Atmospheric Correction Intercomparison of Hyperspectral and Multispectral Imagery over Agricultural Study Sites

Atmospheric correction (AC) is necessary for deriving accurate spaceborne estimates of surface reflectance (SR) for agricultural fields. In this effort, we evaluated the consistency of various AC tools applied to WorldView-3 (WV3) and PRISMA imagery including MODTRAN, FLAASH, and ASI’s Level2D AC. Overall, these AC tools produced consistent SR estimates compared to top-of-atmosphere reflectance (TOAR). MODTRAN and FLAASH were highly consistent when applied to WV3 imagery, yielding SR relative percent differences <10%, and SR value differences <|0.005| for all bands except the coastal blue. For a case study of cover-cropped fields, normalized difference vegetation index (NDVI) values derived from PRISMA and WV3 SR imagery showed only slight variation between different AC tools (|0.00| to |0.05|). Observed NDVI differences between TOAR and SR (|0.06| to |0.20|) and change in NDVI from cover crop termination (up to |0.70|) were greater than AC tool differences. For a crop residue case study, shortwave infrared imagery was converted to spectral indices (SIs) and used for fractional crop residue cover (f R ) estimation. The shortwave infrared normalized difference residue index (SINDRI) demonstrated good f R estimation performance (R 2 = 0.88) with minimal f R RMSE differences (<0.005) for various AC tools, image sources, and SR vs. TOAR. SIs computed with a 2100 nm band, like the cellulose absorption index (CAI) yielded top performance in f R (R 2 = 0.89 to 0.90) but could only be derived from PRISMA imagery. Although SIs covering the wavelength range of 2035-2265 nm performed well, the lignin cellulose absorption index (LCA) - featuring a 2330 nm band - produced larger f R RMSE differences (0.018 for AC tools and 0.035 for SR vs. TOAR). For SINDRI, AC tool type did not impact f R estimation substantially, but AC was found to be critical for deriving stable coefficients for SI-to-f R calibrations, a finding that highlights the importance of AC for time series f R applications like tillage monitoring.