Exploring spatiotemporal meteorological correlations for basin scale meteorological drought forecasting using data mining methods

In this paper, two data mining methods, support vector machine (SVM) and group method of data handling (GMDH), were used to identify spatiotemporal meteorological correlations, which can be used to forecast basin scale seasonal droughts. Standardized Precipitation Index (SPI) was used as a meteorological drought severity index. The case study of this paper consists of the basins of four major dams in Iran that supply domesticwater demands of Tehran, the capital city of Iran. A GMDH and an SVM model optimized by particle swarm optimization (PSO) were used to predict seasonal SPIs in the fall, winter, spring, and some combined seasons. The historical time series of the meteorological variables including air temperature and geopotential height at the surface, and 300, 500, 700, and 850 mbar levels in the geographical zone covering 10 to 60 degrees north latitudes and 0 to 90 degrees east longitudes were selected as the model predictors. Average mutual information (AMI) index was used for feature selection among the mentioned predictors. The selected predictors in the months of April to August were used as the SVM and GMDH inputs. The results showed that the seasonal SPI values could be forecasted by the proposed model with 2- to 5-month lead-time with enough accuracy. Hence, the proposed method can be used in mid-term water resource management in the study area.