Suitability of multispectral satellites for deriving water constituents in high-altitude lakes: A case study from Peru

Abstract

Water quality monitoring in lakes of difficult access can be potentially supported through remote sensing. This study examined the feasibility to derive water constituents using multispectral data from Sentinel-2, Landsat-8/9 and PlanetScope SuperDoves of three high-altitude lakes in the Peruvian Andes: Junín (4085 m.a.s.l., mesotrophic), Lasuntay (4650 m.a.s.l., oligotrophic) and Huacracocha (4460 m.a.s.l., eutrophic). Field spectrometer measurements and water sample analysis for chlorophyll-a were carried out for validation. Atmospheric correction of all satellite images was performed with ACOLITE. WASI-AI was used to correct for reflections at the water surface (sun and sky glint) and errors of the atmospheric correction and to derive the concentrations of Chl-a, TSM and CDOM from both spaceborne and in-situ spectral data. The algorithm’s neural network was trained for the deep water areas of each lake. Sentinel-2 images lead to reliable Chl-a and to plausible TSM and CDOM results for the lakes Junín and Lasuntay. For the latter lake, Landsat-8/9 produced good results as well, with slightly higher TSM concentrations. For lake Huacracocha, its very dark water with high Chl-a concentration made it difficult to obtain reliable results from Sentinel-2. Reasonable results were however achievable with Landsat-8/9, in spite of having less NIR channels than Sentinel-2, and should thus be further studied. The bands of PlanetScope SuperDoves were spectrally too inconsistent, with large differences between the individual satellites, to allow reliable data analysis in any of the studied lakes.