Mapping Soil Organic Matter in Cultivated Land Using Landsat 8 Image and GA-AdaBoost Algorithm

Soil organic matter (SOM) is essential for maintaining soil structure, nutrient supply, and water regulation in cultivated land, significantly impacting agricultural productivity and the health of agricultural systems. However, developing robust inversion methods for SOM using satellite remote sensing technology faces challenges due to the spatial heterogeneity of different land use patterns. This study aimed to improve the accuracy of estimating cultivated land SOM from remote sensing images during the bare soil period. Fifteen spectral features were extracted from Landsat 8 image and the recursive feature elimination based on cross validation (RFECV) was applied to identify the optimal feature combination. Multiple machine learning methods optimized by genetic algorithms (GA) and particle swarm optimization were compared to identify the best method for estimating SOM. The results revealed that the features screened by RFECV showed improved modelling accuracy over unscreened features and that the highest accuracy of estimating SOM, with R2 of 0.66, RMSE of 5.93 g/kg and MAE of 4.70 g/kg, was achieved by the GA-optimized adaptive boosting (GA-AdaBoost) method. Therefore, Landsat 8 remote sensing images acquired during the bare soil period, the combination of RFECV and the GA-AdaBoost method can achieve accurate estimation of cultivated land SOM at the regional scale.