Modelling of Soil Infiltration Rate using Machine Learning Techniques for Hilly Watershed of Sikkim

This paper presents the comparative analysis of basic soil Infiltration Rates (IR) estimated through three different machine learning approaches namely Artificial Neural Network (ANN), Support Vector Regression (SVR), and Multiple Linear Regression (MLR) with the infiltration rate measured using double ring infiltrometer in Ranikhola watershed in Sikkim, India. The input parameters considered for the modelling of infiltration rate were the soil texture, bulk density, particle density, and organic carbon from 81 locations in the watershed. The procured data were divided into a ratio of 70:30 for the training and testing phases of the modelling. In comparison to MLR and ANN models, the SVR model with the Radial Biased Functional (RBF) kernel predicted the basic soil infiltration rate with the highest value of Coefficient of Determination (R2) (0.91 and 0.87), and lowest value of Root Mean Square Error (RMSE) (1.75 and 1.24), Mean Absolute Error (MAE) (0.40 and 0.51) and Mean Absolute Percentage Error (MAPE) (40.1 and 51.19) during training and testing phases, respectively. In terms of performance, the ANN and MLR models followed the SVR model. The findings of the study also emphasised the application of machine learning techniques using easily measurable soil properties for the prediction of basic soil IR in inaccessible hilly terrains.