| Fusarium Head Blight(FHB)is a serious threat to wheat production and has become a worldwide problem affecting the sustainable development of wheat industry.Since the mid-1990s,the disease had gradually evolved from sporadic to continuous occurrence in Hebei province,with an average annual occurrence area of more than 2.67×105 hm2.FHB shows rapid outbreak,large area and heavy loss during epidemic years,indicating that it has risen from a minor disease to a major disease.Based on the analysis of the temporal and spatial characteristics of FHB in Hebei province,the maximum entropy(MaxEnt)model was used to predict the risk area of FHB.The composition and pathogenicity of Fusarium graminearum in the main wheat producing areas in Hebei province were clarified,and the resistance of 103 recommended wheat cultivars was evaluated.The meteorological factors that had a significant impact on the occurrence of FHB were screened,and the prediction models based on multiple linear regression,boosted regression tree and random forest algorithm were established.All these results not only provide technical support for FHB prediction in Hebei province,but also provide reference for the optimization and improvement of FHB prediction model.The main results were as follows:1.Based on the occurrence and distribution characteristics and environmental variable data of FHB in Hebei province,MaxEnt model was used to predict that the area of high-risk and medium-risk of FHB accounted for 14.98%and 10.19%of the total wheat area in Hebei province,respectively,which mainly concentrated in the middle and south of Hebei province.Among them,66 counties were high-risk areas including the south of Baoding,the middle and east of Shijiazhuang,Hengshui,the middle and east of Xingtai,and the middle and east of Handan.The AUC value predicted by MaxEnt model was 0.816.The importance analysis results of environmental variables obtained by jackknife cutting method showed that the average temperature in the warmest quarter(bio 10),the maximum temperature in the warmest month(bio 5),the average temperature in the coldest quarter(bio 11)and the minimum temperature in the coldest month(bio 6)had a great impact on the occurrence of FHB.Among them,bio 10 had the highest relative contribution rate,up to 67.9%,and its importance accounted for 22.2%.2.The time dynamic analysis of FHB in Hebei province showed that the standardized values of FHB occurrence area in 2003,2004,2005 and 2018 were all more than 1.0,and the disease occurrence varied greatly.The standardized values of wheat yield loss in 2003,2009,2010 and 2017 were more than 1.0,with a great variation.The yield was reduced by 4.18×107 kg,3.91 ×107 kg,3.54×107 kg and 3.20×107 kg,respectively,accounting for 0.41%,0.32%,0.29%and 0.21%of the total yield of the year.The spatial analysis of FHB showed that there was more than medium spatial correlation in Hebei province.3.575 Fusarium strains were isolated and identified from 17 counties of 7 cities in Hebei province.F.graminearum was the dominant pathogen,and the isolation frequency was as high as 96.00%,followed by F.semitectum and F.proliferatum,with the separation frequencies of 2.61%and 1.39%,respectively.4.103 wheat cultivars mainly promoted in Hebei province were evaluatedand most of them were susceptible.There was no immune cultivar to FHB.Medium resistant,medium susceptible and high susceptible cultivars accounted for 24.27%,13.59%and 61.17%,respectively.Among them,25 cultivars,such as Henong 7069,Heng 4399 and Yuanda 1,were medium resistant,and 14 cultivars such as Shijiazhuang 8,Heng 136 and Nongda212 were medium susceptible.Meanwile,63 cultivars such as Jimai 21,Jinmai 59 and Henong 826 were highly susceptible to FHB.The length of fawn,spike neck and lower spike node were significantly negative correlated with their disease resistance,and the correlation coefficients were-0.3429,-0.2951 and-0.2841,respectively.There was a lowly negative correlation between plant height and disease resistance,and the correlation coefficient was-0.2118.5.Three prediction models were constructed based on meteorological factors.(1)A prediction model based on multiple linear regressions.The values of model R2 and corrected R2 were 0.8158 and 0.8018,respectively,which was y=-13.2427+0.3145 LT-65-0.9824 MWS.55+0.1209 MRH-55+0.1377 Rain-35一0.4184 MT-25+0.08143 SD15+0.2802 MRH15-0.8832 DRain15.Among them,LT-65 was the lowest temperature from the 26th to 30th days before the initial date of heading stage.MRH-55 was the average wind speed from the 21st to 25th days before the initial date of heading stage.Rain-35 was the average relative humidity from the 21st to 25th days before the initial date of heading stage.Rain-35 was the total rainfall from the 11th to 15th days before the initial date of heading stage.MT-25 was the average temperature from the 6th to 10th days before the initial date of heading stage.SD15 was the sunshine hours from 1st to 5th days after the initial date of heading stage.MRH15 was the average relative humidity from 1st to 5th days after the initial date of heading stage.DRain 15 was the rainfall days from 1st to 5th days after the initial date of heading stage.Three key influencing factors of FHB(MRH15、Rain-35 and MRH-55)were selected.(2)A prediction model based on boosted regression tree.Its learning efficiency(lr),tree complexity(tc),sampling ratio(bf),function loss form,cross validation discount and n.trees were 0.005,6,0.75,Gaussian,10 times and 5000,respectively.The importance of MRH-55,Rain-35,MRH15,SD15,LT-65,MWS-55,MT-25 and DRain15 were 69.62%,14.08%,4.89%,4.34%,3.35%,2.02%,1.20%and 0.50%,respectively.The ROC curve analysis method was used to verify the prediction results of the training set and the test set.The AUC values of the prediction results were 0.950 and 0.931,respectively.(3)The fitting model based on random forest.The values of mtry and ntree were 2 and 300,respectively.The prediction model was evaluated by confusion matrix.The accuracy of training set and test set were 0.9043 and 0.9032,respectively.The prediction results of training set and test set were verified by ROC curve analysis method.The AUC values of prediction results were 0.891 and 0.876,respectively.According to the verification results of historical data in 2008,2010 and 2012,the accuracy of multiple linear regression model,boosted regression tree model and random forest model in predicting FHB ear rate were 88.43%,87.72%and 90.91%,respectively.The prediction coincidences of FHB incidence were 87.88%,81.82%and 93.94%,respectively.The prediction results were basically consistent with the actual observation values. |
PDF Full Text Request