Observation And Spatiotemporal Simulation Of Carbon And Water Fluxes On Complex Underlying Surfaces

In the context of global climate change and the convening of the Global Climate Summit,the issue of carbon emissions has received unprecedented attention.City is the source of global CO₂emissions.Under the policies of"peak carbon dioxide emissions"and"carbon neutrality",studying the mechanism of carbon cycle and reducing carbon emissions is one of the most effective ways to achieve"peak carbon dioxide emissions"and"carbon neutrality".Shanghai is the most populous international metropolis in China.Its urbanization rate is higher than the average level in China.The urban surface process is constantly updated and changed,and the proportion of impervious areas is also increasing year by year.The carbon cycle research in urban complex underlying surface based on the eddy-covariance observation system is also of great theoretical and practical significance for better understanding the carbon cycle process on the land surface and predicting the carbon dynamics of future urban ecosystems,which can ultimately achieve the goal of better responding to the global climate change.Based on the observation data and multi-source remote sensing data obtained from field observation sites and meteorological observation fields in the suburbs of Shanghai,this study on the long-term dynamic analysis of the carbon and water fluxes of the urban complex underlying surface and the simulation of urban complex underlying surface carbohydrate flux by machine learning were carried out in an orderly manner.Firstly,the long-term dynamic change of meteorological background in the study area was analyzed based on the long-term observation data from 2010 to 2019,and the daily impermeable surface area was quantified by using flux footprint model.Then,the long-term dynamic change characteristics of carbon and water fluxes on the complex urban underlying surface and the influences of environmental factors and impermeable surface area on its change were analyzed.Secondly,four kinds of machine learning algorithms were used to simulate and predict the long-term carbon and water fluxes of urban complex underlying surface in a long time series,and the simulation performance of tne four were evaluated and compared.Finally,the optimal model was used to simulate the spatiotemporal distribution pattern of carbon and water fluxes in Fengxian District,Shanghai,and the main factors affecting the spatial and temporal pattern of carbon and water fluxes in Fengxian District were analyzed,so as to clarify the variation characteristics of the spatiotemporal distribution of carbon and water fluxes.The main research contents and conclusions are as follows:(1)Based on the analysis results of the daily average value of atmospheric temperature from 2010 to 2019,the temperature and rainfall in the study area showed a slow growth trend,with an average annual temperature increase of about 0.06?.From the perspective of seasons,it is found that the temperature rised most obviously in autumn,with rainfall increasing by 372 mm,and the rainfall contribution mainly came from summer;The average relative humidity variability between years was low,showing a relatively stable state,with an average annual humidity of 77.92%;The distribution of wind speed and direction was characterized with obvious spatial differentiation characteristics,that is,the main wind direction is the southeast wind,and the wind speed is mainly concentrated in 0-6 m/s,which accorded with the basic wind direction characteristics of the subtropical monsoon climate;The total solar radiation,net radiation and photosynthetically active radiation in 2011,2012,2017,2018 and 2019.The daily mean values of effective radiation all showed an increasing trend with the characteristic of single-peak,besides,the radiation value was the largest in summer and smaller in spring and winter.(2)By analyzing the observation data of carbon and water fluxes in the study area in 2011,2012,2017,2018 and 2019,it was found that the average daily carbon sink time in 2012 was the longest,up to about 10 hours;From 2011 to 2019,the monthly average CO₂ flux showed a decreasing trend as a whole.In the past five years,the average value of CO₂ flux showed a decreasing trend in summer and an overall increasing trend in winter.The carbon sink capacity of the whole ecosystem increased with time.The daily dynamics of CO₂ concentration in 5 years in 2011,2012,2017,2018 and 2019 had obvious temporal differentiation characteristicse with two peaks in a day:the maximum value generally appeared around 5:30 in the morning,while the minimum value generally appeared at about 15:30.The daily variation trend of H₂O flux in 5 years was opposite to that of CO₂ flux.H₂O flux reached its peak around 12:30noon,and the maximum peak range was 11940.440-16559.907 mg/m3,and became steady after 18:30 pm.During the five years,the H₂O flux in the four seasons showed a trend of polyline growth,and the H₂O flux reached the peak in summer on the interannual scale.There were two peaks of H₂O concentration every day,the first peak appeared at 5:30 in the morning,and the second peak appeared at 12:30.Through the analysis of the correlation between environmental factors,impervious surface area and CO₂ and H₂O fluxes,environmental factors such as atmospheric temperature(Ta),relative humidity(RH),temperature of the soil surface(Ts?10cm),and net radiation(Rn)were negatively correlated with CO₂ flux in the order of Ta(-0.819)>Rn(-0.372)>Ts?10cm(-0.298)>RH(-0.287),and positively correlated with H₂O flux in the order of Ta(0.69)>Rn(0.403)>Ts?10cm(0.31)>RH(0.191),among them,the most influential factor on CO₂ and H₂O fluxes was atmospheric temperature.The proportion of impervious surface area was positively correlated with CO₂ flux and negatively correlated with H₂O flux.Atmospheric temperature had the highest contribution to the changes of CO₂ and H₂O flux,Ta contributed more than 60%to the variation of CO₂ and H₂O fluxes,followed by Ts?10cm and Rn,which can explain more than 13%of the change trend of CO₂ flux,and explain more than 7%of the change of H₂O flux.The contribution of the impervious surface area to the change of CO₂ flux was 8.23%,and the contribution to the change of H₂O flux was 3.119%.(3)In the simulation of carbon and water fluxes in urban complex underlying surface based on four different machine learning algorithms,the Long Short-Term Memory Network(LSTM),the Support Vector Machine(SVM),the Random Forest(RF)and the Artificial Neural Network(ANN)all accurately estimated the carbon and water fluxes of urban complex underlying surface over a long period of time.Among them,the RF model had the best performance in simulating CO₂ flux,with the highest R² value(0.852),the lowest RMSE value(0.293?mol·m^-2·s^-1)and MAE value(0.157?mol·m^-2·s^-1),The second was the SVM model,while the ANN model had the lowest simulation performance.The simulation performance of the LSTM was between the SVM model and the ANN model.When the three machine learning models simulated the long-term H₂O flux of urban complex underlying surfaces in Shanghai,the RF model had the highest simulation performance with the highest R² value(0.897),the lowest RMSE value(0.275 mmol·m^-2·s^-1)and the MAE value(0.179 mmol·m^-2·s^-1),followed by ANN model,while SVM model performed the worst.The simulation performance of the LSTM models was second only to RF model.Therefore,the RF model had the highest accuracy in simulating the carbon and water fluxes.(4)The machine learning method-RF can accurately simulate the spatiotemporal variation characteristics of carbon and water fluxes in Fengxian District,Shanghai.By comparing the simulation performance of the RF model trained with and without the impervious surface area input,it was found that when the impervious surface area was input,the R² value(84.46%)of the CO₂ flux model simulated by the RF model was higher than that without the impervious surface area input,and the R² value(89.80%)of the H₂O flux model with RF model is 0.19%higher than the R² value(89.61%)when the impervious surface area was not input.The carbon and water fluxes in Fengxian District of Shanghai in 2011,2012,2017,2018 and 2019 had strong spatial heterogeneity:the CO₂ flux value in the western part of Fengxian District was lower than that in the eastern part,and the CO₂ flux value gradually increased from the western part to the eastern part of Fengxian District.The H₂O flux in the east was lower than that in the west,and the H₂O flux value gradually decreased from the west to the east of Fengxian District.The variation characteristics of interannual of carbon and water fluxes also showed relatively obvious spatial characteristics.The average annual CO₂ flux value in more than 90%of Fengxian District was reduced,distributed in Shanghai Harbour Comprehensive Economic Development Zone in the east of Fengxian District,Nanqiao Town and Zhuanghang Town in the west;The average annual H₂O flux in 55.1%of the regions,mainly distributed in Situan Town and Bay Town in Fengxian District.Among meteorological factors,soil moisture played an important role in the spatial pattern of carbon and water fluxes,while the effects of temperature and radiation were not obvious in the small study area.