Evaluation of data-driven models to downscale rainfall parameters from global climate models outputs: the case study of Latyan watershed

Assessment of climate change in future periods is considered necessary, especially with regard to probable changes to water resources. One of the methods for estimating climate change is the use of the simulation outputs of general circulation models (GCMs). However, due to the low resolution of these models, they are not applicable to regional and local studies and downscaling methods should be applied. The purpose of the present study was to use GCM models' outputs for downscaling precipitation measurements at Amameh station in Latyan dam basin. For this purpose, the observation data from the Amameh station during the 1980–2005 period, 26 output variables from two GCM models, namely, HadCM3 and CanESM2 were used. Downscaling was performed by three data-driven methods, namely, artificial neural network (ANN), nonparametric K-nearest neighborhood (KNN) method, and adaptive network-based fuzzy inference system method (ANFIS). Comparison of the monthly results showed the superiority of KNN compared to the other two methods in simulating precipitation. However, all three, ANN, KNN, and ANFIS methods, showed satisfactory results for both HadDCM3 and CanESM2 GCM models in downscaling precipitation in the study area.