Estimating Soil Organic Carbon Using Sensors Mounted on Unmanned Aircraft System and Machine Learning Algorithms

Abstract

Predicting Soil Organic Carbon (SOC) accurately and generating SOC distribution map holds potential for assisting farmers in assessing soil fertility, optimizing and enhancing the resource use efficiency. This study used Mica Sense Red Edge sensor mounted onboard Idea forge Q4c Unmanned Aerial System (UAS) to assess the distribution of SOC in the experimental site. Random Forest (RF) and Support Vector Machine (SVM) techniques were developed with both UAS as well as Sentinel datasets for SOC prediction. Overall, the UAS dataset exhibited greater accuracy in prediction of SOC compared to Sentinel Datasets. Random forest model provided an accurate prediction of SOC when used with the UAS dataset (RPD = 1.09, R²CV = 0.25, RPIQ = 2.57 and RMSECV = 0.06), whereas the Sentinel 2A dataset provided a better prediction of SOC with SVM model (RPD = 0.96, R²CV = 0.10, RPIQ = 0.96 and RMSECV = 0.07). The prediction map of SOC was generated using the UAS dataset with the RF model because it was found to be more accurate compared to the Sentinel and SVM model. The accuracy assessment indicators indicated that UAS based SOC prediction is having the potential in achieving more accurate predictions of SOC, which will offer an optimized agricultural practice and insights for supporting informed decision-making.