Application of hybrid meta-heuristic-based data-driven models to forecast streamflow drought index

Drought is a complex natural catastrophe threatened markedly the societies. Thus, its precise prediction has a noteworthy effect in numerous sections such as water resources, nutrition, economics, industry, etc. In the current investigation, to evaluate the hydrological drought procedure in Nazlu River basin, Urmia City, West Azerbaijan province, the streamflow drought index (SDINRB) is applied. In this direction, the SDINRB in four different time measures including 3, 6, 9, and 12-month are calculated using 829 mean monthly streamflow datasets recorded from Aug 1951 to Aug 2020 by Tapik hydro-meteorological station. Then, two robust advanced hybrids support vector regression (SVR) with Harris hawk algorithm (HHO) and intelligent water drop (IWD) optimization algorithms i.e., hybrid SVR-HHO and SVR-IWD models, are developed to estimate the fluctuations pattern of SDINRB-3-12. Given this, to accomplish the optimum supportive models' structure, many scenarios are implemented by tuning meta-parameters such as a number of hidden neurons and deterministic factors of SVR, HHO, IWD algorithms. According to the performance assessment criteria and comparison plots, the hybrid SVR-RBF-HHO model under ideal meta-parameters is identified as the suitable model for predicting SDINRB-3-6 droughts, yet the hybrid SVR-RBF-IWD model is recognized as the appropriate model for forecasting SDINRB-9-12 droughts. Likewise, the best modelling performance is achieved by the hybrid SVR-RBF-HHO model in predicting SDINRB-6 drought. It results in an RMSE, R-2, NSE, and MBE of 0.23, 0.94, 0.92, and 0.072, respectively. Nonetheless, for the single SVR-RBF as the benchmark model is attained 0.43, 0.79, 0.78, and 0.045, respectively.