Development of Pedo-transfer Functions for Estimating Soil Aggregation and Erodibility in Kandi Region of Punjab, India

Quantification of soil aggregation and erodibility from easily measurable soil characteristics have been done by using pedo-transfer functions (PTFs) and PTFs developed were compared using statistical and machine learning techniques for the kandi region of Punjab. Dataset 1, having six basic soil properties, was used for the estimation of mean weight diameter (MWD) and erodibility (K), prediction using an artificial neural network (ANN) was slightly better than a generalized linear model (GLM). In dataset 2, six basic soil properties in dataset 1 having high correlation with soil parameters were used and prediction using GLM was slightly better than ANN. In dataset 3 including all 11 basic soil properties, prediction using ANN was significantly better than GLM. Thus, ANN performs better for a complex system having a greater number of variables whereas for a small set having fewer variables, the statistical methods perform better.