| Authors |
Zhengdong Wang, Bingfang Wu, Miao Zhang, Hongwei Zeng, Leidong Yang, Fuyou Tian, Zonghan Ma, Hantian Wu |
| Intro |
Biological soil crusts (BSCs) covering drylands worldwide are an important functional vegetation unit and play an important role in the carbon and nitrogen cycling of dryland ecosystems. However, there is a lack of accurate data on the spatial distribution and change of BSCs. Here, two indices were developed to enhance BSC mapping, including the sandy land ratio crust index (SRCI) for sandy land with moss-dominated BSCs and the desert ratio crust index (DRCI) for desert with lichen-dominated BSCs. The SRCI combines the near infrared band in which moss-dominated BSCs and other surface components have significant spectral differences, and red edge bands in which such differences can be enhanced. Another index, the DRCI, highlights different spectral characteristics of lichen-dominated BSCs, gravel and sand in the shortwave infrared and red-edge bands. A random forest (RF) machine learning framework was developed to map the large-scale BSC distribution with 10 m resolution using Sentinel 2 satellite data. The results show that the BSC area in the Mu Us Sandy Land and Gurbantunggut Desert of China covers 10,647 km2 and 14,036 km2, accounting for 5.7% and 29.4% of the total area, respectively, indicating that BSC is one of the dominant land cover types in the Gurbantunggut Desert. Validations and comparisons demonstrate that the overall accuracy of BSC detection is 90% and 94% with kappa coefficients of 0.80 and 0.86 in sandy land and desert, respectively, and the proposed SRCI and DRCI increase the overall mapping accuracy by 6% for sandy land and desert compared to the non-index scheme. Therefore, indices can enhance the BSC mapping accuracy for sandy land and desert. Moreover, this work provides basic data for exploring the ecological effects of BSCs in arid areas. |
English
中文