Hydrological evaluation of conservation bench terrace system in outer foothills of Himalayas using soft computing techniques

Abstract

Soil erosion has been problematic in the outer foothills of the Himalayas. Therefore, it is necessary to implement suitable conservation measures to check erosion issues. The present study experimented on conservation bench terraces (CBT) as a conservation measure in the outer foot Himalayas in Dehradun, India. The study evaluates the effectiveness of different CBTs of embankment heights 10 cm (CBT-10), 15 cm (CBT-15), and 20 cm (CBT-20) to runoff and soil loss under different rainfall storms. Further, the estimation of Soil loss (Sloss) was evaluated using soft computing techniques (machine learning models). The soft computing techniques were applied to the data obtained from CBT plots for hydrological assessment. The methods used for analysis were support vector machine (SVM), Bagging, and Random Forest (RF) models. The performance of developed models was evaluated using mean absolute error (MAE), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), Pearson's correlation coefficient (PCC), and Wilmott Index (WI) along with graphical representation (Scatter plot and Taylor diagram). The results showed that the CBT techniques were observed to reduce runoff and soil loss. Among CBT, CBT-20 was the most significant as the number of runoff events was less than others. Considering the storm size, the maximum percentage of runoff generated from the storm size is more important than 100 mm. Moreover, soil loss (Sloss) estimation showed that the SVM models perform better in all three scenarios, which are results of three scenarios such as R2 = 0.50, 0.52 and 0.80; MAE = 0.03, 0.03 and 0.10; RMSE = 0.08, 0.07 and 0.27; NSE = 0.39, 0.12 and 0.21; PCC = 0.71, 0.725 and 0.89; and WI = 0.67, 0.794, and 0.49, respectively. Therefore, SVM techniques can be applied to the estimation of Sloss. Results can be recommended for precision farming under climate changes in the hilly and plain areas and similar models can be applied to other areas.