Assessment of adjacency effect correction methods for Sentinel-2 MSI observations over mid-latitude lakes

Mortimer Werther, Mariano Bresciani, Claudia Giardino, Abolfazl Irani Rahaghi, Loe Maire, James Runnalls, Daniel Odermatt

Accurate retrieval of remote-sensing reflectance (Rrs) from lakes requires separating water-leaving radiance from atmospheric, surface and adjacency contributions. We assessed adjacency effect correction (AEC) methods for Sentinel-2 MultiSpectral Instrument (MSI) imagery using 329 satellite--insitu matchups from four mid-latitude lakes: Greifensee, Lake Geneva, Lake Garda and Lake Trasimeno. The AEC methods TMART and RAdCor were combined with atmospheric correction (AC) methods ACOLITE, C2RCC and POLYMER, while GAAC was evaluated as an integrated AEC+AC approach. RAdCor+POLYMER produced the lowest median symmetric accuracy (MdSA) in three of the four lakes. The largest AEC benefit occurred in narrow Greifensee, where MdSA was 49.5 percentage points lower for RAdCor+POLYMER than for standalone POLYMER. Prior AEC reduced MdSA in 98 of 144 comparisons, with improvements most frequent at 705 and 740 nm. Rrs retrieval variation among AEC+AC combinations increased towards longer wavelengths: the median normalised interquartile range was 15.9% at 560 nm and 104.4% at 740 nm. Higher aerosol load was associated with larger AEC benefits in several combinations, whereas associations with SZA were generally small and varied among approaches and wavelengths. These results support routine inclusion of AEC in MSI lake Rrs processing, particularly for small or narrow lakes and products that use red and near-infrared bands. However, correction magnitude, sign and AEC benefit depended on water type, aerosol conditions, wavelength and downstream AC approach. Further method development and broader validation are required before AEC+AC combinations provide consistent benefit across inland waters.