Drought forecasting using data-driven methods and an evolutionary algorithm

• Published in: Modeling earth systems and environment Vol. 3; no. 4; pp. 1675 - 1689
• Main Authors: Hosseini-Moghari, Seyed-Mohammad, Araghinejad, Shahab, Azarnivand, Ali
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.12.2017; Springer Nature B.V
• Subjects: Algorithms; Autoregressive models; Chemistry and Earth Sciences; Computer Science; Correlation coefficient; Correlation coefficients; Criteria; Drought; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Ecosystems; Environment; Evolutionary algorithms; Forecasting; Frameworks; Lead time; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Mathematical models; Original Article; Physics; Root-mean-square errors; Statistical analysis; Statistics for Engineering; Support vector machines; Warning systems; Standard precipitation index; Support vector regression; Multi-layer perceptron; Drought forecasting; Evolutionary algorithm
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-017-0385-x

The present study focuses on quantitative (exact) and qualitative (classifying) drought forecasting in Gorganrood, Iran, based on monthly time-series of standard precipitation index (SPI) with 1–6 months lead-times. In so doing, recursive multi-layer perceptron (RMLP) and recursive support vector regression (RSVR) were optimized via an imperialist competitive algorithm (ICA). A traditional approach, autoregressive integrated moving average (ARIMA), has also been applied in this case. In quantitative forecasting, ICA-RMLP and ICA-RSVR models outperformed ARIMA ones according to three performance criteria namely, correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE). For example, in SPI 24 and one month lead time forecasting; R, RMSE, and, MSE values for ARIMA model equaled to 0.90, 0.484, and 0.322 while, for ICA-RMLP equaled to 0.967, 0.277, and 0.188, respectively. In contrast, the criteria for ICA-RSVR were evaluated 0.969, 0.278, and 0.186, respectively. Increases in lead-times decreased the forecasting accuracy for both qualitative and quantitative forecasting. However, increases in SPI scales provided more accurate results. Whereas, in the quantitative forecasting, model could provide appropriate forecasts for all scales of SPI. According to the performance of the proposed framework, it would be practical for developing a drought warning system.