Comparing the Performance of Dynamical and Statistical Downscaling on Historical Run Precipitation Data over a Semi-Arid Region

Abstract

Precise evaluations of climate model precipitation outputs are valuable for making decisions regarding agriculture, water resource, and ecosystem management. Many downscaling techniques have been developed in the past few years for projection of weather variables. We need to apply dynamical and statistical downscaling (DD and SD) to bridge the gap between the coarse resolution general circulation model (GCM) outputs and the need for high-resolution climate information over a semi-arid region. We compare the requirements of DD (RegCM4) and SD (Delta) approaches, evaluate the historical run of NNRP1 data in comparison with station data, and analyze the changes in wet days and precipitation values through both methods during 1990–2010. In this study, we did not want to use prediction data under different scenarios of climate change, and we have just applied observed data to assess the amount of precise of NNRP1 data, over the observed period. SD method requires less time and computing power than DD. The DD approach performs better over the evaluation period according to efficiency criteria. In general, the Pearson correlation in DD with observation data in evaluation period was higher than (r > 0.72 and R2 > 0.52) SD (r > 0.65 and R2 > 0.41) over three study stations. Similarly, MAE and NSE show better results from DD relative to SD. SD underestimates the number annual mean wet-days for all three stations examined. DD overestimates a number of annual mean wet-days, but with less deviation from the observed mean.