An expert system for predicting the infiltration characteristics

Abstract

Infiltration plays a fundamental role in streamflow, groundwater recharge, subsurface flow, and surface and subsurface water quality and quantity. This study includes a comparative analysis of the two machine learning techniques, M5P model tree (M5P) and Gene Expression Programming (GEP), in predictions of the infiltration characteristics. The models were trained and tested using the 7 combination (CMB1 – CMB7) of input parameters; moisture content (m), bulk density of soil (D), percentage of silt (SI), sand (SA) and clay (C), and time (t), with output parameters; cumulative infiltration (CI) and infiltration rate (IR). Results suggested that GEP has an edge over M5P to predict the IR and CI with R, RMSE and MAE values 0.9343, 15.9667 mm/hr & 8.7676 mm/hr, and 0.9586, 9.2522 mm and 7.7865 mm for IR and CI, respectively with CMB1. Although the M5P model also gave good results with R, RMSE and MAE values 0.9192, 14.1821 mm/hr, and 19.2497 mm/hr, and 0.8987, 11.2144 mm and 18.4328 mm for IR and CI, respectively, but lower than GEP. Furthermore, single-factor ANOVA and uncertainty analysis were used to show the significance of the predicted results and to find the most efficient soft computing techniques respectively. HIGHLIGHTS Infiltration characteristics is predicted using two soft computing techniques; M5P and GEP.; Linear relationship model is generated from M5P and GEP.; Single-factor ANOVA and uncertainty analysis is done for the best selected models.;

Keywords

• cumulative infiltration
• gene expression programming
• infiltration rate
• m5p model tree