Aflatoxin contamination in crops, particularly aflatoxin B1 (AFB1) from *Aspergillus flavus*, poses a substantial threat to human and animal health worldwide. Up to 80% of crops globally show detectable aflatoxin levels, with a significant portion exceeding legal limits. AFB1 is a potent carcinogen, causing liver damage and cancer through ingestion of contaminated products or animal products derived from contaminated feed. Maize (*Zea mays* L.), a globally significant crop, is particularly vulnerable to aflatoxin contamination during both pre- and post-harvest stages. This study focuses on developing and testing a model to predict and manage aflatoxin risk in the maize supply chain, motivated by high rejection rates of maize shipments due to excessive aflatoxin levels at a large processing plant in Hyderabad, India. The high rejection rates (20-55% annually) highlight the urgent need for effective risk prediction and management tools. While existing models predominantly focus on pre-harvest dynamics, this research integrates pre- and post-harvest phases, considering factors such as weather, location, storage conditions, and common processing practices (drying, filtering, bagging) to provide a more comprehensive risk assessment across the entire supply chain. The dual cropping seasons (Kharif and Rabi) and multiple sourcing regions in India further complicate the risk assessment, emphasizing the need for a robust integrated model that accounts for these complexities. The model's primary aim is to predict aflatoxin levels at the factory gate to optimize sourcing strategies and to provide a adaptable framework for use in diverse meteorological environments and across different countries. Due to the scarcity of detailed spatially and temporally resolved data, especially in low- and middle-income countries, the model is parameterized and validated using reliable aflatoxin contamination data from shipments at the processing plant.

Most quantitative risk assessments for aflatoxin have focused on pre-harvest dynamics, modeling the influence of weather and site variables on *A. flavus* growth and aflatoxin production. Several models, such as the AFLA model developed by Battilani et al., have been used to predict aflatoxin contamination in maize and other crops. These models often incorporate aspects of crop growth dynamics and environmental factors to estimate contamination risk. However, these models generally do not consider the post-harvest phase, which plays a critical role in aflatoxin accumulation, particularly in regions with limited storage infrastructure. This study builds upon the existing literature by integrating both pre- and post-harvest dynamics into a single, comprehensive model, thereby addressing a significant gap in current risk assessment approaches.

The study utilizes a discrete-time compartmental model to track *A. flavus* and aflatoxin levels throughout the maize supply chain, from planting to delivery. Meteorological data (temperature, humidity, rainfall) with 3-hourly temporal and 10 km spatial resolution from the UK Met Office Unified Model were used. Aflatoxin concentration data for daily maize shipments received at a Hyderabad processing plant from three sourcing regions (Bellary, Guntur, and Nizamabad) over six years (2012-2017) served for model parameterization and validation. The model simulates maize growth and *A. flavus* dynamics on 1000 simulated farms per region per season, incorporating dual cropping seasons (Kharif and Rabi). The pre-harvest model, adapting from Battilani et al., uses a compartmental framework to track *A. flavus* spores in soil and on silks, infections on maize, and aflatoxin production. The hourly resolution meteorological data drives growth rates and susceptibility. A spatial grid covering the sourcing regions was created. The post-harvest model tracks *A. flavus* and aflatoxin accumulation after harvest, considering on-farm and market storage conditions, and simulates the factory's daily sourcing and sampling process to match the model output with observed data. The harvest processing model incorporates drying, filtering, and bagging, allowing for scenario analysis of various control practices. Five key parameters (sporulation rate, pre- and post-harvest *A. flavus* growth rates, aflatoxin production rate, and drying protection duration) were estimated using Approximate Bayesian Computation (ABC) based on the aflatoxin time-series data (2012-2015), with 2016-2017 data used for validation. Model performance was assessed by comparing model-predicted median aflatoxin levels and monthly rejection rates with observed data. The model's ability to capture the spatial and temporal variability in *A. flavus* growth suitability and aflatoxin production was also evaluated using risk maps. Scenario analyses explored the impact of improved filtration, controlled storage, and optimized sourcing strategies on aflatoxin levels.

The integrated model accurately replicated the overall profile, scale, and variance of the historical aflatoxin data for both the fitting (2012-2015) and validation (2016-2017) periods. The model successfully predicted the magnitude and timing of annual peaks and monthly fluctuations in aflatoxin levels. The model's accuracy in predicting median aflatoxin levels was approximately 85% (within ±4 ppb of observed values), consistently across both periods. Model-predicted monthly rejection rates also followed the observed trends, though with a tendency to overestimate rejection rates during low-risk periods, resulting in an accuracy of around 50% (within ±10% of observed values). Risk maps highlighted significant spatial and temporal variability in *A. flavus* growth suitability and aflatoxin production across the region. Analysis of individual batch trajectories revealed that *A. flavus* colonization increases rapidly during the growing season, with substantial growth continuing during storage. Most aflatoxin production occurred during the storage phase, significantly increasing with prolonged storage times. Scenario analyses indicated that improved filtration at harvest substantially reduced aflatoxin levels at the factory gate, and controlled storage (cooling) showed significant benefits, with non-linear effects of temperature reduction. Optimized sourcing strategies, based on historical data and model predictions, substantially reduced aflatoxin levels compared to the historical sourcing profile. The model also enabled ‘nowcasting’ to aid real-time decision-making by providing insights into current and near-future aflatoxin levels based on available and historical data. This information is crucial for timely decisions about sourcing strategies to mitigate the risk of exceeding aflatoxin thresholds and incurring rejection of shipments.

This integrated model provides a valuable tool for assessing and managing aflatoxin risk in maize supply chains, offering insights into both pre- and post-harvest dynamics. The model's ability to accurately replicate historical data across both fitting and validation periods demonstrates its predictive power. The high accuracy in predicting median aflatoxin levels and the conservative approach in predicting rejection rates (lower false negatives) emphasize the model's usefulness for decision-making. The scenario analyses highlight the potential benefits of various interventions, such as improved filtration and controlled storage, for reducing aflatoxin contamination. The model's adaptability to different geographical locations and cultural practices makes it a potentially valuable tool for assessing and mitigating aflatoxin risks globally. While the model's performance is encouraging, some limitations exist, including data availability, particularly for intermediate stages of the supply chain, and the assumption of equilibrium between environmental humidity and kernel surface moisture during storage. These limitations suggest directions for future research that would improve model accuracy. Furthermore, integrating economic considerations into the scenario analyses would enhance the practical application of the model.

This study presents a novel integrated model for predicting aflatoxin contamination in maize, successfully capturing pre- and post-harvest dynamics. The model's high accuracy and ability to incorporate intervention strategies provide a valuable tool for decision-making in the maize industry. Future work should focus on improving data availability, refining model parameters, and integrating economic factors for a comprehensive cost-benefit analysis of various intervention strategies.

The primary limitation of the study is the reliance on aflatoxin data from the factory gate for model parameterization and validation. The lack of data from earlier stages in the supply chain introduces uncertainty. The model's assumptions, such as equilibrium between environmental humidity and kernel surface moisture during storage, and the use of a 10 km resolution meteorological data due to data availability, could be refined with additional data. The assumption of a uniform distribution of sowing and harvesting dates represents another limitation. The focus on a specific region in India also limits the generalizability of the findings to other regions with different climatic conditions and agricultural practices.