Aziz et al. (2026) Deep Learning Based In-Silico Water Level Prediction and IoT Based Monitoring System

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-01-01
• Authors: Faruque Aziz, Rudraneel Bhattacharya, Arijit De, Sukanta Ghosh, Debashish Pal, Subhajit Das
• DOI: 10.1007/s11269-025-04463-5

Research Groups

• BITS Pilani K K Birla Goa Campus, Zuarinagar, India
• Aecho.ai, Bengaluru, India
• Electronics and Communication Engineering Department, KPR Institute of Engineering and Technology, Coimbatore, India
• School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, India
• University of Engineering and Management, Kolkata, India

Short Summary

This study develops and evaluates machine learning models for water level prediction using Venice Lagoon data and proposes integrated IoT-based systems for real-time water leakage and quality monitoring to enhance water resource management. The Long Short-Term Memory (LSTM) model demonstrated superior predictive performance compared to Random Forest (RF), achieving a Root Mean Square Error (RMSE) of 0.012.

Objective

• To develop and evaluate machine learning models (LSTM and Random Forest) for accurate in-silico water level prediction using historical time series data.
• To propose and implement an integrated IoT-based hardware system for real-time water leakage monitoring.
• To propose a comprehensive IoT-based framework for multi-parameter water quality monitoring, aiming for efficient and sustainable water resource management.

Study Configuration

• Spatial Scale: Venice Lagoon Telemareographic Network, Punta Salute, Giudecca canal (45º25’N, 12º20’E).
• Temporal Scale: Water level time series data from 1983 to 2015 (33 years), recorded hourly. Training data covered 1983–2010, and testing data covered 2011–2015.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) neural network, Random Forest (RF) ensemble learning algorithm.
• Data sources:
  • Water level time-series data obtained from the Kaggle dataset (originally from the automatic station of the Venice Lagoon telemareographic network, Punta Salute).
  • IoT sensors for leakage monitoring: Ultrasonic sensor (HCSR04) for water level, Temperature and Humidity sensor (DHT11).
  • Proposed water quality monitoring system includes sensors for pH, turbidity, dissolved oxygen, temperature, pressure, flow rate, mineral content, contamination levels, ambient temperature, humidity, pump status, and valve positions.

Main Results

• Statistical analysis of Venice Lagoon water level data (1983-2015) revealed distinct seasonal characteristics (higher mean from September to December, lower from March to May), inter-annual variability, and an increasing trend of approximately +0.04 mm/year, indicating climate-warming-driven sea-level rise.
• For water level prediction, the LSTM model achieved a Root Mean Square Error (RMSE) of 0.012 and a correlation coefficient (R²) of 0.73.
• The Random Forest (RF) model achieved an RMSE of 0.086 and an R² of 0.69.
• The LSTM model demonstrated superior performance over RF, particularly in capturing extreme events and non-linear temporal variations, making it more suitable for long-term forecasting and real-time prediction.
• A low-cost IoT-based hardware system for real-time water leakage monitoring was successfully developed and experimentally validated, utilizing an Arduino UNO microcontroller, an ultrasonic sensor, and a temperature/humidity sensor.
• A three-tier IoT architecture for comprehensive water quality monitoring (including pH, turbidity, dissolved oxygen, temperature, etc.) was proposed, designed for real-time data acquisition, edge processing, and cloud-based analytics.

Contributions

• Integration of statistical analysis, advanced machine learning (LSTM and RF) for water level prediction, and IoT-based monitoring systems (leakage and quality) into a unified framework for efficient water resource management.
• Demonstrated the superior performance of LSTM over Random Forest for water level prediction, especially in handling non-linear and temporal variations and capturing extreme events, which is crucial for flood monitoring.
• Developed a practical, low-cost, and scalable IoT-based hardware architecture for real-time water leakage detection, addressing a significant challenge in water conservation.
• Proposed a comprehensive three-tier IoT framework for multi-parameter water quality monitoring, enabling real-time data collection, edge processing, and cloud-based predictive analytics.
• Addresses a gap in existing literature by integrating monitoring and prediction of water levels with a proposed water leakage architecture framework.

Funding

Not applicable. The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Citation

@article{Aziz2026Deep,
  author = {Aziz, Faruque and Bhattacharya, Rudraneel and De, Arijit and Ghosh, Sukanta and Pal, Debashish and Das, Subhajit},
  title = {Deep Learning Based In-Silico Water Level Prediction and IoT Based Monitoring System},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-025-04463-5},
  url = {https://doi.org/10.1007/s11269-025-04463-5}
}