Impact of Weather Extremes on Agricultural Yield: A Spatiotemporal Statistical Analysis Using Bayesian Hierarchical Models

Project Abstract

Weather extremes driven by climate variability pose substantial risks to agricultural productivity, necessitating robust quantitative frameworks that capture spatiotemporal heterogeneity and complex dependencies among climatic drivers, soil properties, crop traits, and management practices. This study develops a Bayesian hierarchical model to quantify the impact of extreme temperature and precipitation events on crop yield across multiple spatial scales and time periods, integrating high-resolution climate reanalysis, soil and land-use data, and farm-level management information. We address non-stationarity in climate-yield relationships by allowing region-specific random effects and by incorporating priors that reflect agronomic knowledge and expert elicitation. The core modeling framework combines a spatiotemporal Gaussian process to capture latent spatial structure, a hierarchical crop-yield submodel that links yield to standardized extreme indices (e.g., heatwaves, drought spells) and ancillary covariates, and a measurement-error layer to account for harvest-report inaccuracies. Model selection and validation employ cross-validated predictive checks, information criteria, and posterior predictive diagnostics to ensure robust inference under data sparsity and missingness. We implement a multivariate approach to simultaneously model multiple crops and phenological stages, enabling assessment of differential sensitivity to extremes and potential yield compensation mechanisms. The analysis addresses key questions (i) which extremes most strongly influence yield across agro-ecological zones, (ii) how the strength and direction of extreme-yield relationships vary temporally (e.g., across growing seasons and decades), and (iii) the extent to which socio-economic and adaptive management factors mitigate adverse effects. Our results reveal pronounced yield declines linked to consecutive hot days during flowering and to intense precipitation shocks during grain-filling, with spatially varying magnitudes due to soil moisture retention and irrigation access. Bayesian inference provides full posterior distributions for effect sizes, enabling probabilistic risk assessment and scenario evaluation under future climate projections. We quantify the attributable fraction of yield loss due to specific extreme events and map risk profiles to guide targeted adaptation strategies, such as cultivar selection, planting windows, and irrigation scheduling. Sensitivity analyses examine the robustness of conclusions to alternative extreme definitions, prior specifications, and spatial resolution. The study contributes methodologically by advancing a scalable Bayesian hierarchical framework capable of disentangling intertwined climatic, agronomic, and economic drivers of yield, and empirically by delivering regionally resolved insights into the vulnerabilities and resilience of agricultural systems in the face of weather extremes. Policy relevance is highlighted through risk communication tools, including posterior predictive risk maps and decision-support metrics that inform stakeholders about probable yield trajectories and mitigation priorities under projected climate scenarios.

Project Overview

What This Project Is About

---

A plain-language overview of how weather patterns like heat waves, heavy rainfall, and drought affect crop yields over time and across different places. The project examines how to model these effects using simple statistics that can handle data that changes over both space and time.

---

---

The Problem It Addresses

---

Crop yields are influenced by extreme weather, but traditional methods may miss how these effects vary by location and year. This project fills that gap by using a method that can account for differences across farms or regions and changes over seasons and years, helping farmers and policymakers predict risks more accurately.

---

---

Objectives of the Project

---

1. Understand how extreme weather relates to crop yields in different places and times.
2. Learn the basics of spatiotemporal analysis and Bayesian ideas at a high level.
3. Create a simple model that can adapt to multiple years and locations.
4. Test the model on real data to see how well it explains yield variations.
5. Provide user-friendly results that can inform decision-making for farmers and planners.

---

---

What You Will Do Step by Step

---

1) Gather publicly available yield data and local weather records; 2) Clean and prepare data for analysis; 3) Learn a basic, accessible modeling approach that links yields to weather extremes; 4) Fit the model using a simple software tool; 5) Check how well the model matches observed data; 6) Interpret results in plain language; 7) Create visuals and a brief report; 8) Discuss limitations and potential improvements.

---

---

Expected Outcome

A readable explanation of how extreme weather affects yields, plus a ready-to-use model framework that can be expanded. Outputs include simple graphs and a short guide for applying the approach to local data, helping stakeholders plan with better risk estimates.