| The issue of climate change has received much attention all over the world.Agricultural production will be confronted with greater risks under the environment of global warming and frequent invisible meteorological disasters.Winter wheat production is greatly affected by meteorological conditions.Waterlogging,drought,high temperature and other meteorological disasters as well as the increase of their frequency pose a serious threat to wheat in the stable and high yield.With the proposal of "Climate Smart Agriculture" by the Food and Agriculture Organization of the United Nations(FAO),researches on climate adaptability and intelligent production in agriculture have being carried out in various countries and regions.On the basis of previous studies,we intended to analyze multiple meteorological factors and wheat yields during the wheat growth periods from 1979 to 2014 in three ecological regions of Jiangsu Province and design an integrated climatic assessment indicator(ICAI)based on machine learning models.By this indicator,the sensitive meteorological factors,the adaptability of wheat in different climatic years and the trends of wheat yield levels in future warming conditions were analyzed.The results can provide a reference for wheat yield prediction in Jiangsu,provide a theoretical basis for early warning of agricultural disasters,and provide technical support for agricultural planting insurance.The main results of this study are as follows:1.The temporal and spatial distributions of meteorological conditions were analyzed during winter wheat growth stages in Jiangsu Province over the past 35 years.The spatial distributions of mean temperature,accumulated precipitation,solar radiation and sunshine hours of 10-daily scales were analyzed during the growing seasons of wheat at 10 meteorological stations in Jiangsu Province from 1979 to 2014.The methods of Mann-kendall trend detection and Sen's slope were used for temporal evolution analyses of meteorological factors during four growth stages of wheat(S1:seedling stage;S2:over-wintering stage;S3:reviving&anthesis stage;S4:anthesis&maturity stage).The results showed that the spatial distributions of average temperature and precipitation in 10 stations during the whole growth periods were low in the north and high in the south,while the distributions of daily solar radiation and 10-daily sunshine hours were the opposite.The average temperature of the whole growth periods of wheat at 10 stations showed significant upward trends and the range was between 0.04?/yr and 0.07?/yr.In the four growth stages,significant warming trends appeared in nearly half of the stations during S2,in 9 stations during S1 and S4,and in all stations during S3.By comparison,the temperature rise of S3 was the highest among the four growth stages,and the highest value appeared in Wujiang,reaching 0.092?/yr.In addition to average temperatures,the daily maximum and minimum temperatures of wheat growing seasons also displayed varying degrees of increase.The increases of daily minimum temperatures at Ganyu,Tongshan,Dongtai and Jiangning stations were close to or exceeded the increases of daily maximum temperature from 1979 to 2014.In the S1,S2 and S3,the minimum temperature at all four stations showed significant upward trends,reflecting the imbalance of temperature increase in wheat growing seasons.The average 10-daily precipitations in the whole growth periods of wheat did not show significant upward or downward trends in 35 years.During the four growth stages,only the trends of precipitations in S2 at Liyang and Wujiang stations reached significant levels with Sen's slope above 0.3mm/yr.The relationship between wheat yields and water conditions over wheat growing seasons in Jiangsu Province was analyzed by SPEIx which was a drought index with 10-daily scales.The results showed that the SPEIx values of S4 at scales of 10,20 and 30 days were negatively correlated with the first-order differential yields,indicating that the excessive rainfall during S4 was adverse to wheat yields in Jiangsu Province.The Sen's slopes of the average daily solar radiation at all stations in the whole growth periods of wheat over the past 35 years were positive,but the upward trends did not reach significant level.The solar radiations of 9 stations showed significant upward trends during S3 and 5 of them in Central and Southern Jiangsu reached extremely significant levels.The trends of the other three growth stages were not significant.The Sen's slopes of sunshine hours during the whole growth periods of winter wheat at most stations were negative,but only the downward trend at Xuyi station reached an extremely significant level.There were 4 sites at S1 and 1 site at S2 showing significant downward trends,while 3 sites at S3 showing significant upward trends,and other sites showing no significant trends.2.Key meteorological factors that affect climatic yields of wheat were extracted and the extraction methods were compared.According to the differences of climatic conditions and planting division,three sub-regions:Northern,Central and Southern Jiangsu were divided by the northern irrigation canal and Yangtze River.The relationships between meteorological factors and climatic yields in four growth stages of wheat were analyzed by different methods,and key meteorological factors affecting climatic yields were selected in three sub-regions.Four different methods of extracting climatic yields(LT:de-linear trend method,3MA:de-trended method with three-year moving average,Diff:first-order difference method and Rdiff:relative difference method)were compared.The results showed that in three sub-regions,the average correlation coefficients of climatic yields obtained by 3MA were low in Northern and Central Jiangsu but were high in Southern Jiangsu.The correlations of climatic yields obtained by LT were high in Northern Jiangsu and low in Southern Jiangsu,while the correlation results with Diff and Rdiff were close and both were stable in three sub-regions.The correlation analysis between four meteorological factors and climatic yields in four key growth stages showed that the number of correlation factors and Spearman correlation coefficient values of Diff and Rdiff were better than those of LT and 3MA.We identified 20 lowest climatic yield values as serious yield reduction in three sub-regions and compared the fittness of climatic yields.The results showed that the climatic yields with 3MA and Rdiff were well fitted in Central Jiangsu,while the climatic yield with LT was well fitted in Northern and Southern Jiangsu.The climatic yields with Diff was fitted well in Southern Jiangsu.In view of the simplicity in calculation and strong explanatory ability of Diff,it was determined as the calculation method of climatic yield.The meteorological factors affecting wheat climatic yield were selected by correlation analysis,AIC of stepwise regression model and%IncMSE index of random forest model.The contrastive analysis showed that the meteorological factors screened by three methods were not identical,but some factors showed high importance in all three methods.Therefore,combining three methods to select meteorological factors may obtain more reliable factors,which can be used as input parameters of the climatic yield fitting models in order to obtain more accurate meteorological index.3.Integrated climatic assessment indicator(ICAI)for wheat yields was established.The spatial and temporal prediction modes of ICAI were designed to meet the practical needs.Spatial prediction was used to predict yields levels for different stations using the data of stations with known yields at the same periods in the same region.Temporal prediction was used to predict unknown yield levels by historical data of the same stations.The regression models were established by Random Forest(RF)and Support Vector Machine(SVM).Training sets and testing sets were divided.The accuracy of testing sets showed that R2 of RF and SVM models were more than 0.5 in Northern Jiangsu,but less than 0.4 in Central and Southern Jiangsu.By Kolmogorov-Smirnov goodness-of-fit test,it was clarified that the distribution of climatic yields was complied with 3-parameter Student T distribution.According to the distribution,the output values of the regression models were standardized and transformed into ICAI index which were comparable in temporal and spatial scales.According to the upper and lower 20%quantiles of ICAI(corresponding climatic yields are:388 kg/ha and-267 kg/ha),the classification thresholds were set.That is,the index was divided into three categories(classification thresholds wereħ0.84)with the probability of 20%,60%and 20%to indicate three levels of climatic yields:reduction,normal and increase.After classification,the prediction accuracy of the index was between 50%and 97%.In addition,the accuracies of the SVM-based index were higher than those of the RF-based index.ROC curves were used to assess the prediction ability of the indices calculated by SVM and RF models for two categories:climatic yield reduction and non-reduction(or increase and non-increase).The results showed that the prediction accuracy was high in Northern Jiangsu for both RF-based index and SVM-based index under the yield reduction probability set by 40%,with the highest AUC values of 0.98.In Central Jiangsu,the prediction accuracy was low for yield reduction and increase.In Southern Jiangsu,both SVM-base spatial index and RF-based temporal index had high prediction accuracy.4.The prediction accuracy of ICAI in assessing wheat climatic yields was evaluated.In order to analyze the accuracy of ICAI in climatic yield assessment,Standardized Precipitation Evapotranspiration Index(SPEI)of 1,2,3 and 4 month scales were calculated from November to May of the next year during wheat growing stages,and the relationship between SPEI indices and climatic yields were studied.The prediction performance of SPEI and ICAI were compared with the same data set.The results showed that in spatial prediction,accuracies of the two indices were the same in Northern and Central Jiangsu,while the accuracy of ICAI was higher than that of SPEI in Southern Jiangsu;in temporal prediction,the accuracy of SVM-based ICAI was the same as that of SPEI in Northern Jiangsu,but higher in Central and Southern Jiangsu.The prediction accuracy of RF-based ICAI was lower than that of SPEI,except in Central Jiangsu.Stepwise regression models were constructed for climatic yield prediction by using meteorological factors which reflect light,temperature and water conditions.The comparison results of the stepwise regression models and ICAI showed that the prediction accuracies of the RF-based and SVM-based ICAI in Northern,Central and Southern Jiangsu were higher than those of the stepwise regression models with multiple meteorological factors.Using a new testing set which was different from the original training sets and testing sets,the prediction accuracy of ICAI in six stations of Jiangsu over 2015?2017 was verified.The results showed that the prediction accuracy of the index was more than 61%aggregated by all three regions,but the accuracy is low in Southern Jiangsu.Under two adjusted thresholds the accuracy declined.It suggested that the index has practical values under the original thresholds.5.Climate adaptability of wheat production in Jiangsu Province and sensitive meteorological factors in different regions were analyzed.The sensitive meteorological factors,the climate adaptability in different years and the changes of wheat yields under future warming environment were analyzed by using ICAI.The global sensitivity indices were calculated by using the RF temporal prediction models in Northern,Central and Southern Jiangsu.The results showed that the most sensitive meteorological factor for wheat in Northern Jiangsu was the solar radiation at S2,in Central Jiangsu,it was the solar radiation at S4,and in Southern Jiangsu,it was the average temperature at S2.The variations of index values at 10 stations of Jiangsu Province during 1981-1990,1991-2000 and 2001-2014 were analyzed.It showed that the ratio of lean year was the highest in 1990s among three decades.In the early 14 years of this century,the climate adaptability of wheat in Jiangsu has generally improved.The ratio of lean year was the lowest in the three decades,falling below 20%;the ratio of normal year reached the highest value in three decades,namely,in Northern,Central and Southern Jiangsu,the values were 82%,69%and 68%respectively;the ratio of increase year was over 14%in all three sub-regions,reflecting the steady growth trend of wheat yield in Jiangsu since 2001.The RF temporal prediction model was used to analyze the changes of wheat yields under the warming environment in the future.The Sen's slope of average temperature in 10 stations of Jiangsu during 35 wheat growth periods was 0.05?/yr,hence we set average temperature of wheat growth period increased by 0.025?/yr,0.05?/yr and 0.1?/yr,respectively,to investigate the change of ICAI.The results showed that in Northern Jiangsu,the changes of index values under three warming levels displayed consistent decline in the ratio of normal year and rise in the ratio of lean year and increase year;in Central and Southern Jiangsu,it displayed decline in the ratio of normal year and rise in the ratio of increase year under three warming levels.It suggested that the index can provide assistant decision-making for wheat production under uncertain climate conditions in the future. |
PDF Full Text Request