Remote sensing identification of seasonal pasture based on Sentinel-2

Shiqi Su, Panfei Fang, Rui Qian, Pengfei Zheng, Zhenping Qiang

Abstract: In order to research the distribution of seasonal pastures, we used the Sentinel-2 remote sensing data of Golog Tibetan Autonomous Prefecture in 2017–2020 warm season (270-300d) to construct 22 spectral features and topographic features based on GEE platform. Random forest classifier was used to identify seasonal pastures. The results showed that: (1) the overall accuracy and Kappa coefficient of seasonal pasture identification were 82.41% and 0.74, respectively; The producer's accuracy and user's accuracy were (cold season pasture: 84.08%, 80.57%; Warm season pasture: 71.92%, 75.26%); (2) The total area was about 7320972.81 hm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , including 3080067.88 hm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of cold-season pasture and 4240904.93 hm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of warm season pasture. The cold season pasture was mainly distributed in Banma, Jiuzhi and Maqin counties, while the cold-season pasture was mainly distributed among Maduo, Dari and Gande counties. (3) The imagery of the end of the warm season (270-300d) during 2017–2020 were synthesized to ensure the integrity of the image of the study area as much as possible and eliminate the influence of clouds;(4) The water loss and surface exposure to herbage after grazing and treading by livestock are similar to the carbonized surface after fire. Therefore, the spectral characteristics of NDVI and NDMI which sensitive to water and CSI, NBR and NBR2 which sensitive to bare surface have good effects on the identification of seasonal pasture. (5) Most of the areas with high altitude belong to other types, indicating that elevation factors perform well in distinguishing seasonal pastures from other types of land. The results of this study can provide a reference to using remote sensing images to quickly identify seasonal pasture distribution and formulate scientific grazing management policies.