Precision soil macro-nutrient prediction and fertilizer recommendation for cotton cultivation using IoT sensor fusion and explainable gradient boosting: a multi-site field deployment study from Vidarbha, Maharashtra, India

https://doi.org/10.55529/ijaap.52.93.107

Authors

  • Dr. Mayur R Bhoyar Assistant Professor, Jagdambha College of Engineering and Technology, Yavatmal, India.

Keywords:

IoT Precision Agriculture, Soil Nutrient Prediction, Gradient Boosting, Fertilizer Recommendation, Smallholder Agriculture.

Abstract

Aim: The aim was to identify macro-nutrient deficiencies and devise suitable nutrient management practices to boost yields in cotton-growing areas of the Vidarbha region of Maharashtra, India, where yields have been stagnant. Background: Soil macro-nutrient imbalances and inadequate fertilizer management are persistent problems in the Vidarbha region of Maharashtra, India, where cotton yields have been stagnant. The traditional soil testing methods are time-consuming, costly and too complex for smallholder farmers. The SoilSense-FR is an IoT precision soil monitoring system integrated with an explainable gradient boosting framework to predict soil N, P₂O₅, K₂O and fertilizer recommendation in real-time to optimize soil test crop response (STCR). The 12 sensor nodes were installed across 12 cotton farms in 4 Vidarbha districts (Yavatmal, Wardha, Amravati, Akola) over 3 Kharif seasons (2021-2023) and recorded 1440 labelled soil-sensor-yield observations, which were checked against standard laboratory analyses. Results: The soilSense-FR (ensemble of XGBoost and LightGBM models) performed better than 6 baseline ML models with R² of 0.89, 0.86, and 0.91 for N, P₂O₅, and K₂O prediction, respectively, and RMSE of 12.8, 3.4, and 10.2 kg/ha. The three biggest soil nutrient predictors are identified in the SHAP attribution: Organic Carbon, Soil pH and Electrical Conductivity. Field trials showed that 36.3% higher yield was achieved (1820 to 2480kg/ha) and there was a 12.3% reduction in the fertilizer costs, resulting in a net gain of ₹18,400/ha. The significance is that this is the first Multi-district IoT-ML deployment study conducted for the cotton soil health in the Vidarbha district cluster which is the biggest cotton growing district cluster in India. SoilSense-FR is a simple, accessible, low cost solution for precision nutrient management to farmers who are smallholders who do not depend on laboratory facilities.

Downloads

Download data is not yet available.

Published

2025-10-15

How to Cite

Dr. Mayur R Bhoyar. (2025). Precision soil macro-nutrient prediction and fertilizer recommendation for cotton cultivation using IoT sensor fusion and explainable gradient boosting: a multi-site field deployment study from Vidarbha, Maharashtra, India. International Journal of Agriculture and Animal Production, 5(2), 93–107. https://doi.org/10.55529/ijaap.52.93.107

Similar Articles

1 2 3 4 5 6 7 8 9 10 > >> 

You may also start an advanced similarity search for this article.