Support vector regression to downscaling climate big data: an application for precipitation and temperature future projection assessment

Abstract

The techniques for downscaling climatic variables are essential to support tools for water resources planning and management in a climate change context in the entire world. Support vector machines (SVM) through regression approach (SVR), constitute an artificial intelligence method to downscaling climatic variables. Since that statistical downscaling based on regression methodologies is susceptible to the predictor variables, the aim of this study was exploring a big database of predictor variables to achieve the best performance of a statistical downscaling model using SVR to predict precipitation and temperature future projections. Data from regional climate models of Ecuador and information of three meteorological stations was used to apply this approach in the Tomebamba river sub-basin, located in southern Ecuadorian Andean region. The results show that the downscaling model has a better performance with the climatic averages. The precipitation extremes do not estimate in a good manner, but the model achieves an effective behavior with the temperature extremes values. These results could serve to improve water balance projections in the future for formulating suitable measures for climate change decision-making.