Analyzing cropland fragmentation evolution and driving mechanisms in Shandong through MSPA and landscape metrics integration

Abstract

Abstract As the foundational and core resource for agricultural development, arable land plays a critical role in optimizing regional agricultural resource allocation and ensuring food security. This study employs Morphological Spatial Pattern Analysis (MSPA) and the Landscape Pattern Index (LPI) to comprehensively analyze the erosion and fragmentation of arable land in Shandong Province from 1990 to 2023. Bivariate spatial autocorrelation (BSA), Pearson correlation coefficients and the Self-Organizing Map (SOM) algorithm are applied to examine the relationships among landscape pattern indices and spatial clustering patterns. Finally, the driving factors are explored using Multiscale Geographically Weighted Regression (MGWR). The results demonstrate the following: (1) The results indicate a rapid decline in core cultivated zones across central and southeastern coastal Shandong between 1990 and 2023. The proportion of primary core areas (Level 1) dropped sharply from 82.44 to 30.28% of total arable land, suggesting marked degradation of the agricultural landscape. (2) An assessment of landscape pattern metrics reveals increasing fragmentation of cultivated land. In contrast, the western regions have retained more intact agricultural landscapes, marked by a higher share of cultivated land, more cohesive patch patterns, and stronger spatial aggregation. (3) Bivariate spatial autocorrelation (mean Moran’s I = 0.6516) and Pearson correlation (mean r = 0.9225) both point to a strong positive association between the Aggregation Index (AI) and the Percentage of Landscape (PLAND). Among the six cluster types, Clusters C, D, and F—characterized by large area shares, high aggregation, and strong connectivity—represent the most suitable regions for agriculture. Although the proportions of Clusters C and D declined (from 26.5 to 3.7% and 36.8 to 25.7%, respectively), farmland protection policies contributed to a notable rise in Cluster F, from 17.6 to 27.9% over the study period. (4) The influence of key drivers varies across different landscape metrics. Slope gradient shows the greatest explanatory power for the Perimeter-Area Fractal Dimension (PAFRAC), Patch Density (PD), and AI. Meanwhile, NDVI, nighttime light index, and GDP emerge as primary drivers for Connectance Index (CONNECT), Patch Cohesion Index (COHESION), and PLAND respectively, with regression coefficients ranging [0.43, 0.76], [−0.99, −0.67], and [−0.98, −0.12].

Keywords

• Cultivated land
• MSPA
• Landscape pattern index
• Bivariate spatial autocorrelation
• SOM clustering
• MGWR