Effective global agricultural land management is crucial for ensuring food security amidst rapid population growth, especially in Northern China's semi-arid and desert regions, where uncontrolled fallowing has led to increased cropland abandonment. Traditional remote sensing methods often face accuracy challenges in these harsh climatic conditions. This study introduces a novel approach to enhance the Normalized Difference Vegetation Index (NDVI) by reconstructing the magnitude image using ground-truth samples and regional-scale Vegetation Health Index (VHI) data. This allows for flexible, pixel-level parameter assignment to detect cropland transitions more accurately. Rather than relying on single-value assessments to differentiate between active and inactive croplands, we employ pixel-wise magnitude images, integrated into the LandTrendr algorithm for trajectory-based change detection, to better address the variability of large, diverse agricultural regions. Tested on the Google Earth Engine (GEE) platform in Inner Mongolia-a representative arid and semi-arid region of Northern China-our method identified the year of cropland abandonment with 82.02 % accuracy and showed a correlation (R2) of 0.6065 between observed and actual abandonment durations. This research extends the applicability of the LandTrendr algorithm, offering a robust solution for optimizing land use change detection across regions with significant climatic variation.