The Estimation of Forest Carbon Storage in Northwest Yunnan Based on Different Machine Learning Classification Algorithms

Taking northwest Yunnan as the study area, this research utilizes the Google Earth Engine(GEE) platform and applies four machine learning algorithms—Random Forest(RF), Support Vector Machine(SVM), Decision Tree(DT), and K-Nearest Neighbors(KNN)—to classify the region into five land cover types: forest land, cropland, built-up land, water bodies, and others. By integrating Landsat 8 multispectral remote sensing data and Yunnan Province's forest resource inventory data, 23 significantly correlated feature factors were selected, and five principal components were extracted through factor analysis. Subsequently, four regression models—Random Forest Regression, Support Vector Machine Regression, ExtraTrees Regression, and LightGBM Regression—were employed to estimate above-ground forest carbon storage, with the optimal model selected for carbon storage estimation.The results indicate that among the four machine learning classification algorithms, the Random Forest classifier achieved the highest classification accuracy, with an overall accuracy of 0.88 and a Kappa coefficient of 0.84, meeting the requirements for subsequent carbon storage estimation. Regarding the four carbon storage estimation models, the Random Forest Regression model demonstrated the best performance, with an R² of 0.89, RMSE of 9.22, rRMSE of 15%, and MAE of 2.75, showing strong model fitting capabilities. The total above-ground carbon storage of forests in northwest Yunnan was estimated at 371.7 Tg, with an average carbon storage of 42.64 t/hm². Specifically, the carbon storage in Dali Bai Autonomous Prefecture, Diqing Tibetan Autonomous Prefecture, Lijiang City, and Nujiang Lisu Autonomous Prefecture was 109.61 Tg, 96.7 Tg, 86.44 Tg, and 78.94 Tg, respectively.Among the various machine learning algorithms, the Random Forest-based model exhibited the highest accuracy for estimating above-ground forest carbon storage in northwest Yunnan. The integration of machine learning classification and carbon storage estimation demonstrates clear advantages and holds promising application potential.