Crop Model Simulations And Uncertainty Analyses For The Responses Of Winter Wheat Production To Climate Change In The Loess Plateau

Global climate change has posed adverse impacts on wheat growth and development and changed its distribution of climate-suitable planting areas.At present,based on statistical models,crop models(CMs),and global climate models(GCMs),some relevant studies were conducted to investigate the impacts of climate change on unit yield,water use efficiency(WUE),and climate-suitable planting areas of winter wheat in the Loess Plateau of China.Then,some adaptive measures were recommend based crop model method.In assessment of impacts of history climate change on wheat production,most of the studies just focus on meteorology but lack insight of soils,management,and varieties,which could not able to get a comprehensive assessment.Most of the simulation results were mainly projected with a single crop model in these previous studies,which might lead to heavy uncertainties in the relevant simulation results due to the low level of model ensemble.The combinations of species distribution models(SDMs)and GCMs were the modish method in the predicted suitable planting distribution of crops under climate change.However,the prediction uncertainty caused by different sources was little know.Therefore,it is necessary to comprehensively assess the responses of production,WUE,and distribution of climate-suitable planting areas of winter wheat to climate change,and quantify the uncertainties in model simulation results before providing references for agronomic management measures and land policy adjustment for winter wheat production in the Loess Plateau.To this end,this study firstly evaluated the responses of winter wheat yield to historical(1981-2009)climate factors,extreme climate factors,management and soil through interpretable machine learning methods based on the observation data of winter wheat from26 agro-meteorological stations on the Loess Plateau.Then,we combined six different crop models(CMs)of wheat growth(e.g.,APSIM,Aqua Crop,DSSAT-CERES-Wheat,DSSAT-CSCRP-Wheat,DSSAT-NWheat,and STICS)and 27 global climate models(GCMs)to simulate winter wheat growth and development in three periods(1971-2010,2041-2060,and 2081-2100)under the SSP585 and SSP245 emission scenarios of Shared Socioeconomic Pathway(SSP)across 166 meteorological sites in winter wheat growing area of the Loess Plateau.The responses of winter wheat phenology and yield formation to climate change were investigated with the projections.Then,we combined 9 different species distribution models(SDMs)and 27 global climate models to project potential distributions of climate-suitable areas of winter wheat in three periods(1971-2010,2041-2060 and2081-2100)under SSP585 and SSP245 based on topography,soil mechanical composition,land use,and meteorology-related variables in the Loess Plateau.The unit yield of winter wheat simulated by the crop models and the climate-suitable planting areas of winter wheat predicted by the species distribution model were combined to project the response of reginal winter wheat production to climate change in the Loess Plateau.Quantifications of the uncertainties in the projections of winter wheat yield,climate-suitable planting areas and regional production under climate change were conducted with the ANOVA(analysis of variance)method on the Loess Plateau.Finally,based on five wheat growth models(e.g.,APSIM,DSSAT-CERES-Wheat,DSSAT-CSCRP-Wheat,DSSAT-NWheat,and STICS)and27 global climate models,we quantified the uncertainties in projected yield and WUE of winter wheat considering the suitable adaptive measures(e.g.,sowing date,irrigation,and fertilization)with three different methods(ANOVA considering the interaction of different sources,ANOVA without considering the interaction of different sources,and range precent)at three typical sites of Linfen,Yangling and Changwu in the Loess Plateau.The main findings and conclusions of the study were as drawn as follows.(1)Analysis of the impacts of historical climate change on the yield of winter wheat with interpretable machine learning methods in the Loess Plateau.Based on the Mann-Kendall(MK)trend test,the vegetative growth period was shorter than the reproductive growth period,which shortened winter wheat growing seasons from1981 to 2009 in the Loess Plateau.An obvious trend was found at some sites in Shanxi and Gansu province.The results indicated that the climate change from 1981 to 2009 could less explain the changes of winter wheat yield in the Loess Plateau through the first-order difference method.Finally,based on three interpretable machine learning methods(e.g.,PLSR,Randomforest,and RPART),it was found that soil and meteorology related variables were the most important factors contributed in the spatial-temporal changes of winter wheat yield,with about 19.8%～22.3% importance contributed by meteorological factors in the spatial-temporal changes of winter wheat yield in this region.(2)Assessment of the responses of winter wheat phenology and yield formation to climate change with crop model ensemble in the Loess Plateau.Based on the data typical field experiments of winter wheat the Loess Plateau,six different crop growth models were calibrated and validated,respectively.The relative root mean square error between the simulated and observed values was between 2% and 10%.The standard deviations of phenological period and yield simulated by the DSSAT-CERES-Wheat and Aqua Crop models were the smallest and largest among the six models,respectively.Based on the ensemble of the six crop growth simulation models,winter wheat yield were projected to increase by 12.4%-27.5% compared with the baseline(1971-2010)in the entire Loess Plateau.(3)Investigation of the response of potential distributions of climate-suitable areas of winter wheat to climate change with an ensemble of species distribution models(SDMs)in the Loess Plateau.The AUC(The area under the ROC curve),COR(The true skill statistic),and TSS(Pearson correlation coefficient)of the nine SDMs ranged from 0.7 to 1.0.The distribution of simulated climate-suitable areas of winter wheat with the nine SDMs shared a similar spatial pattern with the distribution of winter wheat planting areas based on remote sensing survey in the testing process.There were large variations in the trends and change magnitudes of projected climate-suitable areas of winter wheat by different SDMs.Generally,the SDM models of Glmnet,Maxlike,MLP and RBF projected small variation range(< 15%)of climate-suitable areas of winter wheat,while some models(e.g,GAM and GBM)projected variations as lager as ±50% in the climate-suitable areas of winter wheat.Based on an ensemble of SDMs,the climate-suitable areas of winter wheat would decrease by 14.95%and 51.06% in 2081-2100 under SSP245 and SSP585 scenarios in Gansu and Ningxia,respectively,which could lead to losses about 10.91% and 24.71% of the climate-suitable planting areas of winter wheat in the Loess Plateau.(4)Assessment of the impacts of climate change on winter wheat production with ensemble model-based approach in the Loess Plateau.Considering the changes of winter wheat climate-suitable planting area,there was larger variations in the projected trends with combinations of CMs and ensemble-SDM than with CMs of winter wheat in the Loess Plateau.The changes of winter wheat acreage had larger impacts on the changes of regional winter wheat production than the change of unit yield.The yield per unit of winter wheat would be increased while the climate-suitable area would be decreased under the two emission scenarios in 2081-2100,which showed projection uncertainty with combinations of CMs,GCMs,and SDMs.Based on the ensemble of 27 GCMs,6 CMs,and 9 SDMs,regional winter wheat productions were projected to increase by 14.57%,19.67%,4.85%,and 3.47% compared with the baseline in the Loess Plateau,respectively.(5)Analysis of the uncertainties in the projected unit yield,climate-suitable planting areas,and regional production of winter wheat under climate change in the Loess Plateau.The CMs were the largest source of uncertainty in the predicted unit yield of winter wheat in the Loess Plateau,with an uncertainty contribution of 38.8%.Among the sources of prediction uncertainties of climate-suitable areas of winter wheat,SDMs accounted for about25.4% of the total uncertainties,followed by the interaction between SDM and GCM.The main factors were no longer the largest uncertainty contributors in the predictions of regional productions of winter wheat in the Loess Plateau.The interaction between SDM and GCM was the largest uncertainty source in the prediction of regional production of winter wheat in the Loess Plateau,contributed about 20.9% of the total uncertainty.The uncertainty contribution by the SDMs was about 20.3%,which was the largest uncertainty source among the four main factors of CMs,SDMs,GCMs,and emission scenarios.SVM,RF and RPART might reduce the uncertainty of SDM in the prediction of winter wheat production,while Aqua Crop and CSCRP-Wheat might increase the uncertainty of CM in the prediction of winter wheat production in most of the Loess Plateau.(6)Quantification of the uncertainties caused by different adaptive measures to climate change for winter wheat production in the Loess Plateau with different methods.Based on the three uncertainty analysis methods of ANOVA considering the interaction of different sources(ANOVA all),ANOVA without considering the interaction of different sources(ANOVA main),and range precent,the CMs were always the largest uncertainty source for predicted unit yield of winter wheat at Linfen and Changwu stations,with uncertainty contributions of 30.6%-65.4%.However,adaptive measures would be the largest uncertainty source in unit yield projections at Yangling station.Additionally,CMs were consistently the largest uncertainty source in WUE forecasts.The relative uncertainty contributed by CMs in WUE projections was larger than the contribution in unit yield projections.In addition,the models of DSSAT-CERES-Wheat and STICS might increase the uncertainty in unit yield forecasts compared with the other crop models.