Application of Novel Hybrid SVMD–NARX–ACO Model for Precise Predicting Monthly Pan Evaporation

Abstract Pan evaporation ( E pan ) is a direct measurement of compound climatological interactivities reflected as a decisive criterion of universal warming. Since numerous factors impact E pan , its precise prediction is a difficult procedure. So, in the present study, a novel advanced coupled predictive disintegration–optimization-based method is recommended to improve the prediction accuracy of the E pan . The recommended predicting method is a coupling version of SVMD (successive variational mode decomposition) with NARX (nonlinear autoregressive with exogenous inputs) neural network optimized by ant colony optimization (ACO) meta-heuristic algorithm (i.e . , hybrid SVMD–NARX–ACO model) with the seq2seq regression module of forecasting. The Nazlu basin, Urmia City, West Azerbaijan province, Iran, is used as the study area. Through Shannon entropy process, the operative predictor variables in modeling among a list of on-site available potential meteorological parameters on E pan recorded in the Nazlu basin ( ME pan NB ) are identified as T min , T ave , and T max . In all models, to achieve appropriate estimation results for decreasing the overfitting influence and enhancing the skill of models, meta-parameters including number of hidden neurons ( NHN ), number of time delays ( d ), and compactness of mode ( α ) are precisely tuned. After a lot of experiments, the optimal value for NHN/d for NAR, NARX, hybrid SVMD–NARX, and SVMD–NARX–ACO models are achieved to be 15/4, 10/3, 15/3, and 12/3, respectively. Performance judgment criteria verify that the coupled SVMD–NARX–ACO technique via the operative predictor and ideal hyper-parameters outperforms with an R 2 of 0.92, NSE of 0.94, MBE of 2.27 mm, and RMSE of 9.65 mm. Comparatively, the individual NARX as the basis model yields in an R 2 of 0.77, NSE of 0.8 , MBE of 16.1 mm , and RMSE of 54.5 mm.