ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume V-3-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2020, 401–408, 2020
https://doi.org/10.5194/isprs-annals-V-3-2020-401-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2020, 401–408, 2020
https://doi.org/10.5194/isprs-annals-V-3-2020-401-2020

  03 Aug 2020

03 Aug 2020

ESTIMATION OF WATER QUALITY IN A RESERVOIR FROM SENTINEL-2 MSI AND LANDSAT-8 OLI SENSORS

F. M. C. Pizani1, P. Maillard1, A. F. F. Ferreira2, and C. C. de Amorim2 F. M. C. Pizani et al.
  • 1Department of Geography, Universidade Federal de Minas Gerais, Belo Horizonte, Brasil
  • 2Department of Sanitary and Environmental Engineering, Universidade Federal de Minas Gerais, Belo Horizonte, Brasil

Keywords: Water Quality, Statistical Models, Regression Analysis, Reflectance, Landsat-8 OLI, Sentinel-2 MSI

Abstract. The low operational cost of using freely available remote sensing data is a strong incentive for water agencies to complement their field campaigns and produce spatially distributed maps of some water quality parameters. The objective of this study is to compare the performance of Sentinel-2 MSI and Landsat-8 OLI sensors to produce multiple regression models of water quality parameters in a hydroelectric reservoir in Brazil. Physical-chemistry water quality parameters were measured in loco using sensors and also analysed in laboratory from water samples collected simultaneously. The date of sampling corresponded to the almost simultaneous overflight of Sentinel-2B and Landsat-8 satellites which provided a means to perform a fair comparison of the two sensors. Four optically active parameters were considered: chlorophyll-a, Secchi disk depth, turbidity and temperature (the latter using Landsat-8 TIR sensor). Other six optically non-active parameters were also considered. The multiple regression models used the spectral reflectance bands from both sensors (separately) as predictors. The reflectance values were based on averaging kernels of 30 m and 90 m. Stepwise variable selection combined with a priori knowledge based on other studies were used to optimize the choice of predictors. With the exception of temperature, the other optically active parameters yielded strong regression models from both the Sentinel and Landsat sensors, all with r2 > 0.75. The models for the optically non-active parameters produced less striking results with r2 as low as 0.03 (temperature) and as high or better than > 0.8 (pH and Dissolved oxygen).