Maniam et al. (2026) Enhancing Agricultural Sustainability: An IoT-Based RNN-LSTM Model for Precision Sub-Surface Moisture Monitoring and Irrigation Optimisation

Identification

• Journal: Annals of Emerging Technologies in Computing
• Year: 2026
• Date: 2026-01-01
• Authors: Shamala Maniam, Erfan Memar, Tee Yei Kheng, Pragyan Kumari, HY Wong, Mukter Zaman
• DOI: 10.33166/aetic.2026.01.005

Research Groups

• Multimedia University, Malaysia
• Malaysian Cocoa Board, Malaysia

Short Summary

This study developed and evaluated an Internet of Things (IoT)-based real-time sub-surface soil moisture monitoring and irrigation optimization framework, integrating Time Domain Reflectometer (TDR) sensors with a Recurrent Neural Network (RNN) employing Long Short-Term Memory (LSTM) to predict soil moisture levels with high accuracy, thereby enhancing agricultural sustainability.

Objective

• To develop an automated Internet of Things (IoT)-based real-time soil moisture monitoring and irrigation framework integrated with a recurrent neural network (RNN) employing long short-term memory (LSTM) for moisture prediction, addressing key limitations in existing precision agriculture (PA) systems.
• To optimize irrigation schedules tailored for specific crops by providing improved comprehension of root behaviors, aiming to conserve water and reduce wastage of soluble fertilizers.

Study Configuration

• Spatial Scale: A real plantation site in a humid tropical environment. Five customized sub-surface soil moisture probes were deployed in a "Star Topology," each equipped with five TDR sensors placed at depths of 5 cm, 10 cm, 15 cm, 20 cm, and 25 cm from the surface. The system is capable of gauging moisture content up to a one-meter depth.
• Temporal Scale: A continuous six-month (180-day) dataset was collected for training and validation. The model was designed to predict soil moisture levels one day in advance.

Methodology and Data

• Models used: Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) units. The model consisted of two LSTM layers, each with 64 units, a dropout rate of 0.2, and a batch size of 32. The Adam optimizer was used, and the loss function was Mean Squared Error (MSE). A MinMax scaler was applied for data standardization.
• Data sources:
  • High-resolution sub-surface soil moisture measurements from customized TDR sensor probes deployed at a real plantation site.
  • Wireless Sensor Network (WSN) using ESP32-based low-range communication modules for transmitting sensor data to the cloud.
  • Temperature and humidity data obtained from the weather department.
  • Real-time rain data acquired via a real-time satellite API.
  • Soil specimens collected from the research site for sensor calibration.

Main Results

• The proposed IoT-based RNN-LSTM model demonstrated strong predictive capability for sub-surface soil moisture dynamics under real field conditions.
• Achieved a prediction accuracy of 95 ± 2%.
• Recorded a Mean Absolute Error (MAE) of 0.6362.
• Exhibited a Root Mean Square Error (RMSE) of 1.1544.
• Obtained an R² value of 0.3331.
• The system provides a scalable and data-driven solution for improving irrigation efficiency, leading to significantly reduced water usage and environmental impacts of farming practices.

Contributions

• Development of an integrated, field-ready framework that combines deep-root moisture prediction, secure IoT-based actuation, and holistic agronomic-economic performance metrics for sustainable, cost-effective, and data-driven irrigation.
• Design and implementation of a real-time subsurface soil moisture detection system capable of reaching depths of up to one meter, crucial for assessing moisture at root levels where most water uptake occurs, particularly for tree-living crops.
• Integration of IoT-enabled Time Domain Reflectometer (TDR) technology with an RNN-LSTM model for advanced predictive analytics, providing precise moisture tracking at various root depths and yielding water savings and economic benefits.
• Introduction of an end-to-end, solar-powered ESP32 mesh network that encrypts every uplink, estimates its own energy reserves, and autonomously triggers irrigation valve events based on predictive model outputs, enabling secure, energy-efficient, and closed-loop irrigation in real time.
• Compilation of a 180-day, five-depth soil moisture dataset in a humid tropical environment, addressing limitations of existing models often confined to shallow depths, single growth seasons, or temperate regions.
• Integration of ion-selective probes with the moisture network to compute holistic performance indicators, including yield-normalized water utilization, cost savings per hectare, and nutrient-leaching indices, shifting the focus from predictive accuracy alone to decision quality and economic viability.

Funding

• Graduate Research Scheme (Project ID: MMUI/180273), Multimedia University, Malaysia.

Citation

@article{Maniam2026Enhancing,
  author = {Maniam, Shamala and Memar, Erfan and Kheng, Tee Yei and Kumari, Pragyan and Wong, HY and Zaman, Mukter},
  title = {Enhancing Agricultural Sustainability: An IoT-Based RNN-LSTM Model for Precision Sub-Surface Moisture Monitoring and Irrigation Optimisation},
  journal = {Annals of Emerging Technologies in Computing},
  year = {2026},
  doi = {10.33166/aetic.2026.01.005},
  url = {https://doi.org/10.33166/aetic.2026.01.005}
}