عنوان مقاله [English]
Extreme weather conditions have an important role on strategic planning of water resource and developing adaptation plans and natural disaster management. Therefore, it is necessary to present a detailed perspective of upcoming extreme patterns of rainfall events. In the context of climate change, pattern extraction of extreme events can only be achieved by using of daily downscaling methods. In the current paper, two single site downscaling methods SDSM, DMDM and a multisite approach based on Singular Value Decomposition (SVD) technique are used and their results are evaluated. The case study is located on Tehran province with over 10 precipitation and synoptic stations in the period 1985 to 2005. The used climate change scenarios were generated for (2021-2050) period. In addition, the daily NCEP/NCAR dataset and results of climate change scenarios (RCP2.6, RCP4.5, and RCP8.5) were achieved from the Canadian Centre for Climate Modeling and Analysis.
For each downscaling models, based on their own concepts, suitable predictors have been selected via backward stepwise regression as a preprocessing step (Hessami et al. 2008). The implemented multisite approach is based on combination of two multiple regression models to simulate precipitation amount and occurrence and also using SVD to capture stochastic behavior of precipitation to preserve accurately the space–time statistical properties of daily precipitation (Khalili and Nguyen, 2017). Beside the SDSM as a regression based downscaling method (Wilby et al. 2002), DMDM as a regression based tool box including Multiple Linear Regression (MLR), Ridge Regression (RR), Multivariate Adaptive Regression Splines (MARS) and Model Tree (MT) have been used as well (Tavakol et al. 2013b).
To achieve the goal of the current research, temporal downscaling method to simulate extreme precipitation values is needed. In this regard, numerical model based on scaling invariant concept is used to do temporal downscaling (Nguyen et al. 2007). The sub daily extreme rainfall, are estimated from daily downscaled rainfalls by analyzing the non-central moments of observed rainfalls, single time regime (from 6 h to 24 h) and using scaling factor. Finally, as the major output of this study, Intensity Duration Frequency (IDF) curve is calculated affected by climate change in the period 2020 to 2050. The results based on statistical assessment both in calibration and validation periods of daily precipitation show the effectiveness of SDSM and DMDM models, respectively, in terms of long-term monthly average, and multisite model in preserving the trend of computed information in comparison with observed values. Based on uncertainty assessments results, DMDM provided the most precsion results versus the other methods over the study area. In addition, the models performance rank in estimating unseen station belong to the DMDM and multisite and SDSM methods, respectively. In the second step, quantities of IDF curves for return periods of (2-100) years and durations of (6, 12, 24 hour) at Mehrabad station are estimated using the results of coupled three different spatial downscaling and GEV distribution. Results show higher accuracy of DMDM and SDSM models respectively in comparison with multisite model. Based on the linear structure of SDSM and Multisite downscaling models versus the complex structure of DMDM, it seems that limitations of linear methods cause DMDM to be superior to the other ones. In addition, evaluation of the results of extreme values by three different climate change scenarios based on the DMDM downscaling model indicates an increase in rainfall intensity using scenario RCP8.5 and a decrease under scenarios RCP4.5 and RCP2.6.