Estimating soil carbon content using easily obtainable parameters

The successful measurement of carbon sequestration efficiency requires large-scale soil carbon quantification, a costly and time-requiring process. This work aimed to train a machine learning model to estimate the carbon content of farms based on easily acquired data, such as soil texture and management practices. Data from 53 farms spread across four regions of Brazil were used, being preprocessed and split into train, test, and validation sets. We obtained a model capable of estimating soil carbon concentration with a percent root mean square error of 32.9% using only simple inputs. If climate, soil class, and plant residue data were included in the model, we believe the results would be more accurate.