Comparing Multiple Regression, Principal Componant Analysis, Partial Least Square Regression and Ridge Regression in Predicting Rangeland Biomass in the Semi Steppe Rangeland of Iran

Abstract

In this paper, the prediction of rangeland biomass using different methods including Multiple regression, Principal Component Analysis, Partial Least Square regression and Ridge regression were compared. For this goal, environmental factors such as elevation (m), rainfall (mm), slope (٪), caco3 (٪), Sand (٪), Sand (٪), Clay (٪), pH, EC (ds.m-1), Saturation (SP (٪)) were used to determine a relationship between environmental factors and the forage yield. The results showed that PLS, and ridge regressions were among the best models to predict rangeland biomass followed by multiple regressions and Principal component Analysis. PLS and ridge regression had a higher predicted accuracy than other calculation methods. It was shown that, the Partial Least Square regression values to R, RMSE and MAE were 0.83, 34.9 and 26.23, respectively.