Real time and accurate monitoring of forest disturbance plays an important role in maintaining a healthy and stable cycle of forest ecosystem. Constructing and selecting the optimal feature variables is an important step in forest disturbance monitoring.This study focused on three typical types of forest disturbance: forest fires, deforestation, and forest geological disasters. Based on Sentinel-1 and Sentinel-2 data, 116 features of four categories were calculated, including spectral features, texture features, index features, and scattering features. The JM（Jeffries-Matusita） distances were used to evaluate the separability of each feature between disturbed and non-disturbed samples. Features were ranked in descending order of JM distance and sequentially added to a One-class Support Vector Machine（One-class SVM） classifier for disturbance classification and accuracy evaluation. By integrating JM distance and classification accuracy, the best features for monitoring each type of forest disturbance were analyzed, and Principal Component Analysis（PCA） was applied to compress features, constructing optimal features simultaneously applicable to all three disturbance types.The research results indicated that spectral features, texture features, and index features had higher contributions compared to scattering features in the three types of forest disturbance.With the increase of the number of features, the monitoring accuracy tended to stabilize or even decrease after significant improvement. After feature compression, the optimal features simultaneously applicable to all three disturbance types were the first principal component of Band6, Band7, Band8, Band8A, and Band9, the first principal component of mean texture features of Band6, Band7, Band8, Band8A, and Band9, and DVI. The recall rates of samples of forest fire, deforestation, and forest geological disaster based on the compressed optimal features were 90.57%, 75.74%, and 79.07%, respectively, with corresponding F1 scores of 0.927, 0.855, and 0.694. This study provide a theoretical basis and methodological reference for feature selection in forest disturbance monitoring, significantly improving monitoring accuracy and efficiency. It holds important practical significance for the protection and management of forest ecosystems.