Abstract:

 Jingyuan County is one of the main production areas of Lycium barbarumL., an important cash crop in Northwest China. It is important to quickly and accurately obtain the information about planting structure and spatial distribution of L. barbarum for local agricultural adjustment and regional economic sustainability. Based on Sentinel-2A image data and objectoriented classification method, a random forest classifier was established, with the planting information of L. barbarum being extracted using spectral and textural features. The results showed that the random forest classification combined spectral features with textural features had the highest accuracy. The overall accuracy was 88.14%, the Kappa coefficient was 0.81, and the user accuracy of wolfberry was 81.03%. The planting area of L. barbarumin Jingyuan was 297.12 km², characterized by large-scale concentrated and sporadic distribution. The planting bases in several towns, including Jing’an, Wuhe, Dongsheng and Beitan, were the main distribution sites. In addition, Sentinel-2A was good ateffectively extracting scattered planted L. barbarum. Our results provided support for the adjustment of planting structure and development and utilization of L. barbarum in Jingyuan County. The method used here provided reference for remote sensing monitoring of L. barbarum in a large area.

Key words: Lycium barbarum, remote sensing extraction, random forest, Sentinel-2A, Jingyuan.