Janani et al. (2025) Dynamic SG-SKRDX hybrid framework for precision weather forecasting and crop suitability in the Cauvery Delta

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-13
• Authors: K. Janani, R. Alageswaran, Rengarajan Amirtharajan
• DOI: 10.1038/s41598-025-31717-4

Research Groups

• School of Computing, SASTRA Deemed University, Thanjavur, India, Tamil Nadu
• School of Electrical & Electronics Engineering, SASTRA Deemed University, Thanjavur, India

Short Summary

This study introduces the Dynamic SG-SKRDX hybrid framework for precision weather forecasting and adaptive crop recommendation in the Cauvery Delta Region. The framework integrates an SVR-GRU (SG) model for predicting future weather conditions with a dynamic ensemble of machine learning models (SKRDX) for recommending suitable crops, demonstrating superior accuracy and robustness for sustainable agriculture.

Objective

• To develop and evaluate the Dynamic SG-SKRDX hybrid framework for precision weather forecasting and adaptive crop recommendation in the Cauvery Delta Region.
• To predict current and future weather conditions using an SVR-GRU (SG) model, enhanced through hyperparameter tuning and cross-validation.
• To recommend region-specific crops based on these forecasted weather parameters using a dynamic ensemble of machine learning models (SKRDX) that adapts to changing weather variables.

Study Configuration

• Spatial Scale: Cauvery Delta Region, Tamil Nadu, India (specifically Thanjavur district for weather data). The entire geographic land area of the Cauvery Delta Zone (CDZ) is 1.447 million hectares.
• Temporal Scale:
  • Weather Data: Daily meteorological data collected from 2012 to 2024 (12 years).
  • Forecasting Horizon: Futuristic weather parameters for a certain period of months (e.g., monthly forecasts for January 2025, July-August 2025).
  • Sliding Window: 30-day window for time-series data generation (optimal configuration).

Methodology and Data

• Models used:
  • Proposed Hybrid Framework: Dynamic SG-SKRDX
    • Weather Prediction (SG Model): Hybrid of Support Vector Regressor (SVR) and Gated Recurrent Unit (GRU) models, combined using a simple averaging technique.
    • Crop Recommendation (Dynamic SKRDX Model): Dynamic ensemble of Support Vector Machine (SVM), K Nearest Neighbour (KNN), Random Forest (RF), Decision Tree (DT), and eXtreme Gradient Boosting (XGBoost) models, using a Hard Voting Classifier for dynamic selection and weight updates.
  • Comparison Models (Weather): GRU, GRU+CrossValidation, GRU+SVR (individual components), LSTM+CrossValidation, RNN+Cross_Validation.
  • Comparison Models (Crop): Decision Tree, Random Forest, KNN, SVM, XGBoost.
  • Pre-processing: Min-Max scaling for weather data, StandardScalar for crop data, LabelEncoder for target crop types.
  • Optimization: Adam optimizer (GRU), GridSearchCV for hyperparameter tuning (SVR and base ML models).
  • Cross-validation: 8-fold cross-validation (weather models).
• Data sources:
  • Weather Data: Public Works Department (PWD) Thanjavur, for the delta regions (2012–2024). Features include datetime, temperature, dew point, humidity, precipitation, wind speed, wind direction, sea level pressure, solar radiation, and UV index.
  • Crop Recommendation Data: Kaggle benchmark dataset. Features include N (Nitrogen), P (Phosphorus), K (Potassium), temperature, humidity, pH, and rainfall. Comprises 22 distinct crop types.

Main Results

• SG Weather Model Performance (Test Dataset):
  • Mean Squared Error (MSE): 0.0069
  • Root Mean Squared Error (RMSE): 0.0835
  • Mean Absolute Error (MAE): 0.0481
  • R-squared (R²): 0.6408
  • The SG model demonstrated superior performance compared to GRU, GRU+CrossValidation, GRU+SVR, LSTM+CrossValidation, and RNN+Cross_Validation models.
• Dynamic SKRDX Crop Recommendation Model Performance:
  • Accuracy: 93.41%
  • Precision: 93.72%
  • Recall: 93.41%
  • F1-Score: 93.33%
  • The Dynamic SKRDX model outperformed individual Decision Tree, Random Forest, KNN, SVM, and XGBoost models.
• Hybrid Model Application:
  • Successfully forecasted weather for January 2025 (temperatures between 21.47 °C and 22.29 °C, humidity 70.03% to 72.17%, rainfall 0.36 mm to 0.55 mm) and recommended blackgram.
  • Successfully forecasted weather for July-August 2025 (temperatures between 20.36 °C and 21.09 °C, humidity 61.34% to 66.71%, rainfall 0.34 mm to 0.39 mm) and recommended maize.

Contributions

• Introduces a novel Dynamic SG-SKRDX hybrid framework that integrates deep learning for weather forecasting with dynamic ensemble machine learning for adaptive crop recommendation, specifically tailored for the Cauvery Delta Region.
• Develops an SG (SVR-GRU) weather model capable of capturing both long-term and short-term weather dependencies and non-linear relationships, addressing limitations of existing models in terms of forecasting horizon and linearity.
• Proposes a Dynamic SKRDX crop recommendation model that intelligently selects and updates the weights of base classifiers based on changing weather parameters, ensuring robust and accurate region-specific crop suggestions.
• Offers a practical, technology-driven solution for farmers to make informed decisions on crop cultivation based on futuristic weather forecasts, thereby promoting climate resilience, sustainable agriculture, and improved economic stability in the region.

Funding

Nil

Citation

@article{Janani2025Dynamic,
  author = {Janani, K. and Alageswaran, R. and Amirtharajan, Rengarajan},
  title = {Dynamic SG-SKRDX hybrid framework for precision weather forecasting and crop suitability in the Cauvery Delta},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-31717-4},
  url = {https://doi.org/10.1038/s41598-025-31717-4}
}