Phenological Changes Of Typical Woody Plants In Northeast China And Their Response To Climate Warming

In the past few decades,the temperature of the northern hemisphere has risen rapidly,and the plant phenology such as leaf unfolding date(LUD)and first flowering date(FFD)has been significantly advances.Plant phenology has been shown to be a sensitive biological indicator of climate change,and phenological models can effectively extrapolate historically missing phenological data to study long-term relationships between plants and climate change.Although a large number of ecophysiological and statistical models have been developed and performed well in the training data set of the model,the prediction effect of these models on the test set is not satisfactory,and the poor universality leads to the failure of the model to be popularized.In natural science research,although compared with the traditional statistical method,machine learning(ML)algorithm has shown obvious advantages,but it has not been widely used in phenological model research.In this study,the phenological observation data of typical northeastern woody plants collected from 1963 to 2018 were combined with meteorological data of observation stations,and the ecophysiological models,traditional statistical models and machine learning algorithm models were applied to predict phenology,and the performance of different types of models was compared.The results show that the differential evolution algorithm is superior to the other four algorithms in parameter optimization of physiological ecological model.By comparing the predictions of the four types of ecophysiological models(20 in total)for spring phenology(LUD and FFD),the results suggest that the relatively simple One-phase model is more pervasive at the species level,and the RI(Remainder Index)model of all the models is the best.T he performance of K Neighbors Classifier in the machine learning models is superior to that of the other 17 dichotomous machine learning models.The results show that the phenology of typical woody plants in Northeast China changed significantly from 1963 to 2018,including the weather before and after the temperature changepoint(Colder:1962-1988;Warmer:1988-2018).The four phenological stages of the observation station have changed significantly,and the average phenological date before and after the changepoint is significantly different.From 1989 to 2018,the mean values of FFD and late flowering date(LFD)were 7.7,3.9 and 2.8 days earlier than those of 1962 to 1988.The end of season(EOS)was significantly delayed by 6.5 days.The LUD period of woody plants showed an advance trend,with an average advance of 0.40 days/year.In 88%of the species,LUD started earlier.For FFD and LFD,the phenological period were advanced,with an average advance of 0.45 and 0.42 days per year.Among them,95%of species showed early FFD and 89%showed LFD.Different from other phenological time variation trends,the EOS of woody plants showed a significant trend of delay,with an average delay of 0.31 days/year,and 77%of species showed late EOS.Combined with the phenological model,the response of typical northeastern woody plants to climate change was studied.The results showed that the onset period of typical northeastern woody plants exhibited an advance trend of phenological events with the increase of temperature in the relevant sensitive period,and the average temperature sensitivity coefficient(ST)was-2.94 days/?.In 99%of the species,the onset date of LUD was significantly advanced with the increase of temperature in the relevant sensitive period.At the FFD and LFD,the phenological stage tended to advance with the increase of temperature in the temperature sensitive period,and the average temperature sensitivity coefficient was-3.54 and-2.98 days/?.Among them,97%of the species showed a trend that the phenological period advanced with the increase of temperature in the temperature sensitive period.96%of the species showed a trend that the phenological stage at LFD advanced with the increase of temperature in the temperature sensitive period.Different from the distribution of temperature sensitivity coefficients in other phenophase,woody plants showed both advance and delay trend with the increase of temperature sensitive period at EOS.The average temperature sensitivity coefficient was 3.29 days/?,and 75%of the species showed significant delay in EOS with the increase of temperature at temperature sensitive period.Gradient Boosting Decision Tree model and three widely used physiological and ecological models were respectively used to fill the phenological data of the missing years in all species in the onset period of LUD whose observed years were greater than 20 years,so as to analyze the variation of temporal change of phenological response to climate change.The results showed that from 1962 to 2016,the LUD of all species was significantly advanced with the increase of temperature in the response period,with an average advance of 2.76±0.24 days/?.There were significant differences in temperature sensitivity between the warm period and the cold period,averaging 2.18±0.25 days ? 1962-1987 and 2.83±0.41 days/? 1988-2016(consistent with the data conclusions filled in by three ecophysiological models).82.5%of species(33 out of 40)experienced an increase in ST between the two periods.The time dynamic changes of ST in the 15-year moving window during the cold period and the warm period showed that the time change of ST decreased significantly in both periods.During the cold period(1962-1987),ST of all species decreased significantly by 0.87 days/? every 10 years,from 2.89±0.34 days/? in 1962-1976 to 2.11±0.30 days/? in 1973-1987,a decrease of 37.2%(p<0.001).Most species have a similar decline in ST,with the magnitude of the decline varying from species to species.During the warm and cold periods,the decline of ST in late spring was larger.In this study,we compared the performance of five optimization algorithms in the multi parameter optimization process of ecophysiological model,and compared the performance of 20 ecophysiological models in the phenological prediction in spring(LUD and FFD).The binary classification machine learning algorithm model is successfully applied to plant phenology prediction for the first time.The results show that compared with the ecophysiological model and the traditional statistical model,the binary classification machine learning model can greatly improve the accuracy of phenological prediction,and can be applied to all phenophase,and integrate various meteorological factors and even environmental factors that affect phenological changes.There is a certain gap between the predicted value and the observed value of machine learning model,which may be caused by the quality of phenological observation data,the sample size of observation data and the quality of meteorological data.Ecophysiological phenological models may underestimate phenological responses to climate warming.We found that temperature sensitivity generally decreased with the increase of temperature fluctuation in response time at the species level.Our study suggests that machine learning algorithms should be used more widely in future phenological model studies,and that time-scale changes in temperature sensitivity should be studied more deeply to expand our understanding of plants' ability to adapt to future climate change.