Performance evaluation of the genetic programming and support vector machine models in reconstruction of missing precipitation data

Incomplete rainfall datasets with missing gaps is a major challenge in climatology and water resource studies. In the present study, two intelligent models, namely Genetic Programing (GP) and Support Vector Machines (SVM) were used to reconstruct the monthly rainfall data of four rain-gauges located in Hamedan province, Iran during the period of 1992 to 2011. The incomplete rainfall data was reconstructed first by using the data of one, two and three stations respectively. The results showed that increasing the memory and the number of stations involved in the training phase, will improve the performance of the models. In reconstruction of monthly precipitation data of Sarabi and Maryanj stations, the Support Vector Machine method showed better performance with RMSE of 12.9 mm and 11.4 mm, and correlation coefficients (r) of 0.93 and 0.95, respectively. The corresponding values of RMSE for GP approach were 13 mm and 12.21 mm, which indicated the superior performance of SVM.