Assessment of soil fertility indicators based on remote sensing data

Abstract

The article discusses the possibility of using artificial intelligence and machine learning methods to predict soil fertility based on remote sensing data. The research object was 70 soil samples taken in the village of Vshchizh, Bryansk region. Multispectral imaging of the earth's surface from the Sentinel-2 satellite was also used to assess the soil condition. The study in this article is based on the indicators of humus (%), P2O5 (mg/kg), and K2O (mg/kg). The prediction is done using a neural network model based on the Rosenblatt perceptron. Data analysis is conducted using the statistical software environment RStudio. The results of the model show the values of the total mean square error (MSE): MSE=0.178 for humus prediction, MSE=0.138 for P2O5 prediction, MSE=0.171 for K2O prediction. Additionally, the program calculated the correlation values between the predicted and calculated soil fertility. K(humus)=0.548, K(P2O5)=0.287, K(K2O)=0.244. Thus, the neural network most accurately predicted soil fertility based on the humus content.