in most seasons, the trend and variability were well simulated. It should be noted that the results from H3A2 and H3B2 were less satisfactory compared with the NCEP/NCAR data especially for px1d and pxcdd. In summary, the simulation results from NCEP/NCAR data were closer to the observations than the results from H3A2 and H3B2. Projected changes for future climate scenarios 1. Temperature Changes in extreme temperature between the baseline period (1961–1990) and the future period (2011–2099) were shown in Figure 6. Under the H3A2 scenario, all six Hydrol. Process. 26, 3510–3523 (2012) 3518 T. YANG ET AL. 2050s 2020s 2080s 5 4 4 TNx (°C) TNx (°C) 2020s 5 3 2 Winter Spring 2020s Summer 2050s 2 5 4 4 3 2 5 Summer 2050s Autumn 2080s 3 2 1 Winter Spring 2020s Summer 2050s 0 Autumn 2080s 5 Winter 2020s Spring Summer 2050s Autumn 2080s 4 tnq10 (°C) 4 3 2 1 0 Spring 6 5 0 Winter 2020s 2080s 1 tnq10 (°C) 3 0 Autumn TNn (°C) TNn (°C) 6 2080s 1 1 0 2050s 3 2 1 Winter Spring Summer Autumn 0 Winter Spring Summer Autumn Figure 6. (Continued ) temperature indices will increase in future 90 years. Txx (6.2 C) and Tnx (4.8 C) showed the highest increase in summer, while Txn (5.5 C) and Tnn (4.9 C) increase most considerably in spring. Txq90 and tnq10 will increase with similar magnitude during different seasons. Under H3B2 scenario, the projected Txx (in 2020s and 2050s) and Txn (in 2050s) will decrease slightly in spring, while the other four indices (Txq90, Tnx, Tnn, and Tnq10) showed upward trends. Therefore, the extreme temperature events will be more frequent in the future. 2. Pan evaporation Figure 7 showed that all the indices of pan evaporation in H3A2 and H3B2 scenario would increase by 10% (in 2020s) and 40% (in 2080s) in summer. However, the change trends of H3A2 and H3B2 projections are opposite in winter: the projections from H3A2 scenario are decreasing while a slight increase was projected from H3B2 scenario. Ex3d, Ex5d, and Ex7d would decrease in spring during 2020s, but they would increase during 2050s and 2080s under H3A2 scenario. Under the H3B2 scenario, they will decrease by 5% during 2020s and 2050s and increase by 2% to 12% in 2080s. 3. Precipitation The projected changes of precipitation extremes (Figure 8) were inconsistent with temperature extremes. It can be Copyright © 2011 John Wiley & Sons, Ltd. seen that under H3A2 scenario, the pav and p90 would decrease in winter and spring and increase in summer and autumn, while in H3B2, they showed decreasing trend only in winter. As for pnl90, the number of events higher than long-term 90th percentile will decrease in winter and spring and increase in summer and autumn, and this is more obvious under H3A2 scenario. Projection of pxcdd under H3A2 scenario showed considerable increases only in winter. Under H3B2 scenario, pxcdd showed increases in all seasons. For the px1d and px5d, the results of H3A2 had distinct change patterns in different seasons and periods. In the future, the maximum daily precipitation (px1d) and the cumulative 5-day total precipitation (px5d) under H3B2 scenario will increase. DISCUSSION In this section, we attempt to identify the linkages between the underlying driving forces and skill scores in downscaling precipitation extremes over the Dongjiang basin. During the calibration and validation of SDSM with the NCEP/NCAR reanalysis data, the temperature indices were downscaled rather perfectly, but SDSM was not very effective in downscaling precipitation extremes. This can be attributed to the reasons below. Dongjiang River basin located in southern China suffers frequent rainstorms, and the major driving forces are more complicated than in other inland regions Hydrol. Process. 26, 3510–3523 (2012) STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES A2 scenario Winter 2050s Spring 2020s Winter Spring 2020s Winter 50 40 30 20 10 0 -10 -20 -30 -40 60 50 40 30 20 10 0 -10 -20 -30 -40 Summer Autumn 60 50 40 30 20 10 0 -10 -20 -30 -40 Winter 2020s 2080s Summer Autumn Winter 2020s 2080s 2050s Spring 50 40 30 20 10 0 -10 -20 -30 -40 2080s Summer Autumn 2050s Ex1d (%) 60 50 40 30 20 10 0 -10 -20 -30 -40 2020s Summer Autumn 2020s Ex3d (%) Ex5d (%) 60 50 40 30 20 10 0 -10 -20 -30 -40 Spring B2 scenario 2080s Ex5d (%) Ex3d (%) 50 40 30 20 10 0 -10 -20 -30 -40 Winter 2050s Ex7d (%) Ex1d (%) 50 40 30 20 10 0 -10 -20 -30 -40 Ex7d (%) 2020s 3519 Winter 2020s Winter 2050s Spring Summer Autumn 2050s Spring 2080s Summer Autumn 2050s Spring 2080s Summer Autumn 2050s Spring 2080s 2080s Summer Autumn Figure 7. Changes (%) in extreme evaporation between the period (1961–1990) and the period (2011–2099) under the H3A2 and H3B2 scenarios (See Fig. 9). Hereby, the flood season (April to October) was divided into pre-flood and post-flood seasons for sake of discussion. The pre-flood season (April to June) in South China is composed of the frontal precipitation period and the summer monsoon precipitation period (Qiao et al., 2010). In pre-flood season, the main atmospheric general circulation system dominated in middle high latitude of the Eurasia is two-trough and one-ridge, which help cold air move toward the South China. The Western Pacific Subtropical High was stable at 18 N, which creates favorable conditions for the prevailing of the southerly airstream in South China and coastal areas. Meanwhile, the active cold air in the southern Hemisphere and strengthening of the crossequatorial flow contributed to form and intensify low tropospheric jet in China and northern South China Sea. A large amount of moisture and unstable air-mass with Copyright © 2011 John Wiley & Sons, Ltd. high humidity and temperature is transported to the upper level. In this favorable situation, along with the special topography and underlying surface, difference of sea land distribution, non-uniform heating, thermodynamic and dynamical processes in atmosphere and the interaction in different scales would release heavy rain to the South China. Besides, unbalance force of atmospheric motion and the coupling reaction among convective cloud cluster and moisture frontal zone and low level jet lead to the continuation of strong storm. In post-flood season (July to October), the rainstorms are triggered by tropical system, such as tropical cyclone, inter-tropical convergence zone, and easterly wave. The tropical cyclone would not only bring tremendous moisture; they form big rainstorm directly due to the strong convergence and updraft. If combined with outside system (cold air and westerly belt system), it will bring more intense rainfall into the region. Hydrol. Process. 26, 3510–3523 (2012) 3520 T. YANG ET AL. A2 scenario pnl90 (%) 2080s 2020s 30 20 20 pav (%) 30 10 0 Winter Spring Winter -10 -20 -20 2080s 150 100 100 0 Winter Spring Summer Autumn -100 px1d (%) 2080s Winter Spring Summer Autumn Spring Winter -100 2050s 2050s 2080s 0 -50 2020s Summer Autumn 50 -50 180 160 140 120 100 80 60 40 20 0 -20 Spring 2020s 150 50 2080s 0 Summer Autumn 2050s 2050s 10 -10 2020s px1d (%) B2 scenario 2050s pnl90 (%) pav (%) 2020s 180 160 140 120 100 80 60 40 20 0 -20 2020s Winter Summer 2050s Spring Autumn 2080s Summer Autumn Figure 8. Changes in extreme precipitation between the period (1961–1990) and the period (2011–2099) under the H3A2 and H3B2 scenarios For example, under the influence of the hitting of typhoon and the cold air traveling from the north and northwest China, a heavy rainstorm occurred in southern Guangdong providence on 24 September 1979. The highest rainfall of Huiyang exceeded 400 mm. The monsoon trough is another important driving force compared with tropical cyclone. It brings persistent rainfall to South China. As for the precipitation in winter and spring, the anomalous vapor transport of the western Pacific and the low level in the South China Sea were the main impact factors, which was caused by the ENSO teleconnection. The El Niño made the low-level anticyclone of the Philippine Sea abnormal, which offered favorable water vapor condition for the rainstorm. In addition, prevailing south wind contributed to the continuous water vapor convergence in south China. While in case of the La Niña, the opposite phenomenon occurs. Therefore, the complex precipitation processes in Dongjiang River basin increase the difficulty in precipitation simulation. This explains why the indices that described very wet events (maximum of daily precipitation, maximum 5-day total, number of heavy events) were not simulated well. In addition, SDSM is not sufficiently powerful to capture the features of extreme precipitation events similar with other SDSMs (e.g. Srikanthan and McMahon, 2001). The defect of stochastic precipitation models need to be improved (Gregory et al., 1993). According to Wilby et al. (2004), this Copyright © 2011 John Wiley & Sons, Ltd. might attribute to the more stochastic nature of precipitation occurrence and magnitude, and the regression-based SDSMs often cannot explain entire variance of the downscaled variable. Additionally, while there is a strong seasonal consistency between stations for a number of predictors (e.g. geopotential heights and humidity), the seasonal specific predictor also play an important role (e.g. surface divergence during the summer months, Fealy and Sweeney, 2007). Hence, it is recommended the selected predictors at seasonal scale (or month scale) improve the downscaling performance to a certain degree. CONCLUDING REMARKS In this study, the large-scale atmospheric variables from GCMs output were downscaled to the regional scale in order to investigate the spatial-temporal changes in extreme precipitation, temperature, and pan evaporation over the Dongjiang River basin during 2010–2099 under H3A2 and H3B2 emission scenarios. It will improve current understanding on hydrological impacts under future climate change in the subtropical regions. The results for downscaling temperature under scenarios H3A2 and H3B2 showed that the temperature extreme events would be more significant in the rest 21st century (2010–2099). Despite the similar changes supplied by both scenarios, the magnitudes of the changes projected Hydrol. Process. 26, 3510–3523 (2012) STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES 2020s 2050s 2080s 80 80 60 60 40 20 0 -20 Spring Summer Autumn -20 2050s Winter 2080s Spring 2020s 175 125 125 pxcdd (%) pxcdd (%) 20 Summer Autumn -40 2020s 75 2050s 2080s 75 25 25 Winter Spring 2020s Summer Autumn 2050s -25 30 30 20 20 10 0 Spring Spring 2020s 40 Winter Winter 2080s 40 p90 (%) p90 (%) 2080s 40 175 -10 2050s 0 Winter -40 -25 2020s 100 px5d (%) px5d (%) 100 3521 Summer Autumn 2050s 2080s 10 0 -10 -20 Summer Autumn Winter Spring Summer Autumn -20 Figure 8. (Continued ) Westward extension and northward of Western Pacific subtropical high Two-trough and one-ridge circulation system in middle high latitude of the Eurasia Southward warm moist air Southward cold dry air Forming cold and stationary front Meso-and small-scale system convergence, shear, convective activity The special topography and underlying surface Rainstorm in south China Tropical system Non-uniform heating, difference of sea land ENSO cycle Figure 9. Conceptual diagram explained the heavy rain processes in South China by the two scenarios are generally different. As to the pan evaporation, the predicted value from H3A2 indicated that the maximum 1, 3, 5, and 7 days evaporation will decrease in winter while increase in other three seasons in 2010–2099. For H3B2, a general upward trend was identified in future. However, the projected changes for precipitation-related indices are uncertain. Copyright © 2011 John Wiley & Sons, Ltd. Although some preliminary results of changes in downscaled extreme indices are obtained in the present work, a number of uncertainties still exist in assessing the changes of regional-scale extreme indices. More research work in the future, particularly the ensemble projections by higher resolution GCMs or especially RCMs, as well as analyzing the uncertainties related to the model spread, Hydrol. Process. 26, 3510–3523 (2012) 3522 T. YANG ET AL. are needed for a more profound understanding of the futures changes in climate extremes. ACKNOWLEDGEMENTS The work was jointly supported by grants from the National Natural Science Foundation of China (40901016, 40830639, 40830640), a grant from the State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering (2009586612, 2009585512), and the Fundamental Research Funds for the Central Universities (2010B00714), the Australian Endeavour Fellowship Program, and CSIRO Computational and Simulation Sciences Transformational Capability Platform. Finally, cordial thanks are also extended to the editor, Professor Malcolm G. Anderson and two anonymous referees for their valuable comments which greatly improved the quality of this paper. REFERENCES Arnell NW, Hudson DA, Jones RG. 2003. Climate change scenarios from a regional climate model: Estimating change in runoff in southern Africa. Journal of Geophysical Research-Atmospheres 108(D16): AR 4519. Bardossy A, Plate EJ. 1992. Space-time model for daily rainfall using atmospheric circulation patterns. Water Resources Research 28: 1247–1260. Bouraoui F, Vachaud, G, Li LZX, Le Treut H, Chen T. 1999. Evaluation of the impact of climate changes on water storage and groundwater recharge at the watershed scale. Climate Dynamics 15: 153–161. Buma J, Dehn M. 2000. Impact of climate change on a landslide in South East France, simulated using different GCM scenarios and downscaling methods for local precipitation. Climate Research 15(1): 69–81. Cannon AJ, Whitfield PH. 2002. Downscaling recent streamflow conditions in British Columbia, Canada using ensemble neural network models. Journal of Hydrology 259(1): 136–151. Dibike YB, Coulibaly P. 2005. Hydrologic impact of climate change in the Saguenay watershed: comparison of downscaling methods and hydrologic models. Journal of Hydrology 307: 145–163. Faucher M, Burrows WR, Pandolfo L. 1999. Empirical-statistical reconstruction of surface marine winds along the western coast of Canada. Climate Research 11(3): 173–190. Fealy R, Sweeney J. 2007. Statistical downscaling of precipitation for a selection of sites in Ireland employing a generalised linear modelling approach. International Journal of Climatology 27: 2083–2094. Ghosh S, Mujumdar PP. 2008. Statistical downscaling of GCM simulations to streamflow using relevance vector machine. Advances in Water Resources 31(1): 132–146. Gordon HB, Whetton PH, Pittock AB, Fowler AM, Haylock MR. 1992. Simulated changes in daily rainfall intensity due to the enhanced greenhouse-effect–implications for extreme rainfall events. Climate Dynamics 8: 83–102. Gregory JM, Wigley TML, Jones PD. 1993. Application of Markov models to areaaverage daily precipitation series and interannual variability in seasonal totals. Climate Dynamics 8: 299–310. Hassan H, Hanaki K, Matsuo T. 1998. A modeling approach to simulate impact of climate change in lake water quality: Phytoplankton growth rate assessment. Water Science and Technology 37(2): 177–185. Hay LE, McCabe GJ, Wolock DM, Ayers MA. 1991. Simulation of precipitation by weather type analysis. Water Resources Research 27: 493–501. Haylock MR, Cawley GC, Harpham C, Wilby RL, Goodess CM. 2006. Downscaling heavy precipitation over the UK: a comparison of dynamical and statistical methods and their future scenarios. International Journal of Climatology 26: 1397–1415. Hennessy KJ, Suppiah R, Page CM. 1999. Australian rainfall changes, 1910–1995. Australian Meteorological Magazine 48: 1–13. Hundecha Y, Bardossy A. 2008. Statistical downscaling of extremes of daily precipitation and temperature and construction of their future scenarios. International Journal of Climatology 28(5): 589–610. Copyright © 2011 John Wiley & Sons, Ltd. Huth R. 2002. Statistical downscaling of daily temperature in Central Europe. Journal of Climate 15: 1731–1742. IPCC,. 2007. Climate Change 2007: The Physical Science Basis, Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Jasper K, Calanca P, Gyalistras D, Fuhrer J. 2004. Differential impacts of climate change on the hydrology of two alpine river basins. Climate Research 26(2): 113–129. Katz RW. 1996. Use of conditional stochastic models to generate climate change scenarios. Climatic Change 32: 237–55. Kettle H, Thompson R. 2004. Statistical downscaling in European mountains: verification of reconstructed air temperature. Climate Research 26(2): 97–112. Kim JW, Chang JT, Baker NL, Wilks DS, Gates WL. 1984. The statistical problem of climate inversion: determination of the relationship between local and large-scale climate. Monthly Weather Review 112: 2069–2078. Kim MK, Kang IS, Park CK, Kim KM. 2004. Super ensemble prediction of regional precipitation over Korea. International Journal of Climatology 24(6): 777–790. Kunkel KE, Pielke RA, Changon SA. 1999. Temporal fluctuations in weather and climate extremes that cause economic and human health impacts: a review. Bulletin of the American Meteorological Society 80: 1077–1098. Lamb HH. 1972. British Isles weather types and a register of daily sequence of circulation patterns, 1861–1971. Geophysical Memoirs 116: HMSO, London. Misson L, Rasse DP, Vincke C, Aubinet M, Francois L. 2002. Predicting transpiration from forest stands in Belgium for the 21st century. Agricultural and Forest Meteorology 111(4): 265–282. Qiao Y, Zhou X, Jian M, Huang W. 2010. Characteristics of droughts in Spring and its relationship with water vapor transportation in South China. Acta Scientiarum Naturalium Universitatis Sunyatseni 49(2): 125–129. Samel AN, Wang WC, Liang XZ. 1999. The monsoon rainband over China and relationships with the Eurasian circulation. Journal of Climate 12: 115–131. Schmidli J, Frei C, Vidale PL. 2006. Downscaling from GCM precipitation: A benchmark for dynamical and statistical downscaling. International Journal of Climatology 26: 679–689. Srikanthan R, McMahon TA. 2001. Stochastic generation of annual, monthly and daily climate data: A review. Generation of annual, monthly and daily climate data: a review. Hydrology and Earth System Sciences 5(4): 653–670. STARDEX,. 2001. Statistical and regional dynamical downscaling of extremes for European regions. Description of work. http://www.cru.uea. ac.uk/cru/projects/ stardex/description. pdf. Accessed 17 August 2005. Storch H, Zorita E, Cubash U. 1993. Downscaling of global climate change estimates to regional scales: an application to Iberian rainfall in wintertime. Journal of Climate 6: 1161–71. Tripathi S, Srinivas VV, Nanjundiah RS. 2006. Downscaling of precipitation for climate change scenarios: A support vector machine approach. Journal of Hydrology 330(3–4): 621–640. Wang P, Rong Y, Wang W, Wei L. 2011. Downscaling extreme precipitation in Dongjiang River Basin of China based on SDSM. Highlights of Sciencepaper Online 4(14): 1312–1320. (In Chinese with English abstract). Weisse R, Oestreicher R. 2001. Reconstruction of potential evaporation for water balance studies. Climate Research 16(2): 123–131. Wigley TML, Jones PD, Briffa KR, Smith G. 1990. Obtaining subgridscale information from coarse-resolution general circulation model output. Journal of Geophysical Research 95: 1943–1953. Wilby RL. 1998. Statistical downscaling of daily precipitation using daily airflow and seasonal teleconnection indices. Climate Research 10: 163–178. Wilby RL, Harris I. 2006. A framework for assessing uncertainties in climate change impacts: low-flow scenarios. Water Resources Research 42: W02419. Wilby RL, Wigley TML. 2000. Precipitation predictors for downscaling: observed and general circulation model relationships. International Journal of Climatology 20: 641–661. Wilby RL, Charles SP, Zorita E, Timbal B, Whetton P, Mearns LO. 2004. The guidelines for use of climate scenarios developed from statistical downscaling methods. http://ipcc-ddc.Cru.Uea.ac.uk/guidelines/StatDown_Guide.pdf. Wilby RL, Dawson CW, Barrow EM. 2002. SDSM- a decision support tool for the assessment of regional climate change impacts. Environmental Modeling and Software 17: 147–159. Wilby RL, Hay LE, Leavesley GH. 1999. A comparison of downscaled and rawGCM output: implications for climate change scenarios in the San Juan River basin, Colorado. Journal of Hydrology 225(1–2): 67–91. Hydrol. Process. 26, 3510–3523 (2012) STATISTICAL DOWNSCALING OF HYDROMETEOROLOGICAL EXTREMES Xu CY. 1999. From GCMs to river flow: a review of downscaling techniques and hydrologic modeling approaches. Progress in Physical Geography 23(2): 229–249. Yang T, Wang X, Zhao C, Chen C, Yu Z, Shao Q, Xu Q, Xia J, Wang WG,. 2011. Changes of climate extremes in a typical arid zone: Observations and multimodel ensemble projections,. Journal of Geophysical Research 116: D19106,. DOI:10.1029/2010JD015192. Copyright © 2011 John Wiley & Sons, Ltd. 3523 Yang T, Zhang Q, Chen YD, Tao X, Xu C,. 2008. A spatial assessment of hydrologic alternation caused by dam construction in the middle and lower Yellow River, China. Hydrological processes 22: 3829–3843. Zhang XC, Nearing MA, Garbrecht JD, Steiner JL. 2004. Downscaling monthly forecasts to simulate impacts of climate change on soil erosion and wheat production. Soil Science Society of America Journal 68(4): 1376–1385. Hydrol. Process. 26, 3510–3523 (2012)