Assessment of adjacency effect and its impact on retrieving turbidity over small lakes using Sentinel-2 MSI data

Small lakes are essential for aquatic processes at a local scale, but are often overlooked in monitoring programs. Remote sensing can help bridge this gap, however there is still uncertainties regarding retrievals in small lakes, especially due to adjacency effect. Our objective was to assess the potential effect of adjacency over retrievals of satellite reflectance in four small lakes (4 to 29 ha) in Southwest France, and its propagation on retrieving turbidity. We used GRS for correcting for the atmospheric and sunglint effects, and a semi-analytical algorithm to retrieve turbidity over these low turbidity lakes. With 10 field campaigns during Spring and Summer, when the adjacency is strongest, and 65 match-ups, we found that GRS retrieved very good Rrs for bands 3 to 5 (r = 0.94), with some underestimation. The results for band 2 are also fairly good, but for bands 6 to 8A, we found a significant overestimation of the Rrs (r2 = 0.1 at 783 nm), as a result of adjacency effect. Despite this, the turbidity retrieval performs well using the semi-analytical model with visible bands (r2 = 0.72, RMSE = 3.1 NTU), emphasizing the importance to consider the whole spectrum information for water quality retrieval. We showed that Sentinel-2 MSI can be used to produce time series of water quality parameters for lakes as small as 4 ha, when the NIR bands are not used. Following studies will address the adjacency effect over more turbid lakes, where the NIR bands are necessary to retrieve the parameters.