The potential for aflatoxin predictive risk modelling in sub-Saharan Africa: a review

Abstract

This review presents the current state of aflatoxin risk prediction models and their potential for value actors throughout the food chain in sub-Saharan Africa, with a specific focus on improving smallholder farmer management practices. Several empirical and mechanistic models have been developed either in academic research or by private sector aggregators and processors in high-income countries including Australia, the USA, and Southern Europe, but these models have been only minimally applied in sub-Saharan Africa, where there is significant potential and increasing need due to climate variability. Predictions can be made based on historic occurrence data using either a mechanistic microbiological framework for aflatoxin accumulation or an empirical model based on statistical correlations with climate conditions and local agronomic factors. Model results can then be distributed to smallholders through private, public, or mobile extension services, used by policymakers for strategy or policy, or utilised by private sector institutions for management decisions. Specific agricultural advice can be given during the three most critical points in the phenological cycle: preseason insight including sowing timing and crop varieties, preharvest advice about management and harvest timing, and postharvest optimal practices including storage, drying, and market information. Model development for sub-Saharan Africa is limited by a dearth of georeferenced aflatoxin occurrence data and real-time high resolution climate data; the wide diversity of farm typologies each with significant information and technology gaps; a prevalence of informal market structures and lack of economic incentives systems; and general lack of awareness around aflatoxins and best management practices to mitigate risk. Given advancements towards solving these challenges, predictive aflatoxin models can be integrated into decision support platforms to focus on optimisation of value for smallholders by minimising yield and nutritional losses, which can propagate value throughout the production and postharvest phases.

Verification of the Accuscan gold reader and RIDA smart phone application rapid test kits in detection and quantification of aflatoxin levels in maize from selected regions in Kenya

“…Likewise, in Temba et al (2021), low rainfall combined with high temperatures increase the chances of aflatoxin contamination in maize which proves the case in the study sites. Equally, in Keller et al (2022), the evolving climatic conditions in sub -Saharan Africa, Kenya included, are conducive for aflatoxin production. As stated by Kangethe et al (2017), Nandi and Makueni are hot spot regions for aflatoxin contamination in maize and fall in the same agro ecological zones of lower humid highlands to upper midlands, and upper high land zones (LH2, LH3, LM4, LH5), (UM3 UM4).…”

Making food systems more resilient to food safety risks by including artificial intelligence, big data, and internet of things into food safety early warning and emerging risk identification tools

“…Such data would be needed to recalibrate the empirical part of the model. The Partnership for Aflatoxin Control in Africa is currently collecting data and establishing an information management system, which could help provide relevant warning (Keller et al., 2021).…”

“…Several models are amenable for use in Africa. Keller et al (2021) listed 15 methods for mycotoxin prediction that would be suitable for application in sub-Saharan Africa, that is, AFLA-maize, AFLA-maize + carryover, AFLApistachio, APHLIS+, APSIM+ Risk Model, AVHRR-based, CROPGRO, drought index (ARID), Maxent2, multilevel modeling, multivariate regression, risk in storage, spatial Poisson profile regression, and stacked Gaussian. Their accuracies varied between 54% and 99% for 10 methods (with the remaining ones being not predictive, future projections, or unvalidated).…”

“…Such data would be needed to recalibrate the empirical part of the model. The Partnership for Aflatoxin Control in Africa is currently collecting data and establishing an information management system, which could help provide relevant warning (Keller et al, 2021). Chen et al (2022) applied a new predictive modeling method for microbial growth to wheat infection with Helminthosporium and Alternaria Fungi.…”

Dynamic geospatial modeling of mycotoxin contamination of corn in Illinois: unveiling critical factors and predictive insights with machine learning

“…Satelliteacquired databases, including the normalized difference vegetation index (NDVI), have been previously used in studies evaluating mycotoxins in European wheat and led to higher mycotoxin model accuracy (Wang et al, 2022). Several studies have shown significant correlations among climate, soil properties, NDVI, and agricultural management practices with AFL contamination in Africa, as reviewed by Keller et al (2022). AFL studies conducted in Eastern and Western Kenya (Mutiga et al, 2014) support the utility of remote-sensing data such as NDVI, rainfall, and soil properties (such as organic carbon content, pH, total exchangeable bases, salinity, texture, and soil type) to accurately predict AFL contamination (Smith et al, 2016).…”