Applicability Of Phenological Models In The Main Maize Planting Regions Of Northeast And North China Under Climate Change

Climate change characterized by warming has profoundly affected crop production systems.How crop production systems respond and adapt to climate change is a major scientific issue related to national food security strategies,and related researches have attracted great attention from the scientific community.In a warming environment,the phenology of major crops may be advanced/delayed,and the number of days required to complete the developmental stage also changes accordingly.Crop growth simulation models are the primary instrument for predicting crop developmental responses to climate change.The phenology models based on the assumption of constant accumulated temperature/photothermal time have been mainly used.Different types of phenological models have been developed over the past few decades.It has been recently reported that the hypothesized phenological model of nonconstant accumulated temperature/photothermal also has better adaptability.These different hypothetical models are different in mechanism and structure,which leads to large uncertainties in predicting the impact of warming on developmental stages.It is urgent to evaluate the applicability of different phenology models under climate warming conditions.Provide scientific basis for prediction of developmental period under climate change environment.Based on observation data from meteorological stations and agrometeorological stations,this paper analyzed agroclimatic resources of maize growth season and their climatic trend rates of spring maize in Northeast China and summer maize in North China from 1980 to 2020.The simulation effects were discussed by comparing accurate modeling of maize phenology from selected phenology models.The models supported two different hypotheses: constant accumulated temperature/thermal time hypothesis and non-constant accumulated temperature/thermal time hypothesis.The seven of the models were taken from MAIS,SIMCOY,EPIC,MCWLA,WOFOST,Beta and CERES to represent the former hypothesis,and 2 models were taken from RAM(response & adaptation model)and NSM(no simulation model)to represent the latter.Root mean square error(RMSE),normalized root mean square error(NRMSE),simulation accuracy(SA),mean absolute error(MAE),index of agreement(d)and systematic bias were used to test and evaluate the deviation between simulated and observed maize phenology of these models.The machine learning algorithm(MLP),MAIS model,RAM model,and Beta model were compared in a warming climate,in order to provided scientific support and reference for predicting the phenological development period of maize under climate warming.The main results are as follows:(1)Over the past 40 years,significant agroclimatic resources change has occurred across the Northeast China,corresponding to global climate change.In this study,the average temperature,maximum temperature and minimum temperature during the growing season of maize in Northeast China showed a significant increasing trend.The natural precipitation of planting area can basically meet the needs of maize growth and the sunshine hours showed a significant decreasing trend.Historical data showed that the flowering date of spring maize in Northeast China was advanced and the timing of maturity was delayed.Effect accumulated temperature over 8 ℃ increased from northeast to southwest and showed significant increasing trend at many sites.The number of days in the reproductive growth stage increased,and the number of days in the developmental stage was negatively correlated with the average temperature in the same period.The relationship was particularly obvious at the emergenceflowering and emergence-maturity stages.The agroclimatic resources changed across the North China Plain due to significant climate change,the average,maximum,and minimum temperature during the summer maize growing season showed a significant increasing trend,the natural precipitation remained basically stable across the study area,and the sunshine hours during the maize growing season decreased significantly.The emergence period of summer maize fluctuated at different sites,but the timing of flowering and maturity were dominated by advanced and delayed trends,respectively.Effect accumulated temperature over 8 ℃increased from northeast to southwest and showed significant increasing trend at many sites.The number of days from emergence to maturity was negatively correlated with the average temperature over the same period.(2)There were large differences in the performance of nine phenology models used to simulate the flowering and maturity dates of spring maize in Northeast China and summer maize in North China Plain.Among non-constant accumulated temperature/thermal time hypothesis models,the RAM model has the best simulation performance,followed by NSM model.For models with different mechanisms under the constant thermal/photothermal accumulation assumption,MAIS was the best of the models that did not consider the three cardinal temperatures and photoperiod,followed by SIMCOY.For the models that considered the three cardinal temperatures and did not consider photoperiod,MCWLA was the best,followed by EPIC.For the models that considered the three cardinal temperatures and photoperiod,Beta was the best,followed by WOFOST and CERES.Comparing the overall performance of the nine models,RAM was the most suitable model to simulate the maize growth period in Northeast China and North China Plain.(3)The four models with the lowest RMSE values during RGP(RAM,Beta,NSM,and MAIS)had one feature in common,i.e.,they could simulate a remarkable development rate even at temperatures only slightly higher than the base temperature.Therefore,appropriately combining a function that can remarkably simulate development rate at low temperatures in the models based on non-constant thermal/photothermal assumption may be a promising way to further improve phenology models,especially for models that are to be applied under climate variation conditions.(4)The machine learning algorithm(MLP)and three mechanistic phenology models(MAIS,Beta and RAM)were compared in this study in a warming climate.It showed that the machine learning algorithm simulation outperforms the mechanistic phenology model at calibration in cold years.However,when the temperature in the growing season increased,the simulation accuracy of MLP decreased significantly,while the performance of Beta and RAM in the mechanistic phenological models performed well.Overall,models did not benefit from both calibration and validation.MLP performed well at calibration in cold years,but poorly in warm years.The overall performance of the mechanistic phenology models was worse than MLP at calibration,but it performed well in warm years.Different models were appropriate for various contexts.MLP can be recommended to precisely reverse the impact of historical climate change on the growth period.However,mechanism models should be used to precisely predict the impact of future climate change on growth period.It is worth noting that we should use as much data as possible to improve the generalization ability of MLP when apply the model.