Exploitation of Sentinel-2 for mapping phytoplankton in productive inland waters

The aim of the study is to test the suitability of Sentinel-2 MSI for mapping phytoplankton in turbid extremely productive inland waters with recurrent cyanobacterial bloom that represent a major concern. These blooms can have an influence on health care costs, on the costs associated with the treatment of water intended for human consumption and on the tourism business. The study sites investigated in this work include the three lakes of Mantua (Italy), a system of fluvial lakes characterised by a patchy distribution of phytoplankton. These shallow, nutrient-enriched and turbid waters host different primary producers and undergo pronounced discharge variations. The other site is the Curonian lagoon (Lithuania and Russian Federation) and with a surface of 1584 km2 is largest lagoon of Europe. It is also very shallow with waters readily mixed by wind action, although weak stratification can occur temporarily during the summer. The system is also highly eutrophic due to external nutrient loads from the tributary Nemunas River. The presence of cyanobacterial toxic metabolites, is a major concern for water quality and human health in the region. The study areas are part of two research projects (FP-7 INFORM (copernicus-inform.eu) and BLASCO (projectblasco.it)) which have been supporting field activities in the last three years gathering fundamental data for evaluating Sentinel-2 data. Five days match-ups with satellite acquisitions resulted available for evaluating satellite products in terms of concentrations of chlorophyll-a (chl-a), as proxy of phytoplankton. Field measurements include aquatic optical properties and limnological data that allowed the development of semi-empirical adaptive algorithms and of spectral inversion techniques of bio-optical modelling to retrieve chl-a. Depending on water leaving reflectance the approaches were integrated because bio-optical modelling inversion performed better for lower chl-a concentrations (<10 mgm-3), while adaptive semi-empirical band-ratio like algorithms were more suitable for higher chl-a concentrations (up to 150 mgm-3). Chl-a products obtained from MSI data, after removing the atmospheric effects, showed a good agreement with field measurements (r2 0.9; MAE 5.3 mgm-3; relative RMSE 21\%). The Mantua lakes maps exhibited a high dynamic both in space and time due to the circulation of the fluvial waters. In the Curonian lagoon, the MSI products revealed frequent cyanobacteria blooms with chl-a concentrations higher than 50 mgm-3 with rather numerous surface scum phenomena. The performances of the water quality algorithms are largely depending on the quality of inputs. Reliable estimates are expected when both water leaving reflectance (both in shape and in magnitude) and algorithms calibrations are appropriates. The preliminary results show the capability of Sentinel-2 in describing phytoplankton distribution and concentrations in productive waters.