Can the MODIS EVI formula be directly applied to other sensor data? A case study with Landsat data

The question of whether the MODIS enhanced vegetation index (EVI) can be directly applied to other satellite sensors has remained unexplored for nearly three decades since its development. This is important because the EVI's empirical parameters are specifically designed for MODIS. For a long time, users have applied these MODIS-based parameters directly to other sensor data without modification. Therefore, this study systematically explores this issue using Landsat-8/9 data. We introduce a novel methodological framework: using NDVI (parameter-free) as a reference benchmark and designing an auxiliary index (EVIb) to isolate the impact of the blue band, as it is introduced in EVI. Results show that, based on the synchronized observations of MODIS and Landsat-8/9 across eleven vegetation test areas, the deviation between MODIS and Landsat-8/9 EVI is, on average, 1.7 times larger than that of their NDVI, with an R 2 deviation reaching 10%. Analysis confirms that the blue band is not the primary cause of the deviation; instead, the discrepancy arises from the inapplicability of the MODIS-specific parameters to Landsat-8/9 data. Sensitivity analysis reveals strong interactions among the four parameters of EVI, with the soil adjustment factor (L) being the most sensitive. Based on these findings, the study proposes and tests two parameter optimization strategies for Landsat-8/9 EVI: a site-specific cross-calibration method that reduces the mean absolute percentage bias (|PB|) between the MODIS and Landsat-8/9 EVI from 13% to nearly 0%, and a generalized parameter optimization using the Nelder-Mead algorithm, which reduces the average |PB| to 8.33%. Cross-validation indicates that the optimized parameters can reduce bias by approximately 32.5% in unseen areas from similar biomes. This study concludes that the direct cross-sensor application of MODIS EVI parameters introduces significant bias. We recommend sensor-specific parameter optimization and provide a transferable methodological framework to guide future efforts towards consistent multi-sensor vegetation monitoring.