TOXIMAP: Computational Framework for Prediction of Geographical and Temporal Incidence of Mycotoxins in US Crop Fields

The major goal of this proposal is to develop a general predictive modeling framework for calculating mycotoxin occurrence in US crop fields before harvest. This mathematical framework will be accompanied by a web/mobile interface to explore temporal and geographical mycotoxin hazard maps for the entire United States.Objective #1. Build an inventory of favorable environmental conditions for aflatoxin production in corn using extensive field measurements. The PIs have already acquired a network of meteorological stations (soil moisture and temperature, atmospheric temperature, relative humidity, wind speed and direction) using a recent USDA/NRCS data collection grant. The sensors will be used to instrument a cornfield in partnership with a local farmer. In this project the PIs propose to leverage this existing infrastructure and supplement the data collection obtained from the sensor network with a plethora of remote sensing data at various resolutions and frequencies obtained using Google Earth Engine (imagery, geophysical, climate and weather). In addition, weekly corn samples will be collected to measure the levels of aflatoxin concentration during various developmental stages of corn. This data will be used to develop advanced aflatoxin predictive models with quantified uncertainties.Objective #2. Develop novel probabilistic data-driven models for general mycotoxin prediction with quantified uncertainties. A novel approach is proposed to obtain analytical expectations of mycotoxin concentrations as a function of environmental factors. These external variables (temperature, humidity etc.) are determined at a location of interest using spatial interpolations, which is subject to interpolation errors. The proposed approach accounts for both model uncertainties and interpolation errors in generating predictions for mycotoxin concentrations. Separate Gaussian processes are trained using data collected in Objective #1 for both physical models and forcing models, and then they are stacked to obtain prediction of mycotoxin production. Analytical expressions will be derived for first and second-order moments of the proposed stacked Gaussian process. While in this project only the aflatoxin will be modeled, the proposed nonparametric models can be generally applied across the mycotoxin spectrum and in other environmental science projects.Objective #3. Web/mobile interface to explore temporal and geographical aflatoxin hazard maps for the entire Continental United States (CONUS). Predictions generated in Objective #2 will be made broadly available to the community via a dedicated web/mobile interface that will be built and hosted as part of this project. The PIs propose to leverage the data collected in Objective #1 and computational framework derived in Objective #2 to provide both a realtime monitoring aflatoxin tool for CONUS and a forecasting tool to study the impact of climate change on aflatoxin production. The proposed software platform will be based on a plug-and-play framework, where new mycotoxin models, crops, geographical regions, and environmental data can be easily integrated.