Retrieving chlorophyll-a levels, transparency and tss concentration from multispectral satellite data by using artificial neural networks

The construction of the Hydroelectric Power Plants has influenced the environment with their accompanying dams and reservoirs, resulting in changes to these ecosystems. Monitoring the reservoirs is important for the decision-making process. In this paper, we present different methodologies for estimating the concentration of Chlorophyll-a, transparency and total suspended solids in reservoir from remote sensing multispectral data, based on artificial neural networks and Leave-One-Out cross-validation. Simulations on different sample station of the study area were performed to determine a relationship between these physico-chemical parameters and the spectral response of the reservoir. Based on the in situ measurements, empirical models are established in order to relate the reservoir reflectance measured by Landsat 7 Enhanced TM+ with the water optical parameters. Four images for each year from 2007 to 2014 are calibrated and atmospherically corrected. Statistical analysis using error estimation is employed, aiming at evaluating the most accurate methodology. Artificial neural networks are trained by hydrological cycle and are shown to be useful in estimating the physico-chemical parameters of water from reflectance of visible and NIR bands of satellite images, with better results for the period with little rain and few clouds in the analyzed region.