Avaliação da eficiência de modelos de estimativa de clorofila-a por sensoriamento remoto em reservatórios do semiárido brasileiro

Abstract

Estimating the water quality of reservoirs in the Brazilian Semiarid Region (SAB) by remote sensing, using satellite, represents an important monitoring tool, assisting the management of water resources in these environments. Chlorophyll-a (Cla-a) is a parameter commonly estimated by satellite, serving as an indicator of eutrophication. Therefore, this study aimed to assess the performance of 22 spectral models available in the literature for Cla-a estimation in 170 SAB reservoirs. Field data from Public Agencies responsible for monitoring these water bodies were utilized. Satellite images were obtained from Sentinel 2, level 2A and 1C on the Google Earth Engine platform. The Sentinel 2 3 2A data included atmospheric correction by the Sen2cor algorithm, while Sentinel 2 3 1C images underwent correction using the SIAC model (Sensor Invariant Atmospheric Correction). Datasets were analyzed in typologies of groupings: by reservoir volumes, by field Cla-a concentrations, and individually by reservoirs. In order to evaluate the performance of the models in different scenarios, different time periods were used in the analyzes (2015-2022 and 2019-2022). Correlations of Sen2cor-corrected Sentinel 2 images were slightly higher compared to SIAC-corrected images. Model efficiency was low for a significant number of reservoirs, and the quantity of correlations decreased as the temporal period increased. It was concluded that there might be spatial and temporal dependence of the tested models on environments with characteristics similar to the locations where they were generated. Therefore, the challenge of acquiring an effective model to estimate Cla-a by satellite in all reservoirs in the Brazilian semi-arid region still remains, offering support to local water resource managers. However, it is recommended to recalibrate the models for the reservoirs in which the R² showed good performance, as an alternative to reduce the errors evaluated by the NMRSE and NSE, and to enable the safe use of these models in these water bodies.