Evaluation of Sentinel-2 and MODIS NDVI Accuracy in Heterogeneous Agricultural Landscapes Using UAV Observations: A Case Study in the Heihe River Basin

The Sentinel-2 and the Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) products are the most widely used datasets for vegetation researches by remote sensing. However, their accuracy and uncertainty require further analysis to enhance their applicability. Spatial scale mismatches between in situ measurements and satellite observations pose challenges for accurate validation. Unmanned aerial vehicles (UAVs) can bridge between plot scale used in field measurements and coarse scale in satellite monitoring. This study assessed the accuracy and precision of MODIS (MOD09GQ) and Sentinel-2 NDVI products using UAV-derived NDVI data collected over six 1.5 km × 1.5 km plots in the middle research of Heihe River Basin, China. The UAV-derived NDVI data were obtained during early, middle, and late crop growth stages and spatially averaged to match the native resolutions of Sentinel-2 and MODIS products for direct comparison. Key statistical metrics, including R², RMSE, MAE, accuracy (A), and precision (P), were calculated to evaluate correlations and discrepancies. Results indicated that Sentinel-2 NDVI outperformed MODIS NDVI, achieving higher R² (up to 0.91) and lower RMSE (as low as 0.09). However, both products exhibited systematic biases: MODIS consistently overestimated NDVI across all ranges, while Sentinel-2 showed overestimation in dense vegetation (>0.8) and underestimation in moderate ranges (0.2–0.8). These discrepancies highlight the influence of spatial resolution and mixed-pixel effects. This study demonstrates the importance of UAV data in enhancing NDVI product validation and addressing scale mismatches. The findings provide valuable insights for improving satellite product calibration and support the application of NDVI products in agriculture and ecosystem monitoring.