Coupling artificial intelligence techniques and remote sensing data for water quality simulation of lakes

Freshwater lakes are a major resource for human populations. To support water quality (WQ) management for lakes, both WQ monitoring and WQ modeling are essential. Conventional approaches, such as process-based models, are usually used for WQ modelling, however, these approaches require a large number of data (meteorological, topographical, hydrological, and WQ data) with high computational demands. Recently, artificial intelligence (AI) techniques are increasingly recommended in WQ modelling to tackle these challenges. In this study, the application of AI techniques for simulating/predicting water quality for large lakes using remote sensing (RS) is investigated. Specifically, the study aims to develop a robust AI model for turbidity in Lake Victoria, using the lake basin precipitation data and the sediment concentrations of the inflow rivers. To develop the AI model, the freely available remote sensing turbidity data for the lake is used as a reference data. Two models using a multi-layer perceptron neural network (MLPNN) and least square support vector regression (LSSVR) have been trained based on three different scenarios. Some performance indices such as mean absolute relative error and percent bias have been selected for model evaluation. According to the obtained results, LSSVR is more accurate than MLPNN in both training and testing phases of all scenarios. The results indicate that AI-based models are potential tools that can be adopted for WQ simulations of large lakes. Additionally, this study illustrates the potential of the use of remote sensing data to support model development, as an alternative to in-situ measurements, especially in data-scarce regions.

Keywords: Water quality, artificial intelligence, remote sensing, sediment concentration, turbidity