To address the uncertainty in the extraction and dynamic monitoring of cropland cropping intensity (CI) caused by frequent cloud cover, fragmented farmland, and multi-cropping systems in southern China, this study aimed to fully leverage the advantages of multi-scale remote sensing observations to achieve efficient and accurate CI mapping for Hubei Province from 2000 to 2021, and to analyze the spatiotemporal evolution of regional agricultural production patterns.

Time-series NDVI data from 250 m MODIS and 30 m Landsat were integrated using four representative spatiotemporal fusion algorithms: STARFM, ESTARFM, STNLFFM, and GF-SG. Fusion performance was comprehensively evaluated based on spectral fidelity (AD, RMSE) and spatial detail accuracy (Edge, LBP). The optimal algorithm was used to generate a 30 m/8-day NDVI dataset for 2000-2021. Cropland CI was extracted using a phenology-based peak detection method, and then its spatiotemporal variation was analyzed.

Compared with the other three spatiotemporal fusion algorithms, the GF-SG algorithm demonstrated the best performance in both spectral fidelity and spatial detail accuracy (|AD|<0.021, RMSE<0.111; |Edge|<0.55, |LBP|<0.10). The reconstructed NDVI time series using this algorithm improved the accuracy of cropland CI extraction by 0.02%-5.53%. Based on ground samples, the overall classification accuracy of cropland CI in Hubei Province reached 86.60%. From 2000 to 2021, approximately 20%-25% of croplands in the study area experienced CI transitions every five years, with the most significant changes occurring between 2005-2010 (25.79%) and the least between 2010-2015 (20.07%). The dominant transition type shifted from 'single-cropping to double-cropping' (13.49%) in the early years to 'double-cropping to single-cropping' (9.35%) and 'single-cropping to fallow' (4.90%) in the later years.

Over the past two decades, Hubei Province has developed a diversified cultivation pattern dominated by single cropping, with coexistence of double cropping and fallow practices. The evolution of cropland CI has been jointly driven by policy guidance, labor force changes, resource input, and adjustments in cropping structure. By integrating multi-scale remote sensing data from MODIS and Landsat, this study constructed a high spatiotemporal resolution NDVI dataset, which enabled efficient and accurate extraction of long-term cropland CI in complex agricultural landscapes. The findings offered the critical support for agricultural production management and the development of cropland protection policies.