Statistical Downscaling of Extremes of precipitation and Construction of Their Future Scenarios in Kashfroud Basin

The Intergovernmental Panel on Climate Change (IPCC) stated that there is high confidence that recent climate changes have had discernible impacts on physical and biological systems. Impacts of climate change are felt most strongly through changes in extreme climate events, which are responsible for a major part of climate-related economic losses (Jiang, et. al. 2012). The state-of-the art General Circulation Models (GCMs) can reproduce important processes in global and continental scale of atmosphere and predict future climate under different emission scenarios. Since spatial resolutions of GCMs are often coarse (hundreds of kilometer), there is a mismatch of scale between GCMs and the scale of interest for regional impacts. Therefore, a range of downscaling methods have been developed to bridge the gap between the coarse resolution of the climate model outputs and the need for surface weather variables at finer spatial resolution (Wang et. al. 2011). Downscaling methods can be divided into two classes: dynamical downscaling (DD) and statistical (empirical) downscaling (SD). In this study, SD Model was evaluated by downscaling precipitation in the Kashafroud Basin. The statistical downscaling model (SDSM) used in our study here is a hybrid of a stochastic weather generator and regression methods (Wilby et al. 2001). This method includes a built-in transform functions in order to obtain secondary data series of the predictand and/or the predictor that have stronger correlations than the original data series (Wilby et al. 2004). Study area The KashafRoud basin, located between 58° 2´ and 60° 8´ E and 35° 40´ and 40° 36 ´N, totally has an area of about 16500 km2. To the north east of the catchment is the HezarMasjed Mountain, to the south west is the Binaloud Mountain and in the center of the catchment is the Mashhad plain. The climate of KashafRoud river basin ranges from severe semiarid to arid climate. The multi-year average precipitation and air temperature of the basin is about 220 mm and 12/2 °C respectively (Sayari et. al., 2011). Data: The data used for evaluation were large-scale atmospheric data encompassing daily NCEP/NCAR reanalysis data during 1961-2001 and the daily mean climate model results for scenarios A2 and B2 of the HadCM3 model during 1961-2099. Areal average daily precipitation data of the KashafRoud basin (Mean of four weather stations daily precipitation data) during 1969-2001 was used for downscaling. Modeling of four extreme precipitation In which R means the local predictand, L (l1, l2,..., ln) represents n large scale atmospheric predictors, and F is the built quantitative statistical relationship. SDSM uses large-scale atmospheric variables to condition the rain occurrence as well as the rainfall amount in wet days. It can be expressed as follows (Wetterhall et al. 2009; Wilby et al. 2004): in which i is time (days), ωi is the conditional possibility of rain occurrence on day i, i uˆ is the normalized predictor, αj is the regression parameter and ωi−1 and αi−1 are the conditional probabilities of rain occurrence on day i−1 and lag-1 day regression parameters, respectively. These two parameters are optional, depending on the study region and predictand. We used a uniformly distributed random number ri (0≤ri≤1) to determine the rain occurrence and supposed that rain would happen if ωi≤ri. On a wet day, rainfall can be expressed by a z-score as:In which Zi is the z-score on day i, βj is the calculated regression parameter, and βi−1 and Zi−1 are the regression parameter and the z-score on day i−1, respectively. As mentioned above, they are also optional; ε is a random error term represented by the normal distribution.Downscaling precipitation Calibration and validation of SDSM First, all of the 26 atmospheric variables in the region were taken as potential predictors, then most sensitive predictors for the region were analyzed month by month. The analysis results were integrated; and finally, 3 predictors were selected for predictand. The results showed that the pattern of change and numerical value of precipitation can be reasonably simulated. Although some differences existed between values of observed and simulated indices but the pattern of change in most of months were good. In the next 30 years, total annual precipitation would decrease by about 3.3 % in A2 scenario and 3.6% in B2 scenario and summer might be the most distinct season among all the changes in extreme precipitation indices.