Mandlik et al. (2026) An Oracle-Distilled Neural Network for Hardware-Free Precision Water Management across Indian Agroclimatic Zones

Identification

• Journal: International Journal for Research in Applied Science and Engineering Technology
• Year: 2026
• Date: 2026-09-10
• Authors: Shiv Mandlik, Aarush Khilosia
• DOI: 10.22214/ijraset.2026.84820

Research Groups

• Adani International School, Ahmedabad, Gujarat, India
• Shanti Asiatic School, Ahmedabad, Gujarat, India

Short Summary

This paper presents an artificial intelligence system that uses only openly available weather forecasts to optimize irrigation decisions in Indian agriculture. The system achieves 97–99% of the maximum attainable yield while reducing water consumption by 28–44% relative to sensor-based scheduling and by 59–67% relative to prevailing farmer practice.

Objective

• Develop an AI-driven irrigation system that matches or surpasses sensor-based performance without requiring capital-intensive hardware.
• Validate the system across four major crops (cotton, groundnut, maize, and soybean) at 15 locations spanning the major agroclimatic zones of India.
• Quantify water savings relative to existing baselines without compromising food security.

Study Configuration

• Spatial Scale: 15 locations across 12 Indian states, representing arid, semi-arid, and sub-humid climates.
• Temporal Scale: Long-term weather records (2010–2022) with a three-day public weather forecast used for deployment.

Methodology and Data

• Models used: FAO-56 single crop coefficient method, Stewart multiplicative yield model.
• Data sources: Open-Meteo Historical Weather API, ISRIC SoilGrids 2.0.

Main Results

• The behavioral cloning system achieves near-identical performance to the theoretically optimal oracle, with water savings ranging from 28% to 44% relative to sensor-based scheduling and by 59–67% relative to prevailing farmer practice.
• The system generalizes across crops, climates, and geographies while delivering substantial water savings.

Contributions

• This study demonstrates that irrigation policies distilled from optimal control demonstrations can generalize across crops, climates, and geographies while delivering substantial water savings.
• The methodology of using an oracle to create a deployable policy may prove applicable well beyond irrigation, to resource management problems more generally.

Funding

• This research was supported by the authors' home institutions (Adani International School and Shanti Asiatic School) and the Open-Meteo API.

Citation

@article{Mandlik2026OracleDistilled,
  author = {Mandlik, Shiv and Khilosia, Aarush},
  title = {An Oracle-Distilled Neural Network for Hardware-Free Precision Water Management across Indian Agroclimatic Zones},
  journal = {International Journal for Research in Applied Science and Engineering Technology},
  year = {2026},
  doi = {10.22214/ijraset.2026.84820},
  url = {https://doi.org/10.22214/ijraset.2026.84820}
}