Soil Nitrogen Detection Based on Random Forest Algorithm and Near Infrared Spectroscopy

Abstract

In order to achieve rapid detection of soil nitrogen content, a method of soil nitrogen content detection by near-infrared spectroscopy combined with a random forest regression algorithm (RF) was proposed. The spectral data and nitrogen contents of 143 soil samples were collected to establish the detection model by combining the random forest algorithm. The results show that the preferential selection of RF modeling data by ∆Gini can extract the spectral information related to soil nitrogen content and reduce the redundant information of the data. The prediction accuracy of the established RF model is high with a correlation coefficient of 0.909 and root mean square error of 0.1412 for the test set prediction. This study proves the feasibility of NIR spectroscopy combined with a random forest algorithm for soil nitrogen content prediction. The result also demonstrated the feasibility of combining NIR spectroscopy with a random forest algorithm to predict soil nitrogen content and the theoretical basis for the subsequent development of soil composition testing instruments.