Development of chlorophyll-a and total suspended solids algorithms for monitoring water quality in the Inanda Dam, Kwa-Zulu Natal, using remotely sensed Landsat 8 and Sentinel 2 imagery.

Abstract

Freshwater is a crucial resource necessary for the survival of all life on earth. Globally, freshwater resources only comprise two percent of all available water on the planet, however it is utilized in various ways that include agriculture, industries, socio-economic development, domestic purposes, drinking water, recreational purposes, and energy production. On a much larger scale, inland water bodies also aid in climate regulation, carbon cycling, habitat provision for various organisms and ensure the maintenance and balance of ecosystems. Despite the many uses and services provided by these freshwater systems, they have been negatively affected by anthropogenic activities and climate change, leading to the deterioration of their quality. It has therefore become imperative to monitor these inland water bodies to ensure the sustainability of these resources for current and future generations. Traditionally, water quality monitoring is done by collecting in-situ water samples which are then analysed in a laboratory. This method, while extremely accurate, is time-consuming, labourintensive, expensive, and does not capture the variations of water quality parameters in a water body. Recent advancements in the field of remote sensing have now made it possible to monitor water quality remotely using satellite imagery. The launch of Landsat 8 and Sentinel 2, with improved specifications compared to prior satellites, makes it possible to retrieve water quality parameters remotely via the use of algorithms. There are many water quality parameters measured to determine the quality of an inland water body, however chlorophyll-a (chl-a) and total suspended solids (TSS) are among the most retrieved parameters using remote sensing methods. This is due to their ability to influence the spectral signature of water and their importance in determining the quality of a water body. This study attempted to develop algorithms to monitor chl-a and TSS in the Inanda Dam, KwaZulu-Natal using Landsat 8 and Sentinel 2 imagery. In-situ water samples were collected from the Inanda Dam and the corresponding Landsat 8 and Sentinel 2 satellite imagery were downloaded. The water samples were analysed to determine chl-a and TSS concentrations and the satellite imageries were pre-processed to obtain radiance and reflectance values. The radiance and reflectance values for individual bands and band combinations were correlated with in-situ chl-a and TSS using a Pearson correlation to determine which band/band ratios can be used to retrieve chl-a and TSS. The band/band ratios which obtained a correlation of 0.8 and more were then used to develop the algorithms to retieve the water quality parameters. The results indicate that Landsat 8 B1 to B5 radiance and reflectance, either individually or as part of a band ratio, obtained high correlations with in-situ chl-a, and the band/band ratios of Landsat 8 B1 to B5 radiance and B1 to B7 reflectance correlated with insitu TSS. Successful correlations were found between in-situ chl-a and Sentinel 2 B1, B3, B4, B5, B6 and B10 TOA reflectance and B1 to B5 BOA reflectance band/band combinations, and in-situ TSS correlated well with band/band combinations of Sentinel 2 B1, B3 to B8a and B10 TOA reflectance and B1 to B5, B7, B11, B12 BOA reflectance. Algorithms were then developed using SPSS with all band/band combinations which obtained a correlation of 0.8 and higher with in-situ chl-a and TSS, however the most accurate algorithms, after validation, were those developed with highest correlated band/band combinations. The Landsat 8 B4:B2 radiance algorithm and B4:B1 reflectance algorithm obtained R2 -values of 0.869 and 0.842, and the Sentinel 2 B3 TOA reflectance algorithm and B1 BOA reflectance algorithm achieved R2 -values of 0.828 and 0.816 respectively when retrieving chl-a remotely. In terms of TSS estimation from satellite imagery, the Landsat 8 B1 radiance and reflectance algorithms acquired R2 -values of 0.955 and 0.966 respectively, and the Sentinel 2 B1:B3 TOA reflectance and B1 BOA reflectance algorithms obtained R2 -values of 0.790 and 0.902. These algorithms were then used to produce surface distribution maps of chl-a and TSS for the Inanda Dam. The Landsat 8 B4:B2 radiance algorithm most accurately mapped chl-a distributions and the Landsat 8 B1 radiance algorithm mapped TSS distributions with a high level of accuracy, however all algorithms were found capable of chl-a and TSS retrieval from Landsat 8 and Sentinel 2 imagery. While the empirical algorithms developed in this study proved viable in the estimation of chl-a and TSS from Landsat 8 and Sentinel 2 imagery for the Inanda Dam, it must be noted that the algorithms are time and site specific. Therefore, more research must be conducted in the field of remote water quality monitoring across South Africa to develop algorithms independent of location and time that can be used to aid in water quality monitoring.