Use of Remote Sensing and Chemometrics as Tools to Estimate Parameters of Water Quality in Intermittent Lakes of the Lower Doce River

In this work, chemometric prediction models were developed using remote sensing images associated with limnological parameters to evaluate the water quality of intermittent lakes of the Baixo Rio Doce (Southeast - Brazil). The lakes, popularly known as Lagoa Juparana and Lagoa Nova, are located in areas that were affected by the environmental disaster of the Fundao iron ore tailings dam (Minas Gerais). The visible and near infrared reflectance bands were extracted from images of the lakes surface, which were recorded by the Landsat-8 satellite on the dates closest to the days of field collection. After atmospheric correction of the spectral data, the models were built using Regression by Support Vectors to estimate the water quality parameters, which presented results satisfactory by the correlation coefficient of the prediction (R2pred) and by the square root of the mean squared prediction error (RMSEP), respectively: total phosphorus (0.817; 7.305 mu g L-1), turbidity (0.984; 1.467 UNT), transparency (0.705; 0.785 m), chlorophyll-a (0.850; 0.457 mu g L-1) and developed average trophic state index based on the Carlson equation (0.712; 2.617). This technique enables remote analysis of limnological parameters, which can help in environmental monitoring and equipping managers for more efficient decision-making in water resources conservation actions.