Mapping of total suspended solids using Landsat imagery and machine learning

Abstract

The main objective of this work is to propose a new technique for water quality parameters monitoring by applying artificial intelligence methods to optimize remote sensing data processing. A multiple regression model was developed to create a total suspended solids (TSS) prediction model, using unsupervised machine learning. Currently, water bodies throughout the world are poorly supervised in terms of quality, so it is necessary to implement efficient mechanisms to obtain synoptic information for a good diagnosis in TSS evolution, because they are a key indicator of the biophysical state of lakes and an essential marker for continuous monitoring. Conventional methods used to monitor the physical parameters of water bodies, for example, in situ sampling, have proven impractical due to time, cost and space constraints, and remote sensing tools can help to achieve this purpose more efficiently. The proposed multiple regression model requires calibration and to that end, Lake Chapala data from the monitoring time series collected by the National Water Commission (CONAGUA) were used. Lake Chapala is the largest freshwater body in Mexico, and the human intervention that develops around the lake has caused drastic changes such as decrease in the size of the lake and increase in suspended matter and aquatic vegetation. These changes alter the balance of the system, endangering the health of the lake. This work presents a generalized semi-empirical model that uses Landsat image data and machine learning methods for estimating total suspended solids (TSS) in water bodies, with a good prediction precision (R = 0.81, RMSE = 32.52).