Modelos de turbidez, transparência e concentração de clorofila no Reservatório de Três Marias a partir dos dados espectrais dos satélites Sentinel-2

Abstract

Most sutdies of remote sensing involving water quality are applied to water environments that are considerably turbid or highly eutrophic. Três Marias Reservoir is a clear water environment and is classified as ultraoligotrophic to oligotrophic, which is part of the challenge of the study. The main objective of this study consists in developing empirical models to estimate optically active water quality parameters of the Três Marias Reservoir in order to allow its evaluation in a distributed and systematic way from remote sensing data (Sentinel-2 mission). Three parameters were considered: turbidity, Secchi disk depth and chlorophyll-a. The first phase investigated the relationship between the concentration of optically active components measured in situ and the spectral reflectance of water from optical images obtained from the MSI sensor of the Sentinel-2A and -2B satellites. The results confirmed the characteristic of the reservoir as an environment of clear to moderately clear water, with higher spectral reflectance in the 560 nm wavelength (green). With the aid of statistical analyses, the second phase sought to model the three parameters by empirical techniques. Simple and multiple regression models were tested using the MSI bands and band combinations recognized in the specialized international literature. The approach allowed to propose the model with the best performance for each parameters, including a specific model developed for the turbidity of clear water environments. It was concluded that regression models are able to estimate the optical parameters of water quality in the reservoir and characterize how the spatial dynamics can vary over time. The third phase was characterized by the calibration and validation of the selected models. Two specific points of the reservoir (upstream and downstream) were selected to calibrate the models. The results obtained using the consolidated data set for turbidity (r² = 0.80, RMSE = 0.74 FNU), transparency (r² = 0.95, RMSE = 0.70 m) and chlorophyll-a (r² = 0.64, RMSE = 1.24 ug/L) were considered very good, excelent and average respectively. The proposed approach was tested in the Várzea das Flores reservoir where water quality data were acquired, which also returned models with good adjustments for turbidity (r² = 0.62, RMSE = 2, 46 FNU), transparency (r² = 0.94, RMSE = 0.31 m) and chlorophyll-a (r² = 0.80, RMSE = 0.78 ug/L). The research shows that the aquatic environmental monitoring by remote sensing cannot completely replace periodic in situ measurements, but it allows to considerably reduce the quantity and/or frequency of these measurements, facilitating the monitoring of the temporal dynamics of the water body. Remote sensing and its related technologies offer enormous perspectives for the operational monitoring of reservoirs.