Urmia lake water depth modeling using extreme learning machine-improved grey wolf optimizer hybrid algorithm

Abstract

Lake water level changes are relatively sensitive to the climate-born events that rely on numerous phenomena, e.g., surface soil type, adjacent groundwater discharge, and hydrogeological situations. By incorporating the streamflow, groundwater, evaporation, and precipitation parameters into the models, Urmia lake water depth is simulated in the current study. For this, 40 years of streamflow and groundwater recorded data, respectively collected from 18 and 9 stations, are utilized together with evaporation and precipitation data from 7 meteorological stations. Extreme learning machine (ELM) is hybridized with four different optimizers, namely artificial bee colony (ABC), ant colony optimization for continuous domains (ACOR), whale optimization algorithm (WOA), and improved grey wolf optimizer (IGWO). In the analysis, 13 various scenarios with multiple input combinations are used to train and test the employed models. The best scenarios are then opted based on the performance metrics which are applied to assess the accuracy of the methods. According to the results, the hybrid ELM-IGWO shows better performance compared to the ELM-ABC, ELM-ACOR, and ELM-WOA approaches. Results indicate that the groundwater and persistence of the lake water depth have effective roles in models while incorporating higher number of variables can lower the performance of the models. Statistical analysis showed a 62% improvement in the performance of ELM-IGWO in comparison to the ELM-WOA with regard to the root mean square error. The promising outcomes obtained in this study may encourage the application of the utilized algorithms for modeling alternative hydrological problems.