Arid and semi-arid regions, characterized by fragile ecosystems and concentrated land-use pressures, have become high-risk areas of cropland abandonment, posing a significant threat to regional agricultural productivity and sustainable land use. This study constructs a 10 m resolution dataset of cropland abandonment for the period 2016–2022 and, drawing on the Coupled Human–Natural Systems framework, interprets cropland abandonment and systematically investigates its spatio-temporal patterns and driving mechanisms in Inner Mongolia. This paper defines cropland abandonment as farmers, under the pressures of land degradation and environmental constraints and in order to maintain the relative stability of the coupled human–land system, reducing or ceasing cultivation to re-seek a balance between socio-economic returns and ecological carrying capacity, which constitutes a self-organizing adaptive behaviour and response mechanism to degradation. The results show that the mean annual area of abandoned cropland in Inner Mongolia is 639.67 km 2, exhibiting an overall declining trend and pronounced spatial clustering. Precipitation is the dominant natural limiting factor, while accessibility (distance) and production conditions are the main anthropogenic drivers. Further analysis shows that the synergy between mean annual precipitation and total agricultural machinery power has the strongest explanatory power, and that the explanatory power of any pairwise interaction between factors exceeds that of single factors, consistent with the CHANS framework’s expectation of non-linear human–land coupling. The study also delineates four major abandonment hotspots, conducts regional attribution of abandonment, and proposes targeted management strategies. These findings strengthen our understanding of the spatial heterogeneity and driving mechanisms of cropland abandonment in arid and semi-arid regions, and provide theoretical support and empirical evidence for the interpretation and governance of cropland abandonment.