Aflatoxins (AFs) are carcinogenic toxins produced by fungi in many staple grain crops. FDA legislative thresholds for the maximum allowable average AFs concentrations in grain intended for different uses determine if whole batches of grain are rejected or must be used for a less lucrative purpose. Few spatial studies of aflatoxin contamination in the field have been undertaken as AFs analysis is expensive, but such studies are key to improving food security. High risk areas need to be identified to allow management strategies to reduce the amount of wasted grain. AFs production tends to be greatest in high temperature and drought conditions. Previous research has suggested that local Moran's I (LMI) analysis of weather and remotely sensed imagery can identify locations at high risk of AFs contamination but this has not been validated with AFs data. This study uses a 1977-2004 AFs survey of counties in southern Georgia, USA with NDVI and thermal IR data from 1990-2004 imagery to validate this approach. Also investigated in this paper, as they are integral to the validation approach used, are issues relating to the identification of corn growing pixels in historical imagery and a modified kriging approach to interpolate the sparse AFs data for individual years. Comparison tests showed that significantly more high-risk points were identified by bivariate LMI analysis of thermal IR and NDVI in high-risk years than low and medium risk years and these pixels had significantly higher AFs concentrations than other pixels. AFs concentrations were not significantly higher for high-risk clusters in low and medium risk years. LMI analysis from all years suggested that imagery data was better at identifying drought risk than interpolated weather data alone as it also shows the effects of droughty soils. The approach developed here could be used with CropScape data, historic remotely sensed imagery and weather data to determine risk zones for other states in the south or other crops in Georgia. The approach developed for kriging sparse data could also be useful in a range of other environmental studies.
Read full abstract