Determinação de clorofila-a utilizando sensor hiperespectral e análise da qualidade da água da Lagoa da Pampulha - MG

Abstract

Monitoring aquatic environments in urban watersheds is essential to understand water quality dynamics and the effects of incident pollution. The objective of this study was to analyze the use of a hyperspectral sensor to monitor chlorophyll-a (chl-a) in Pampulha Lagoon, where sediment inputs and industrial and domestic effluents cause significant impacts. The empirical methodology was employed, where the Support Vector Regressor (SVR), Random Forest Regressor (RFR), XGBoost (XGB), and linear regression (LR) algorithms were tested in constructing a model to estimate chla concentrations in the reservoir. Laboratory experiments were conducted to validate the sensor's methodology, which was used to estimate chl-a concentration through the Raphidocelis subcapitata algae. Uni- and multivariate statistical analyses were applied to secondary and primary data to characterize the spatial and temporal water quality of the reservoir. Based on the statistical analyses of secondary data between 2013 and 2021, it was found that chl-a, cyanobacteria density, and total phosphorus significantly exceeded the limits established by legislation, especially at the upstream sampling point where the main tributaries, Ressaca and Sarandi streams, discharge into the reservoir. Despite interventions with Phoslock® and Enzilimp® products since 2016, there was no reduction in phosphorus levels, the main focus of the treatment, which is considered the main influence on the high rate of primary production in the lagoon. The monthly primary data collected between March 2022 and February 2023 revealed that the region near the main tributaries had the worst water quality indices. The waste retention curtain installed in 2016 also had an influence, with the upstream region exhibiting worse pollution indices. The proximity of the main tributaries and the shallower water column in the region are possible causes of these results. Cluster analysis confirmed these findings and indicated the presence of a fourth compartment in the lagoon, in addition to the three regions represented by the IGAM monitoring points, just after the curtain, which may indicate the need to expand the monitored points in the reservoir. In the laboratory analysis, a high correlation was observed between the images captured by the sensor and chl-a (R2 = 0.9942 and RMSE = 3.42 µg/L) using partial least squares regression. In the field analysis, among the observed models, the algorithm that best adapted to the data pairs, hyperspectral images, and chl-a concentrations, was SVR (R2 = 0.81 and RMSE = 6.87 µg/L). The hyperspectral sensor demonstrated good performance in estimating both monospecific algal samples and predicting chl-a in an environment with high dynamics of sediment and dissolved compound concentrations, representing an alternative for chl-a monitoring in Pampulha Lagoon. However, further studies are needed to analyze whether the results occurred due to the actions of the government, as the dynamics of parameters, especially phytoplankton, can be influenced by various factors such as climate change.