Mapping land degradation and sand and dust generation hotspots by spatiotemporal data fusion analysis: A case‐study in the southern Gobi (Mongolia)

Abstract

The ongoing desertification and aeolian erosion processes in the southern Gobi Desert are ranked as one of the most significant global environmental disasters. In this study, we analyzed the decadal progress of eolian erosion in the southern Gobi Desert and traced key factors controlling intensified land degradation (LD) and sand and dust (SD) generation employing satellite data and climatic variables. Columnar dust mass density from climatic data re‐analyses as a major SD tracer was combined with the Mann–Kendall (MK) method and the empirical orthogonal function processor. Validation was performed by using ground data sets and field evidence from reference locations. The results revealed that (1) LD/SD patterns and hotspots in the Gobi Desert are significantly controlled by the distribution and trend of precipitation; (2) climatic conditions in the Mongolian Gobi Desert have shifted towards an unfavourable direction with respect to the LD/SD occurrence; (3) surface conditions in southeastern Mongolia have somehow decoupled from the weather factors and the transition zone between the desert and the vegetated terrain has gradually expanded probably due to anthropogenic activities. The correlation analyses between all candidate driving factors of LD/SD indicated that a major control mechanism of spatiotemporal migration of LD/SD in the southern Gobi Desert is the change in precipitation, whereas anthropogenic activity holds a secondary control. The results obtained can be used to prioritize intervention zones in the frame of land use planning processes aimed at adapting to climate change and mitigating LD and SD generation in source areas.

Implementation of a multi‐layered land degradation monitoring and forecasting framework toward integrated countermeasure planning in drylands

“…Evaluating the efficiency of the potential input feature is desirable in order to attain higher‐quality ML prediction. Although we did not assess potential alternative candidates input features, but followed the precedent studies regarding the environmental drivers of LD for input feature selection (Darmenova et al, 2009; Kim et al, 2022), the use of the mean decrease in impurity (MDI, Han et al, 2016) of the employed features was demonstrated (Table 4). The higher the reduction in impurity achieved by a feature split, the more important that feature is considered to be.…”

“…Thus, Layer 0 provides base datasets for the modeling, monitoring, and forecasting components included in Layers 1 and 2, besides producing datasets directly related to end‐users such as policymakers and planners (Table 1). Refer to Kim et al (2022) for using cloud‐based EO datasets in LD mapping. Since these datasets are available in varied native resolutions, they are resampled at 250 m spatial resolution for interoperability and final processing.…”