A data-driven approach to flag land-affected signals in satellite derived water quality from small lakes

The land-affected signal in remotely sensed radiance from nearshore waters is a common problem for remote sensing, introducing uncertainty in atmospheric correction and subsequent water quality constituent concentration estimates. This study proposes a new method for identifying effects of land on satellite remote sensing of water quality. The new optical water types (OWT) containing the land-affected signal were derived from POLYMER-corrected imagery of the Medium Resolution Imaging Spectrometer in reduced resolution (MERIS RR) and Sentinel-3 Ocean and Land Colour Instrument (OLCI). These were then applied, as part of a larger set of existing OWTs corresponding to the variability observed in natural waters, to satellite images. The ability to identify pixels containing both water and land, and those contaminated with radiance from adjacent land, was evaluated. Our test sites include dark lakes of varying size in Sweden (Lakes Rusken, Bolmen, Ringsjo center dot n, and Ivo center dot sjo center dot n) where the classification showed high sensitivity to land near the lake shore. The land-affected signal is shown to lead to underestimations of chlorophyll-a concentration and Forel-Ule colour indices, and overestimations of turbidity in these lakes, which can be corrected after masking out the land-affected pixels. The land-affected signal is strongest in summer, both NDVI and sun zenith angle covaried with the seasonal variation of land-affected signal. Further, the results confirmed that satellite images with coarser spatial resolution are more prone to land-affected signal compared to images with finer spatial resolution, for small inland water bodies. We propose a data-driven approach for water quality processing with 'land-affected water types' as an effective way to improve the lake optical water quality monitoring from water colour sensors.

Automated summary

This study proposes a new method to identify the effects of land on satellite remote sensing of water quality. The results show that the land-affected signal leads to underestimations of chlorophyll-a concentration and Forel-Ule color indices, and overestimations of turbidity.