ENN-SA: A novel neuro-annealing model for multi-station drought prediction

Abstract

This paper presents a new hybrid model, called ENN-SA, for spatiotemporal drought prediction. In ENN-SA, an Elman neural network (ENN) is conjugated with simulated annealing (SA) optimization and support vector machine (SVM) classification algorithms for the standardized precipitation index (SPI) modeling at multiple stations. The proposed model could be applied to predict SPI at different time scales in a meteorology station with lack of data through the intelligent use of SPI series of the nearby stations as the model inputs. The capability of the hybrid model for multi-station prediction of meteorological drought was examined through the cross-validation technique for Kecioren station in Ankara Province, Turkey. To this end, the SPI-3, SPI-6, and SPI-12 at the station were modeled using the same indices of five nearby stations. In the first step, SVM was trained using different kernels in order to generate and classify a set of plausible multi-station prediction scenarios. Then, ENN was used to regress the SPI series at each scenario and finally, the SA component of the integrated model was utilized to improve the ENN efficiency. Various error and complexity measures were used to detect the models' performance. The results showed the ENN-SA is promising and efficient for multi-station SPI prediction.