Improving capability of conceptual modeling of watershed rainfall–runoff using hybrid wavelet-extreme learning machine approach

Rainfall–runoff process identification, due to uncertainties and complexities, requires advanced modeling strategies. For this end, this study presented different strategies to explore spatio-temporal variation of rainfall–runoff process for the Ajichay watershed located in northwest Iran. Extreme learning machine (ELM) was used to predict the runoff in conceptual models. First, a geomorphology integrated ELM (G-ELM) was used to predict watershed runoff in multiple-stations form for the watershed. The spatial and temporal features of sub-basins were selected as input data wherein temporal features were pre-processed by wavelet transform (WT). Results confirmed the capability of G-ELM in successive prediction of watershed runoff. Afterwards, an integrated ELM (I-ELM) was developed based on conceptual reservoir modeling to predict monthly river runoff where the model had the semi-distributed specifications of ELM. This model was capable of exploring spatial variation of rainfall–runoff process without requiring physical characteristics of sub-basins. To meter sufficiency of the modeling strategies, cross-validation technique was performed for station 3 in which G-ELM performed better in comparison to I-ELMs. Furthermore, classic and wavelet-based modeling (W-ELM) of rainfall–runoff was performed for one-step-ahead predictions. Statistical evaluations confirmed the W-ELM, I-ELM, and G-ELM performance, respectively.