A novel nonlinear Arps decline model with salp swarm algorithm for predicting pan evaporation in the arid and semi-arid regions of China

• Published in: Journal of hydrology (Amsterdam) Vol. 582; p. 124545
• Main Authors: Wang, Haomin, Yan, Hui, Zeng, Wenzhi, Lei, Guoqing, Ao, Chang, Zha, Yuanyuan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2020
• Subjects: Cross-station; Machine learning; Meta-heuristic algorithm; Model comparison; Model comparison; Meta-heuristic algorithm; Cross-station; Machine learning
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2020.124545

•A novel nonlinear Arps decline model with salp swarm algorithm (SSA-KNEA) is proposed.•M5 model tree (M5) and the multivariate adaptive regression splines (MARS) was as compared.•The proposed model outperforms other machine learning models in forecasting evaporation.•SSA-KNEA Model with station 51777 was the best general model. Accurate prediction of water surface evaporation (Ep) is important in the fields of both hydrology and irrigation engineering. This study evaluated the potential ability of a new hybrid model based on the salp swarm algorithm (SSA) and the kernel-based nonlinear Arps decline (KNEA) in predicting Ep. Two other common machine learning models, including the M5 model tree (M5) and the multivariate adaptive regression splines (MARS), were also applied in this study for comparison. All models were developed using daily records between 2000 and 2015 from 12 meteorological stations in the arid and semi-arid regions of northwest China. These daily records, including the maximum and minimum temperatures, solar radiation, wind speed and relative humidity, were randomly divided into two parts, with 70% of which used for model calibration and the others applied for validation. Four different parameter input combinations were equipped to explore the possibility of improving model accuracy. Two data application scenarios and five statistical indicators including the root-mean-square-error (RMSE), mean absolute error (MAE), scatter index (SI), d-index and determination coefficient (R2) were used for model evaluation. In the scenario of using local data as inputs for model calibration and validation, the impacts of wind speed and relative humidity on Ep were both greater than that of solar radiation, and SSA-KNEA was consistently superior to MARS or M5 across all the input combinations. In the scenario of using cross-station data, in which models using the best input combination were developed by local data of each station but validated by data from each of the remaining 11 stations, SSA-KNEA models performed better than MARS or M5 models on average. In addition, the SSA-KNEA model established by data from Station 51777 was the most suitable generalized model in this research area. Overall, our findings suggested that the new hybrid algorithm (i.e., SSA-KNEA) has high potential for Ep estimation in the arid and semi-arid regions of China, with local or cross-station data.