Remote-sensing monitoring of rubber plantations using object-oriented characteristics from the vigorous period

Natural rubber is one of the four major industrial raw materials in China, and the demand for it is increasing rapidly in China. However, due to geographic and climatic limitations, the rate of natural rubber production in China is significantly less than required to satisfy this demand. Therefore, to ensure the healthy development of China’s rubber industry, it is urgent to develop a method to rapidly and accurately monitor the planting and distribution of rubber forests in China. Existing studies have exploited the unique phenological characteristics of rubber plantations manifested by the spectral characteristics of the vigorous growth period and the deciduous period to identify rubber plantations through remote sensing. Unfortunately, the cloudy and rainy climate of rubber-growing regions makes it is difficult to obtain remote-sensing data in the optimal period. Therefore, the present study uses Planet images as the basic data and extracts typical spectral and textural features of rubber forests during the single vigorous growth period to construct a remote-sensing method to monitor rubber forests based on the object-oriented random forest algorithm, which uses all the spectral and texture features. The results show that the proposed random forest classification model achieves a high classification accuracy: the total classification accuracy reaches 89%, and the rubber producer’s accuracy and user’s accuracy both exceed 92%. This study can effectively solve the problem of lack image data with cloudless from deciduous period and provides a good theoretical basis for remote-sensing monitoring of China’s rubber forests, thereby facilitating the development of China’s rubber industry.