Object-oriented tea plantation extraction based on GEE: The case of Shuangjiang County, a typical hilly mountainous area in the subtropical monsoon region of south China

Scientifically understanding the spatial distribution of tea plantations is important for preserving the ecological environment and maintaining the sustainable development of agricultural economics. Using the Google Earth Engine with Shuangjiang County, a typical hilly and mountainous area in the subtropical monsoon zone in southern China, Sentinel-2 remote sensing image data were used to construct spectral, vegetation index, texture, and topographic feature sets and combined with the simple non-iterative clustering and machine learning algorithms Random Forest (RF) and Support Vector Machine (SVM) to realize the object-oriented extraction of the tea plantation. This was compared to pixel-based extraction methods for accuracy. Results indicate that, compared to pixel-based methods, object-oriented extraction demonstrates superior performance and higher accuracy in tea plantation extraction. Regardless of whether pixel-based or object-oriented extraction was employed, the RF algorithm outperformed the SVM algorithm. The object-oriented RF method yielded the highest accuracy at 94.9%, a producer accuracy of 86.5%, and an accuracy of 84.2% for tea plantation extraction. This underscores the favorable application advantages and potential of the object-oriented approach and RF algorithm in remote sensing monitoring and extraction of tea plantations. These results can serve as a reference for identifying tea plantations in similar hilly areas and provide decision support for local tea tree cultivation and management.