Input selection for long-lead precipitation prediction using large-scale climate variables: a case study

Abstract

In this study, a precipitation forecasting model is developed based on the sea level pressures (SLP), difference in sea level pressure and sea surface temperature data. For this purpose, the effective variables for precipitation estimation are determined using the Gamma test (GT) and correlation coefficient analysis in two wet and dry seasons. The best combination of selected variables is identified using entropy and GT. The performances of the alternative methods in input variables selection are compared. Then the support vector machine model is developed for dry and wet seasonal precipitations. The results are compared with the benchmark models including naïve, trend, multivariable regression, and support vector machine models. The results show the performance of the support vector machine in precipitation prediction is better than the benchmark models.