Uncertainty Assessment of the Integrated Hybrid Data Processing Techniques for Short to Long Term Drought Forecasting in Different Climate Regions

Accurate prediction of drought indices is a useful method to reduce its undesirable consequences. In this study, the workability of newly integrated hybrid forecasting approach based on Meta model and data processing methods was assessed for forecasting the Standardized Precipitation Evapotranspiration Index (SPEI) in districts with different climates. The short-, mid-, and long-term SPEIs series (i.e. timescale of 3, 9, and 24 month) were computed during the period of 1951–2019 for five sites located in Iran. In this regard, first temporal features of the SPEIs were broken down using Wavelet Transform (WT). Then, for obtaining features with higher stationary properties, Ensemble Empirical Mode Decomposition (EEMD) was applied to further decompose the obtained subseries. Finally, the most efficient subseries were selected and inserted to Meta model approaches [i.e. Feed Forward Neural Network (FFNN), Kernel Extreme Learning Machine (KELM), and Gaussian Process Regression (GPR)] as inputs. Results showed that the proposed methods enhanced the models' capability between 35 to 45%. The capability of the proposed model was verified via Overlap Discrete Wavelet Transform (MODWT) method. Results showed that the distribution range of the Root Mean Square Errors (RMSE) criteria for integrated methods decreased from 0.036–0.172 (in raw data) to 0.025–0.109 (in decomposed data). The Monte Carlo uncertainty analysis was used to assess the applied models dependability. Results showed that the integrated model with having values of 72.8% to 89.2% for the 95PPU indicator had an allowable degree of uncertainty in short- to long-tern drought modeling.