A Hybrid RF–ANN–CNN modelling approach for suspended sediment estimation in Mogan Lake using Landsat-8 imagery

Continuous and reliable monitoring of water quality is crucial for the sustainable management of lake ecosystems. Suspended Solids (SS) concentration is a key indicator, yet traditional measurement methods are costly and offer limited spatial and temporal coverage. Remote sensing addresses these constraints by providing wide area, repeatable observations. This study estimated SS concentrations in Lake Mogan using a hybrid remote sensing approach with Landsat-8 OLI imagery. First, scarce in-situ data were augmented with a Random Forest (RF) model to create a more robust training set. This dataset then supported Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) models, using B2, B3, and B4 bands along with spectral indices. The CNN model yielded the highest accuracy (R² = 0.97, RMSE = 0.17, MAE = 0.13) and was used to generate lake wide SS maps. Overall, the RF–ANN–CNN framework significantly improves SS estimation in small lakes with limited field data, demonstrating the strong potential of remote-sensing-based deep learning for sustain-able water-quality monitoring.