Water Quality Assessment Using Sentinel-2 Imagery Estimating Chlorophyll A, Secchi Disk Depth, and Cyanobacteria Cell Number in Brazilian Reservoirs

Abstract

: Satellite images were used to assess surface water quality based on the concentration of chlorophyll a (chla), light penetration measured by the Secchi disk method (SD), and the Cyanobacteria cells number per mL (cyano). Nine reservoirs are studied in S ã o Paulo State (Brazil); six reservoirs are interconnected, comprising the Cantareira System (CS), and three others are isolated, the Broa, Salto Grande (SG) and Itupararanga (Itu) Reservoirs. For this study, Sentinel-2 images were employed, alongside SNAP image processing software, and the native products conc_chl and kd_z90max, treated using Case 2 Regional Coast Color (C2RCC) atmospheric correction. The database for chla, SD and cyano was obtained from CETESB, the agency legally responsible for operation of the Inland Water Quality Monitoring Network in S ã o Paulo State. For CS, the results demonstrated robustness in the estimates of chla (RMSE = 3.73; NRMSE% = 19%) and SD (RMSE = 2,26; NRMSE% = 14%). Due to the strong relationship between cyano and chla (R 2 = 0.84, p < 0.01, n = 90), both obtained from field measurements, it was also possible to estimate cyano, based on the estimates of chla from the satellite images. For CS, the estimates revealed a clear pattern, with the upstream reservoirs being more eutrophic, compared to those downstream, particularly due to the high cyano. For Broa, a high corre-lation was also observed between chla and cyano (R 2 = 0.6052, RNMSE% = 27, n = 8). Based on the estimates, Broa showed a eutrophic pattern in practically the entire year of 2020, with a predominance of cyanobacteria in the entire water body (from 10,000 to 20,000 cells/mL). For SG, it was possible to observe robustness only for DS, but not for chla. The restricted database available was considered the main explanatory factor for the low robustness observed for (SG), despite the relationships between the field data. For Itu, the C2RCC-Nets demonstrated robustness in the estimates of Chla (RMSE = 4.0 mg/m 3 ; NRMSE = 16.7%) and SD (RMSE = 0.78 m; NRMSE = 19.1%). Despite the good fit of the allometric relationship relating the Chla and Cyano field data, it did not allow validation of the cyano estimates using the conc_chl native S2 product, for Itu. Thus, it is concluded that automatic products are excellent tools for estimating chla and SD, and as a result of the solid relationships between chla and cyano, it is possible to estimate the cyano and observe spatial heterogeneity in water quality, based on SD, cyano, and chla.