Mu Us Sandy Land is a typical semi-arid vulnerable ecological zone, characterized by vegetation degradation and severe desertification. Effectively identifying desertification changes has been a topical environmental issue in China. However, most previous studies have used a single method or remote sensing index to monitor desertification, and lacked an efficient and high-precision monitoring system. In this study, an optimal monitoring scheme that considers multiple indicators combination and different machine learning methods (Classification and Regression Tree-Decision Tree, CART-DT; Random Forest, RF; Convolutional Neural Networks, CNN) was developed and used to analyze the spatial–temporal patterns of desertification from 2000 to 2018 in Mu Us Sandy Land. The results showed that: (a) The random forest model performed best for monitoring desertification based on medium and low-resolution remote sensing images, and the four-index combination (Albedo, NDVI, LST and TGSI) obtained the highest classification accuracy (OA = 87.67%) in Mu Us Sandy Land. Surprisingly, the model accuracy of the three-index combination (NDVI, LST and TGSI) (OA = 85.74%) is comparable to the four-index combination. (b) The TGSI index used to characterize soil information performs well, while the LST is not conducive to the extraction of desertified land in several desertification monitoring indicators. (c) Since 2000, the area of extremely severe desertified land has shown a reversal trend; however, there is significant interannual fluctuation in the total and light desertification land area affected by extreme climate. This research provides a novel approach and a valuable reference for monitoring the evolution of desertification in regional studies, and the results improve the research system of desertification and provide a data basis for desertification cause analysis and prevention.

Biological soil crusts (BSCs) play an essential role in desert ecosystems. Knowledge of the distribution and disappearance of BSCs is vital for the management of ecosystems and for desertification researches. However, the major remote sensing approaches used to extract BSCs are multispectral indices, which lack accuracy, and hyperspectral indices, which have lower data availability and require a higher computational effort. This study employs random forest (RF) models to optimize the extraction of BSCs using band combinations similar to the two multispectral BSC indices (Crust Index-CI; Biological Soil Crust Index-BSCI), but covering all possible band combinations. Simulated multispectral datasets resampled from in-situ hyperspectral data were used to extract BSC information. Multispectral datasets (Landsat-8 and Sentinel-2 datasets) were then used to detect BSC coverage in Mu Us Sandy Land, located in northern China, where BSCs dominated by moss are widely distributed. The results show that (i) the spectral curves of moss-dominated BSCs are different from those of other typical land surfaces, (ii) the BSC coverage can be predicted using the simulated multispectral data (mean square error (MSE) < 0.01), (iii) Sentinel-2 satellite datasets with CI-based band combinations provided a reliable RF model for detecting moss-dominated BSCs (10-fold validation, R² = 0.947; ground validation, R² = 0.906). In conclusion, application of the RF algorithm to the Sentinel-2 dataset can precisely and effectively map BSCs dominated by moss. This new application can be used as a theoretical basis for detecting BSCs in other arid and semi-arid lands within desert ecosystems.

Biological soil crusts (BSCs) are a major functional vegetation unit, covering extensive parts of drylands worldwide. Therefore, several multispectral indices have been proposed to map the spatial distribution and coverage of BSCs. BSCs are composed of poikilohydric organisms, the activity of which is sensitive to water availability. However, studies on dry and wet BSCs have seldom considered the mixed coverage gradient that is representative of actual field conditions. In this study, in situ spectral data and photographs of 136 pairs of dry and wet plots were collected to determine the influence of moisture conditions on BSC coverage detection. Then, BSC spectral reflectance and continuum removal (CR) reflectance responses to wetting were analyzed. Finally, the responses of four commonly used indices (i.e., normalized difference vegetation index (NDVI); crust index (CI); biological soil crust index (BSCI); and band depth of absorption feature after CR in the red band, (BD_red)), calculated from in situ hyperspectral data resampled to two multispectral data channels (Landsat-8 and Sentinel-2), were compared in dry and wet conditions. The results indicate that: (i) on average, the estimated BSC coverage using red-green-blue (RGB) images is 14.98% higher in wet than in dry conditions (P < 0.001); (ii) CR reflectance features of wet BSCs are more obvious than those of dry BSCs in both red and red-edge bands; and (iii) NDVI, CI, and BSCI for BSC coverage of 0%–60% under dry and wet conditions are close to those of dry and wet bare sand, respectively. NDVI and BD_red cannot separate dead wood and BSC with low coverage. This study demonstrates that low-coverage moss-dominated BSC is not easily detected by the four indices. In the future, remote-sensing data obtained during the rainy season with red and red-edge bands should be considered to detect BSCs.