Abstract:

As crucial geographical units within the underlying surface of arid and semi-arid regions， deserts are highly sensitive to climate change and human activities. Timely understanding of the characteristics of large-scale， long-term landscape pattern evolution in desert areas is of significant scientific importance for regional ecological restoration and sustainable development. Using Landsat series satellite imagery from 1980 to 2024 as the data source， various landscape information was extracted via deep learning methods. The spatiotemporal evolution characteristics of desert landscapes over the past 45 years were then systematically analyzed using landscape pattern indices. The study reveals that： （1） The landscape pattern in desert regions exhibited phased evolutionary characteristics. From 1980 to 2000， it was a relatively stable period. The period 2000-2010 marked a phase of significant change， with 14 100.03 km² of grassland landscapes converting to sandy land. Conversely， 2010-2024 was a sandy land recovery period， during which the area of sandy land began to decrease， and 2 138.30 km² of sandy land landscapes gradually converted back to grassland landscapes. （2） Spatially， a distinct differentiation pattern was observed， characterized by natural degradation dominating in the west and significant human disturbance in the east. Sandy land landscapes were highly concentrated in western deserts like the Taklamakan Desert， where they constituted 69.5% of the area with a variation amplitude of less than 0.5%. Conversely， eastern sandy lands （e.g.， Horqin， Mu Us Sandy Land） showed a diverse mosaic of human activity landscapes. （3） Landscape pattern indices revealed that sandy land expansion proceeded in a concentrated and continuous pattern （patch cohesion >99.1%）， while grassland degradation showed characteristics of internal fragmentation （patch cohesion decreased by 0.18）. Overall， desert regions experienced an evolutionary process from stability to homogeneous aggregation （Shannon Diversity Index decreased from 0.8475 to 0.8084）， subsequently adjusting towards increased complexity. This research provides a scientific basis for evaluating the effectiveness of ecological engineering projects and optimizing differentiated prevention and control strategies.

Key words: landscape pattern, desertification, landscape pattern indices, spatiotemporal evolution, desert region