Linking random forest and auxiliary factors for extracting the major economic forests in the mountainous areas of southwestern Yunnan Province, China

Forests are generally extracted from remotely sensed images based on the spectral features, ignoring other important auxiliary information, and the techniques of precise extraction need to be further improved. By using the Sentinel–2 image and auxiliary factors (AFs) including site conditions (SCs) and vegetation indices (VIs), the random forest model with AFs (RF–AFs) was adopted for the extraction of the economic forests in Lancang County, which is a mountainous area with rich biodiversity and is witnessing rapid development of economic forests in Yunnan province of China. The results obtained using the RF–AFs model were compared with those obtained using the random forest model without AFs (RF). The results were as follows: (1) The kappa coefficient for extracting the first–level land use obtained using the RF model was 0.9531. Lancang County is dominated by forests, accounting for 73.76% of the total area. (2) After parameter optimization, the RF–AFs model yielded the highest accuracy in the extraction of the second–level forests, with a kappa coefficient value of 0.9493, which was 14.69% higher than that of the RF model. Thus, the RF–AFs model is more suitable for the precise extraction of economic forests. (3) The evaluation results of the factors’ importance of the RF–AFs model showed that the cumulative importance values of SCs such as temperature (TEM), elevation (EL), precipitation (PRE) and VIs such as plant senescence reflectance index (PSRI), enhanced vegetation index (EVI), transformed soil–adjusted vegetation index (TSAVI) was 76.09%, indicating that they were the main factors for the extraction of economic forests. (4) Economic forests are dominated by Simao pines in Lancang County, which are mainly distributed in the central, southwestern and northern regions, accounting for 31.37% of forests area. The proportion of tea plantations, eucalyptus, and rubber trees is 9.05%, 6.71%, and 3.05% of forests area, respectively. The RF–AFs model is conducive for precisely extracting the economic forests and is thus of great significance in studying the ecological and environmental effects of economic forests, performing forestry management, and maintaining regional ecological security.