Extracting tea plantations in complex landscapes using Sentinel-2 imagery and machine learning algorithms

Chen, Panpan and Zhao, Chunjiang and Duan, Dandan and Wang, Fan (2022) Extracting tea plantations in complex landscapes using Sentinel-2 imagery and machine learning algorithms. COMMUNITY ECOLOGY, 23 (2). pp. 163-172. ISSN 1585-8553 (print); 1588-2756 (online)

Text s42974-022-00077-8.pdf Restricted to Repository staff only Download (1MB)

Abstract

Tea tree is an economically important crop. The rapid and efficient mapping of the distribution and dynamic changes in tea plantations informs decision making for government departments. It also plays an important role in rational tea planting and environmental governance. This study used the 10 m Sentinel-2 image data to map the spatial distribution of tea plantations in Yingde City, China. In this article, we analyzed the differences in vegetation and texture characteristics between the new and mature tea plantations. We found that the texture features of the new and mature tea plantations were significantly different in contrast, which can be used as an index for tea plantation extraction. Moreover, we selected machine learning classifiers, including support vector machine (SVM) and random forests (RF) method, which were utilized to extract the preliminary classification to complete spatial distribution mapping of tea plantations. The overall accuracy of SVM and RF was 90.79% and 89.42%, respectively, and the kappa coefficient was 0.88 and 0.86, respectively. SVM had the highest overall accuracy in terms of tea plantation distribution at the regional scale. These results demonstrate that: (1) separating tea plantations into mature and new tea plantations, taking into account vegetation and texture features such as soil brightness index and contrast, will help improve the accuracy of tea plantation classification and (2) using multi-period images combined with machine learning classification methods can improve the efficiency and accuracy of tea plantation identification.

Actions (login required)

Edit Item