Hybrid Deep Learning Framework for Crop Yield Prediction and Weather Impact Analysis

Abstract

Agricultural productivity is seasonal and reliant on climatic variability and soil conditions. Accurate predictions of crop yield are necessary for food security, predicting resource allocation, and planning for climate-resilient farming. The study introduces a hybrid machine learning framework with Random Forest Regression (RFR) for feature selection and Long Short-Term Memory (RNN-LSTM) networks when weather conditions are variable to predict crop yield. Using simulated datasets, (weather, soil and crop management features) the paper proposes a new methodology formulated mathematically and with algorithms, equations, tables, and flow diagrams. The results indicate, the hybrid model performed better than the baseline regression and deep learning models as measured by RMSE, MAE and R². Sensitivity analysis indicated that rainfall and temperature were the most impactful weather factors for crop yield. The paper ends with implications for precision agriculture and future research work.

Introduction

• Agriculture is vital for global food security and the economy.

• Traditional statistical models fail to capture complex, nonlinear relationships among climate, soil, and crop variables.

• Machine Learning (ML) and Deep Learning (DL) enable better modeling of these high-dimensional, nonlinear interactions.

• Weather variability (rainfall, temperature) strongly impacts crop yield.

• This study proposes a hybrid prediction framework using:

  • Random Forest Regression (RFR) for feature importance

  • Long Short-Term Memory (LSTM) for sequence modeling

2. Related Work

• Prior research applied ML/DL methods like DNNs, CNN-RNNs, Bayesian Networks, and explainability tools (e.g., SHAP) in agriculture.

• Few studies explicitly integrate weather impacts into hybrid prediction models.

• This study fills that gap by merging weather-driven feature engineering with deep learning for yield forecasting.

3. Proposed Framework

Data Pipeline:

1. Input Data: Historical crop yield + meteorological data (rainfall, temperature, humidity, soil moisture)

2. Preprocessing:

   • Normalization, outlier removal

   • Feature selection using PCA and PSO

3. Model Architecture:

   • CNN layer captures spatial patterns in weather and soil data

   • BiLSTM layer captures temporal dependencies (seasonal effects)

   • Combines short-term and long-term climate effects on crop yield

Crop Yield Function:

Y=f(R,T,H,S,N,P,K)+εY = f(R, T, H, S, N, P, K) + \varepsilonY=f(R,T,H,S,N,P,K)+ε
Where:

• Y = crop yield

• R = rainfall, T = temperature, H = humidity, S = sunshine

• N, P, K = nutrient levels

Hybrid Prediction Algorithm:

1. Train RFR → Identify top-k important features

2. Use features in LSTM → Forecast yield over time

3. Hybrid output:
   Ypred=α⋅YRF+(1−α)⋅YLSTMY_{\text{pred}} = \alpha \cdot Y_{\text{RF}} + (1 - \alpha) \cdot Y_{\text{LSTM}}Ypred?=α⋅YRF?+(1−α)⋅YLSTM?
   (where α is a tunable weight, typically 0.3–0.5)

Ensemble Strategy:

• Combines CNN–BiLSTM outputs with SVM and XGBoost

• Uses weighted voting based on model validation accuracy

• Evaluated with RMSE, MAE, R² for optimization

4. Results and Discussion

Model Performance (Prediction Accuracy):

• Hybrid model outperformed all conventional approaches.

• Demonstrated ability to model both spatial and temporal patterns effectively.

Weather-Crop Yield Correlation:

• Rainfall and soil moisture have the strongest impact on yield.

• Consistent with global agricultural studies on water availability.

5. Key Takeaways

• The hybrid CNN–BiLSTM approach significantly improves prediction accuracy.

• Rainfall and soil moisture are the most critical weather parameters.

• The ensemble method increases robustness across regions and crop types.

• Can aid farmers and policymakers in better planning, resource management, and promoting sustainable agriculture.

Conclusion

The study provided a comprehensive framework for crop yield prediction by integrating historical agricultural data and meteorological parameters. The proposed hybrid CNN–BiLSTM model was implemented in conjunction with traditional algorithms of SVM and XGBoost using an ensemble framework, which performed successfully and produced better predictive accuracy than the individual models. Experimental evidences established that rainfall and soil moisture are identified strong factors affecting crop yields, followed by temperature and humidity as moderate factors. RMSE and MAE using accuracy metrics confirmed the proposed approach\'s predictive reliability and accuracy. The model not only offers accurate yield predictions across multiple crop seasons, but it also includes temporal dependencies and reflects non-linear associations in agricultural and weather data. By taking advantage of each of its deep learning and ensemble characteristics, the model also has the strength against sudden climatic factors applied, making it realistic for usage by farmers, policy-makers, and agricultural planners. Future work could be focused on including real-time IoT sensor data, satellite images, or climate change scenarios to improve the prediction accuracy of the yield and take sustainable agriculture to the next level in proactive decision-making.

References

[1] Liakos, K. G., Busato, P., Moshou, D., Pearson, S., & Bochtis, D. (2018). Machine Learning in Agriculture: A Review. Sensors, 18(8), 2674. https://doi.org/10.3390/s18082674 [2] Benos, L., Tagarakis, A. C., Dolias, G., Berruto, R., Kateris, D., & Bochtis, D. (2021). Machine Learning in Agriculture: A Comprehensive Updated Review. Sensors, 21(11), 3758. https://doi.org/10.3390/s21113758 [3] Abbas, F., Afzaal, H., Farooque, A. A., & Tang, S. (2020). Crop Yield Prediction through Proximal Sensing and Machine Learning Algorithms. Agronomy, 10(7), 1046. https://doi.org/10.3390/agronomy10071046 [4] Khaki, S., Wang, L., & Archontoulis, S. V. (2019). A CNN–RNN Framework for Crop Yield Prediction. Frontiers in Plant Science, 10, 1750. https://doi.org/10.3389/fpls.2019.01750 [5] Khaki, S., & Wang, L. (2019). Crop Yield Prediction Using Deep Neural Networks. Frontiers in Plant Science, 10, 621. https://doi.org/10.3389/fpls.2019.00621 [6] Nosratabadi, S., et al. (2020). Hybrid Machine Learning Models for Crop Yield Prediction. arXiv preprint. [arXiv:2005.04155] [7] Shawon, S. M., Barua Ema, F., Mahi, A. K., & Raihan, M. M. S. (2023). Crop Yield Prediction: Robust Machine Learning Approaches for Precision Agriculture. 2023 ICCIT. https://doi.org/10.1109/ICCIT60459.2023.10441634 [8] Abbas, R., et al. (2024). AI Can Empower Agriculture for Global Food Security: Challenges and Opportunities. Frontiers in Artificial Intelligence. https://doi.org/10.3389/frai.2024.1328530 Frontiers [9] Spring framework study: (2024). A Proposed Framework for Crop Yield Prediction Using Hybrid Featur Neural Computing and Applications. https://doi.org/10.1007/s00521-024-10226-x [10] Ma, Y., Zhang, Z., Kang, Y., & Özdo?an, M. (2021). Corn Yield Prediction and Uncertainty Analysis Based on Remotely Sensed Variables Using a Bayesian Neural Network Approach. Remote Sensing of Environment, 259, 112408. https://doi.org/10.1016/j.rse.2021.112408 [11] Lundberg, S. M., & Lee, S.-I. (2017). A Unified Approach to Interpreting Model Predictions. NeurIPS 2017. https://doi.org/10.48550/arXiv.1705.07874 [12] Chen, T., & Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System. Proceedings of KDD 2016, 785–794. https://doi.org/10.1145/2939672.2939785 [13] Gripsy, J. V., Sheela Selvakumari, N. A., Sahul Hameed, S., & Jamila Begam, M. (2024). Drowsiness Detection in Drivers: A Machine Learning Approach Using Hough Circle Classification Algorithm for Eye Retina Images. In Applied Data Science and Smart Systems (pp. 202–208). CRC Press. https://doi.org/10.1201/9781003471059-28 [14] Gripsy, J. V. (2020). Biological Software for Recognition of Specific Regions in Organisms. Bioscience Biotechnology Research Communications, 13(1), 340–344. https://doi.org/10.21786/bbrc/13.1/54 [15] Tan, G., et al. (2022). Winter Wheat Yield Prediction Using CNNs Incorporating Environmental and Phenological Data. arXiv preprint. [arXiv:2105.01282]

Copyright

Copyright © 2025 Dr. J. Viji Gripsy, Mrs. J. Mythili, S. Mohanapriya, G. Kaavya Varshini, R. Divya Shri. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.