Investigating the ability of state-of-the-art remote sensing datasets in capturing the spatiotemporal patterns of cropland area: Implications for agricultural monitoring in the largest river basin of China’s arid region

• Vegetation interference index is proposed to quantify non-cropland vegetation impact • Most products show alarming discrepancies in TRB cropland, with errors up to 222.4% • MCD12Q1 (500m) outperforms higher-resolution products (30m) with lowest MPE (6.5%) • MCD12Q1’s superior performance likely stems from its integration of census data • Current products show limited capability in capturing cropland change trends of TRB Accurate cropland distribution information is crucial for sustainable agricultural management and food security. However, how well current remote sensing-based cropland products capture spatiotemporal patterns of cropland area remains unclear, particularly in the Tarim River Basin (TRB), which is China’s largest arid river basin and cotton production base. This study evaluates seven state-of-the-art cropland products (CACD, CLCD, ESA-CCI, GLASS-GLC, MCD12Q1, CLUD-A, and GLAD) in characterizing cropland area dynamics in the TRB from 2000 to 2020. We found substantial discrepancies among them. Despite showing good inter-dataset consistency in temporal variations (mean correlation=0.84), the multi-year average standard deviation (10.4 × 10 3 km 2 ) of these estimates represented 77.6% of the actual cropland area. Most products significantly misestimated the cropland area, with mean percentage errors up to 222.4%. The MODIS land cover product (MCD12Q1) demonstrated the best overall performance, with the lowest root mean square error (169.5km 2 ) and mean percentage error (6.5%), despite its relatively coarse resolution. MCD12Q1’s superior performance likely stems from its integration of census data which better identifies abandoned croplands common in TRB. However, even the lowest trend error among all products (12.2km 2 /yr) was comparable to the mean reference trend (15.3km 2 /yr), highlighting challenges in accurately capturing agricultural landscape changes. Through proposing a novel Vegetation Interference Index (VII) as an assessment tool, we found that mixed vegetation patterns had stronger influence on cropland mapping than topography. Our findings provide crucial implications for developing robust remote sensing-based decision support systems in arid agricultural regions, emphasizing that appropriate remote sensing product selection is essential for precise agricultural management.