An IoT-Enabled Digital Hydrology Framework for River Water Quality Index Prediction Using Hybrid Deep Learning
TL;DR
A digital hydrology model driven by deep learning capabilities, incorporating IoT-based water quality monitoring with advanced artificial neural networks to estimate the water quality index is proposed, finding that the performance of CNN–LSTM significantly outperforms standard LSTM with lower prediction errors and higher explanatory power.
Abstract
— The paper proposes a digital hydrology model driven by deep learning capabilities, incorporating IoT-based water quality monitoring with advanced artificial neural networks to estimate the water quality index. For analysis, the model focuses on high-resolution time series data containing approximately 46500 samples from IoT sensors in Prayagraj, Uttar Pradesh State, India, from January 2019 to March 2020. The key physicochemical parameters of water were used as inputs for the model. The LSTM and hybrid CNN–LSTM architectures were considered for the modeling task to capture long-term temporal dependencies alongside short-term local patterns in water quality dynamics. In this study, we found that the performance of CNN–LSTM significantly outperforms standard LSTM with lower prediction errors (RMSE = 1.431, MAE = 1.168) and higher explanatory power (R² = 0.925). These results demonstrate that hybrid deep learning models are highly effective at modeling complex nonlinear river water quality processes influenced by seasonal and anthropogenic factors.