Investigating the performance of data mining, lumped, and distributed models in runoff projected under climate change

Abstract

This work evaluates the effects of climate change on the surface water resources (river flow) of the Sanjabi basin, Iran, by comparing data-mining, lumped, and distributed models, namely artificial neural networks (ANN), the identification of unit hydrographs and component flows from rainfall, evaporation, and streamflow (IHACRES) model, and the soil and water assessment tool (SWAT). Climate projections in terms of monthly temperature and rainfall made by 17 atmosphere–ocean general circulation models (AOGCMs) by the 5th Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) under emission scenarios of Representative Concentration Pathways (RCPs) (RCP2.6, RCP4.5, and RCP8.5) during the baseline period 1971–2000 and future periods 2040–2069 and 2070–2099 are applied in the Sanjabi basin. The predictive skill of the AOGCMs is evaluated with performance criteria. The evaluation results indicate the CNRM-CM5 model features the best performance in terms of rainfall, average temperature, and minimum temperature projections, and the GFDL-CM3 provides the most accurate maximum temperature projections. Four downscaling methods (change factor (Delta), ClimGEN, LARS-WG, and Genetic Programming (GP)) are compared based on the R2, RMSE, MAE, and NSE. The predictive skill of the LARS-WG method was the highest. ANN, IHACRES, and SWAT are implemented to project future runoff following calibration and testing. The IHACRES model exhibits the best performance. The IHACRES model is applied to project future runoff under climate-change scenarios. The results indicate a reduction in runoff under all emission scenarios in the two future periods, with the RCP8.5 scenario featuring the largest reductions in runoff in 2040–2069 and 2070–2099 and being equal to 42.0 and 44.3%, respectively.