Priyanka et al. (2025) Machine learning approach for crop planning and resource allocation in the Bargarh Canal Command

Identification

• Journal: Plant Science Today
• Year: 2025
• Date: 2025-12-15
• Authors: M Priyanka, C P Jagadish, M D Dwarika, S. K. Raul, P S Ambika, S Subhasish, M Bimalendu, R Sefali, Sanghamitra Samantaray
• DOI: 10.14719/pst.11084

Research Groups

• Department of Soil and Water Conservation Engineering, College of Agricultural Engineering and Technology, Odisha University of Agriculture and Technology, Bhubaneswar, Odisha, India
• Krishi Vigyan Kendra, Jagatsinghpur, Odisha University of Agriculture and Technology, Bhubaneswar, Odisha, India
• Department of Soil Science and Agricultural Chemistry, College of Agriculture, Odisha University of Agriculture and Technology, Bhubaneswar, Odisha, India
• Krishi Vigyan Kendra, Dhenkanal, Odisha University of Agriculture and Technology, Bhubaneswar, Odisha, India

Short Summary

This study developed an advanced machine learning framework, integrating predictive modeling, clustering, and genetic algorithms, to optimize crop planning and resource allocation in the Bargarh Canal Command, Eastern India. The framework, with XGBoost demonstrating superior performance, provides data-driven insights for enhancing crop yield and net returns under climate variability and resource constraints.

Objective

• To develop and implement an advanced machine learning framework for optimizing agricultural decision-making by integrating predictive modeling, clustering techniques, and genetic algorithms in the Bargarh Canal Command of Eastern India.
• To enhance crop yield and net return predictions while considering environmental factors, resource constraints, and market dynamics.

Study Configuration

• Spatial Scale: Bargarh Canal Command (BCC) in the western part of Odisha, India, encompassing 11 blocks across Balangir, Subarnapur, Sambalpur, and Bargarh districts. Geographically located between 20°43' N and 21°41' N latitude and 83°39' E and 83°58' E longitude, with an elevation of 185 meters above mean sea level. The Culturable Command Area (CCA) is 115.36 thousand hectares.
• Temporal Scale: Historical weather and crop yield data from 2010 to 2023. Input cost data from 2015 to 2023. Model retraining weekly with new weather data, seasonal calibration with yield results, and annual feature set expansion.

Methodology and Data

• Models used: Random Forest (RF), XGBoost, Long Short-Term Memory (LSTM) for predictive modeling; k-means for clustering analysis; Genetic Algorithms for optimization; CROPWAT software for crop water requirement calculation.
• Data sources:
  • Weather Data: India Meteorological Department, NASA POWER (2010-2023).
  • Crop Yield Records: Agricultural Research Institute, College of Agriculture, Odisha University of Agriculture and Technology, Bhubaneswar (2010-2023).
  • Soil Quality Data: National Bureau of Soil Survey and Land Use Planning (NBSS and LUP) (2018 soil survey data).
  • Economic Data: Agricultural Market Information System, Agriculture Statistics Book (market price data 2010-2023), Farm Economic Survey (input cost data 2015-2023).
  • Real-time/Operational Data: Automated weather stations, soil sensors, satellite imagery, farm management software.
  • Input Features: Environmental factors (daily temperature metrics, rainfall characteristics, humidity, solar radiation, wind speed), Agronomic factors (crop type/variety, growth stage, soil characteristics, fertilizer application, irrigation scheduling, pest/disease incidence, previous crop), Management factors (planting density, planting/harvesting dates, labor input, machinery usage).

Main Results

• The XGBoost model demonstrated superior performance for yield prediction, achieving an average R² of 0.87 and a Root Mean Square Error (RMSE) of 0.32 tons per hectare across all crops.
• For net return prediction, XGBoost achieved an R² of 0.83 with an RMSE of 0.52 tons per hectare.
• Feature importance analysis revealed rainfall during critical growth periods as the most crucial factor (23.4% predictive power), followed by soil nitrogen content (18.7%).
• Crop-specific prediction accuracy was highest for paddy (R² = 0.92), moderate for pulses and oilseeds (R² = 0.75-0.85), and lower for pest-prone crops (R² = 0.68-0.72).
• K-means clustering identified four distinct crop clusters based on water requirements and profitability: High-Profit, High-Water (Paddy, Sugarcane); High-Profit, Moderate-Water (Vegetables); Moderate-Profit, Low-Water (Wheat, Maize, Sunflower, Til); and Low-Profit, Low-Water (Mung, Biri, other pulses, Groundnut).
• Weather scenario analysis showed that below-average rainfall led to a 15% reduction in High-Profit, High-Water crops in favor of Moderate-Profit, Low-Water crops, resulting in a 12% decrease in total net returns. Conversely, above-average rainfall allowed an 8% increase in High-Profit, High-Water crops, leading to a 5% increase in total net returns.
• The framework enables decision support applications including optimized crop calendar, risk assessment, and dynamic resource allocation dashboards.

Contributions

• Development of a comprehensive and integrated machine learning framework combining predictive modeling (RF, XGBoost, LSTM), clustering analysis (k-means), and genetic algorithms for agricultural decision-making in a specific canal command area.
• Quantitative comparison of state-of-the-art ML models, identifying XGBoost as superior for crop yield and net return prediction in the study region.
• Identification of key factors influencing crop yield through feature importance analysis, providing actionable insights for precision agriculture.
• Classification of crops into distinct clusters based on water requirements and profitability, offering a strategic basis for crop diversification and resource allocation.
• Demonstration of the framework's adaptability to varying weather scenarios and its utility in generating weather-responsive planning strategies and optimized crop calendars.
• Providing a scalable decision support system for farmers, policymakers, and researchers to enhance sustainable agricultural intensification and economic outcomes.

Funding

• Not explicitly stated as projects or programs. The authors acknowledge hardware and software support from the Hydrology Simulation Laboratory of the College of Agricultural Engineering and Technology at the Odisha University of Agriculture and Technology, Bhubaneswar.

Citation

@article{Priyanka2025Machine,
  author = {Priyanka, M and Jagadish, C P and Dwarika, M D and Raul, S. K. and Ambika, P S and Subhasish, S and Bimalendu, M and Sefali, R and Samantaray, Sanghamitra},
  title = {Machine learning approach for crop planning and resource allocation in the Bargarh Canal Command},
  journal = {Plant Science Today},
  year = {2025},
  doi = {10.14719/pst.11084},
  url = {https://doi.org/10.14719/pst.11084}
}