Researches On Risk Prediction Model Of Aflatoxin Contamination In China

Peanut?Arachis hypogaea L.?is not only the main oil crops and agricultural products with export superiority in China,but also a kind of food with high nutritional value and therapeutic function.It is an economic crop can be used for both direct consumption and oil extraction.It occupies an important position in peanut production in China and the world.However,peanuts are very vulnerable to fungal infection and aflatoxin contamination,which poses a threat to human health and has become a major constraint factor affecting peanut consumption safety and world trade in China.Aflatoxin contamination can occur in the whole process such as planting,harvesting,storage,transportation and marketing.The contamination happened in the field is not only the main cause of aflatoxin contamination of peanut,but also the source of contamination during storage.Aflatoxin contamination control is a worldwide problem.There are many kinds of peanut varieties,soil types and weather patterns in China.Therefore,it is of great significance to understand the variation trend of aflatoxin contamination in peanut and analyze the effects of variety resistance,soil types,aflatoxigenic potential of fungi and climatic conditions on aflatoxin contamination.Thus,it is significant for scientific control of aflatoxin contamination in peanuts to study on risk prediction model of aflatoxin contamination,develop the early warning technology,and then adopt precise and efficient methods to make post-processing as pre-prevention.Based on the data of aflatoxin content in peanut and distribution of aflatoxigenic fungi of main peanut production areas,as well as peanut variety information,soil information and climate data released by authoritative agencies during 10 consecutive years from 2009 to 2018,the variation trend of aflatoxin contamination and its relationship with influence factors were analyzed.A rule-based pre-classification combination with balanced sampling and random forest model suitable for China was established,which can be used for prediction of aflatoxin content in peanuts.It provided crucial technical support for risk management of aflatoxin contamination in peanut.The main research methods and results are summarized as follows:1.The spatial distribution characteristics of AFB₁ in peanut were evaluated and AFB₁contamination map was generated.The main results were as follows:?1?The distribution pattern of aflatoxin contamination in peanuts from China was carried out by geostatistical analysis.It showed that there were significant temporal and spatial differences.?2?The omnidirectional semivariogram was obtained by using GS+10.0 software.It indicated that the content of AFB₁ in peanut had spatial autocorrelation,and the semi-variogram of AFB₁contamination from 2009 to 2017 was spherical model.?3?The contamination map of AFB₁ was generated after the AFB₁ values in unsampled areas were obtained by ordinary kriging interpolation.It revealed that AFB₁ contamination presented scatterred distribution.It may relate with the different varieties,soil types,aflatoxin producing potential of fungi and climatic conditions in these areas.2.The trend of aflatoxin contamination of peanut in China during the nine consecutive years from2009 to 2017 was analyzed.The effects of resistance of different peanut varieties,soil types,aflatoxin producing potential of fungi and climatic conditions on the variation of aflatoxin contamination were discussed.The main results were as follows:?1?Aflatoxin contamination had annual difference.Aflatoxin contamination was heavier in 2017,2014,2015 and 2013 than that in other years,with total aflatoxins of 26.67±116.53?g/kg,17.49±72.01?g/kg,15.33±87.19?g/kg and 13.95±52.65?g/kg.It was the lowest in 2009?2.40±32.48?g/kg?and2010?4.32±35.15?g/kg?.?2?Aflatoxin contamination had regional differences.The contamination level of aflatoxin in peanut was the highest in the main producing areas of the Yangtze River Basin,with total aflatoxins range of 5.59-37.15?g/kg,followed by the southern areas?4.23-19.23?g/kg?and the northern areas?1.78-9.11?g/kg?.It was the lowest in the northeastern areas with total aflatoxins of no more than 1.00?g/kg.?3?Aflatoxin susceptible varieties,sandy soil,high aflatoxin producing potential strains,high temperature had higher proporation in non-compliant samples than compliant samples.High temperature regions were similar with high aflatoxin contamination level regions.The effects of influencing factors on the variation trend of aflatoxin in peanut were verified.It laid a foundation for the establishment of large-scale early warning technology of aflatoxin contamination in peanut.3.A peanut aflatoxin prediction model suitable for China based on rule classifier,balanced sampling and random forest was established.And then it was verified by data of 2018.The main results were as follows:?1?The main climatic variables finally were selected.According to the maximum information coefficient between climatic variables and aflatoxin,as well as considering the multi-collinearity and the comprehensiveness of the factors,the main climatic variables finally selected were latitude,precipitation from 8:00 to 20:00,average air pressure and average air temperature.?2?The rules for preclassification were made.By analyzing the relationship between the distribution of the main climatic variables and aflatoxin content,it can be seen that the effect of rule-making was remarkable in high latitude region?>40°N?,low latitude?<21°N?and low temperature region?average air temperature?16.8°C?.That was to say,in order to improve the accuracy of the model,the samples in these regions can be directly identified as non-compliant samples.?3?The results of the balanced sampling-Random forest model showed that the prediction accuracy was more than 95%.Four main climatic parameters,resistance of peanut varieties,soil types and aflatoxigenic potential of fungi were identified as the input parameters of the model.Balanced sampling-Random forest was used to classify and predict the aflatoxin contamination risk.When the classification threshold was 20?g/kg and 5?g/kg,the cross-validation results showed that the prediction accuracy of samples with AFB₁ content below threshold was 98.94%and 96.97%respectively,and the classification accuracy of samples with AFB₁ content below threshold was 100%.Compared with other commonly used classifiers?Decision tree,Support vector machine,K-nearest neighbor,Back propagation network and Radial basis function neural network?,Random forest can effectively separated the samples whether containing AFB₁ content more than limits?20?g/kg and 5?g/kg?.?4?Importance of the input parameters was analyzed.The importance of the input parameters for the classification model was further analyzed.The results indicated that when the threshold values were20?g/kg and 5?g/kg,importance of the 8 input parameters in descending order was latitude>mean daily temperature>precipitation from 8:00 to 20:00>mean air pressure>peanut varieties>soil types>aflatoxin producing potential of fungi.?5?The model validation results showed that the prediction accuracy was 86.89%.The mean content of AFB₁ in peanuts collected in 2018 was predicted by the rule classifier-balanced sampling-random forest classification model.Considering pre-classified samples,when the classification threshold was 20?g/kg and 5?g/kg,the results showed that the prediction accuracy of all the samples was 86.89%.