SATELLITE-DRIVENASSESSMENTOFTOTALORGANICCARBON ANDDISSOLVEDSOLIDSUSINGARTIFICIAL INTELLIGENCE: CODEDEVELOPMENTANDIMPLEMENTATION

Abstract

A vital resource for life and the advancement of human
 civilization is water. The current water supplies are facing
 serious problems due to the world's expanding population and
 rising demand for clean water. Large-scale assessment of
 Surface Water Quality Parameters (SWQPs) and routine
 monitoring of water bodies are essential; on the other hand,
 remote sensing offers broad geographical and temporal
 coverage. This research models SWQPs by using ground truth
 water quality data, satellite data, and Artificial Neural
 Networks (ANNs). The surface reflectance value of the water
 area has initially been produced using atmospheric correction
 methods, such as Quick Atmospheric Correction (QUAC),
 Dark Object Subtraction (DOS), Fast Line of Sight
 Atmospheric Analysis of Hypercubes (FLAASH), and
 Atmospheric Correction (ATCOR). The coefficient of
 determination (R2) and Root Mean Squared Error (RMSE)
 values for each atmospherically adjusted technique and the
 Landsat8 reference data were then calculated using a Python
 tool. Second, this research builds a Backpropagation Neural
 Network (BPNN) model based on Landsat8 for modeling
 Total Organic Carbon (TOC) and Total Dissolved Solids
 (TDS) using the most accurate atmospherically adjusted
 technique, FLAASH, with ANN. Thirdly, different epochs
 (800–10,000) were used to train the created BPNN models in
 order to evaluate their effect on the training procedure. The
 models will continue to be accurate and cost-effective if the
 epochs are set to 1200, which will save processing time and
 computational expenses for subsequent projects