Assessing the Impact of Weather Variability on Crop Yield Predictions Using Machine Learning

Thesis Abstract

The increasing variability of weather patterns poses a significant challenge to accurate crop yield prediction models, impacting agricultural productivity and food security across temperate and tropical regions. This study aims to assess the extent to which weather variability influences crop yield predictions by employing advanced machine learning techniques, with a focus on identifying robust models that incorporate climate fluctuations. The specific objectives include (1) evaluating the predictive performance of various machine learning algorithms—such as Random Forest, Gradient Boosting Machines, and Support Vector Regression—in modeling crop yields; (2) analyzing the relationship between meteorological variables and crop productivity across different climatic zones; and (3) developing an integrated model that enhances prediction accuracy by accounting for weather variability. The research adopts a quantitative, empirical design based on secondary data analysis, using a stratified sampling approach to select representative agricultural regions within a country with diverse climate zones. The target population comprises crop production records from 2010 to 2020, sourced from national agricultural agencies and meteorological departments, totaling approximately 5,000 records encompassing variables such as temperature, rainfall, humidity, solar radiation, soil parameters, and crop yields. Data collection involves extracting longitudinal weather and yield data from established databases, ensuring that the datasets are comprehensive and representative of different weather patterns. Data preprocessing includes normalization, missing data imputation, and feature engineering to facilitate accurate modeling. The analysis applies a combination of descriptive statistics to profile weather and yield patterns, followed by inferential techniques including multiple regression analysis to establish baseline relationships. Model evaluation and comparison employ cross-validation, root mean squared error (RMSE), mean absolute error (MAE), and R-squared metrics to determine the most predictive algorithm. A theoretical framework grounded in the Ecological Niche Theory guides the interpretation of environmental influences on crop productivity, complemented by the Machine Learning Theory to justify the model selection and optimization processes. Key expected findings include significant correlations between temperature, rainfall, and crop yields, with variability in weather parameters accounting for a substantial portion of the unexplained variance in traditional models. Machine learning models, particularly ensemble techniques like Random Forest and Gradient Boosting, are anticipated to outperform conventional approaches, demonstrating superior predictive accuracy and robustness in the presence of weather fluctuations. The results will underscore the importance of incorporating climate variability explicitly into crop prediction models, thereby enhancing their reliability under changing climatic conditions. This study contributes novel insights into the integration of meteorological data within machine learning frameworks for agriculture, advancing both theoretical understanding and practical applications. By quantitatively demonstrating the impact of weather variability on crop yield forecasts and identifying the most effective modeling strategies, the research offers valuable guidance for stakeholders engaged in agricultural planning, climate adaptation, and food security policy formulation. Additionally, it broadens the empirical evidence base supporting the use of artificial intelligence techniques in climate-sensitive crop prediction. The main conclusion emphasizes that accounting for weather variability significantly improves the accuracy and resilience of crop yield models, with machine learning techniques providing flexible, data-driven solutions adaptable to diverse climatic contexts. Recommendations include the adoption of ensemble machine learning models in operational crop forecasting systems, integration of real-time weather data for dynamic prediction updates, and further exploration of spatial-temporal modeling approaches to capture localized climate impacts. Future research should investigate the incorporation of additional climate extremes, such as drought and flood events, and explore the potential of deep learning architectures for enhanced predictive performance under complex environmental conditions.