Cropland Suitability Prediction Method Based on Biophysical Variables from Copernicus Data and Machine Learning

The goal of this study was to propose and validate a method for predicting cropland suitability based on biophysical variables and machine learning according to an FAO land suitability standard using soybean (Glycine max L.) as a representative crop, aiming to provide an alternative to geographic information system (GIS)-based multicriteria analysis. The peak leaf area index (LAI) and the fraction of absorbed photosynthetically active radiation (FAPAR) from PROBA-V/Sentinel-3 data were calculated according to ground-truth soybean agricultural parcels in continental Croatia during 2015–2021. Four machine learning regression algorithms, including random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB), as well as their combination, were evaluated for predicting the peak LAI and FAPAR on the entire agricultural land in the study area, with RF producing the highest prediction accuracy with an R2 in the range of 0.250–0.590. The translation from K-means classes to the FAO land suitability standard was performed using a relative-based approach, ranking five resulting classes based on their relative mean sums of LAI and FAPAR values. The results of the proposed approach indicate that it is viable for major crops, while cropland suitability prediction for minor crops would require higher spatial resolution, such as vegetation indices from Sentinel-2 imagery.