Cumulative infiltration and infiltration rate prediction using optimized deep learning algorithms: A study in Western Iran

Sixteen different sites from two provinces (Lorestan and Illam) in the western part of Iran were considered for field data measurement cumulative infiltration, infiltration rate, other effective variables that affect process. Soil is recognized as a fundamental process hydrologic cycle affecting surface runoff, soil erosion, groundwater recharge. Hence, accurate prediction one most important tasks hydrological science. As direct difficult costly, empirical models are inaccurate, current study proposed standalone, optimized deep learning algorithm convolutional neural network (CNN) using gray wolf optimization (GWO), genetic (GA), an independent component analysis (ICA) rate prediction. First, 154 raw datasets collected including time measuring; sand, clay, silt percent; bulk density; moisture rate; survey. Next, 70 % dataset used model building remaining 30 was validation. Then, based on correlation coefficient between input outputs, combinations constructed. Finally, power each developed evaluated visually-based (scatter plot, box plot Taylor diagram) quantitatively-based [root mean square error (RMSE), absolute (MAE), Nash-Sutcliffe efficiency (NSE), percentage bias (PBIAS)] metrics. Finding revealed more while characteristics (i.e. content) significant This shows area, parameter, which dominant constituent can control effectively. Effectiveness present study, order importance time, silt, content, density. be related to fact area rangeland thus, overgrazing leads compaction lead slow content density not highly our because these factors do significantly change across area. Findings demonstrated optimum variable combination, all considered. The results illustrated CNN algorithms have very high performance, metaheuristic enhanced performance standalone (from 7% 28 %). also showed CNN-GWO outperformed algorithms, followed by CNN-ICA, CNN-GA, both All underestimated overestimating rates.