Assessment of adjacency-effect correction for satellite-derived reflectance and water quality in agriculturally impacted rivers: a case study in Eastern Canada

The adjacency effect (AE) is a major challenge for optical remote sensing of small inland waterbodies. We evaluated satellite-based retrieval accuracy for two rivers in agriculturally intensive Eastern Ontario, Canada, namely the Ottawa River and its tributary, the South Nation River. Satellite-derived reflectance and water quality parameters, with and without AE correction, were compared to in situ measurements. Applying T-Mart AE correction to Sentinel-2 MSI and Landsat OLI/OLI-2 imagery significantly improved ACOLITE-derived water reflectance, reducing the average RMSE and two bias metrics across all bands by 19.4 %, 32.2 %, and 24.1 %, respectively, with the largest improvements in the red-edge and near-infrared bands. Turbidity was well retrieved using the 705 nm MSI band, achieving an RMSE of 5 FNU across a 2–72 FNU range. While AE correction minimally affected this band, it reduced RMSE of the more affected 783 nm band by 48.5 %, which is important for estimating turbidity in highly turbid waters. AE correction also improved the retrieval of chlorophyll-a concentrations (Chl-a) and colored dissolved organic matter absorption (CDOM) at 440 nm, though performance remained relatively poor (RMSE of 30 μg/L for Chl-a and 1.26 m−1 for CDOM within respective ranges of 2–88 μg/L and 2.0–5.2 m−1). Simulated reflectance based on in situ measurements helped explain these retrieval outcomes. Further improvements in CDOM retrieval may require enhanced atmospheric correction in the visible range, while accurate Chl-a retrieval likely requires hyperspectral sensors with high signal-to-noise ratios between 600 and 800 nm, where reflectance spectra of optically complex waters are sensitive to changes in Chl-a.