Enhancing Turbidity Prediction in Saki Lake Using a Hybrid Gradient Boosting and Neural Network Model

Lakes are vital ecosystems, providing fresh water for drinking, agriculture, and biodiversity. However, industrialization and human activities have led to significant water pollution, threatening these ecosystems. Monitoring water quality, particularly turbidity (TUB), is essential for ecological balance and sustainable resources. Remote Sensing (RS) technology offers real-time, large-scale data, enhancing environmental monitoring efficiency. Incorporating RS data into machine learning (ML) models improves the forecasting of water quality trends. However, the weak, complex, and nonlinear correlation between in-situ data and satellite-derived information presents a significant challenge. To address this, the study presents an innovative hybrid model combining Gradient Boosting Regressor (GBR) and Neural Network (NN) for predicting TUB in water, focusing on Saki Lake in Telangana, India. This hybrid model is particularly well-suited for managing small datasets and effectively capturing both linear and non-linear patterns in data. The methodology includes data preparation, a three-layer NN architecture with ReLU activation and batch normalization, hyperparameter tuning using a Random Search tuner, and extensive training of both models. Final predictions are derived by merging the outputs from both models and assessed using Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Mean Absolute Error (MAE). Results are visualized and exported for reproducibility, aiming to enhance TUB prediction accuracy and provide a robust framework for assessing water quality in polluted lakes.