Dynamic Simulation And Analysis Of Carbon And Water Fluxes In China Based On Model-data Fusion

Accurately quantifying and predicting terrestrial ecosystem carbon water flux is of great significance for understanding interactions between land and climate,managing resources,constructing regional ecological civilization,predicting future climate change and controlling greenhouse effect.Flux observation and model simulation are two main methods which are used to study carbon and water flux.The flux observation has a higher observation accuracy,but its observation range is limited,and its site distribution is not uniform.In addition,it is easy to be affected by the environment and it is difficult to conduct a zone spreading.Model simulation can realize different scales of parameter estimation,but due to the limitation of idealized assumption,model parameters and driving data,there is often a bigger deviation between simulation results and real values.Model-data combination method establishes a mutually restricting and adjusting optimization relation through parameter estimation and data assimilation which are two kinds of technology integration observation and model information,so as to improve the matching degree between model results and real values.Based on this idea and on the basis of meteorological,soil and CO₂ data,this study focuses on the optimization of sensitive parameters of LPJ-DGVM and obtains the suitable parameterization scheme for China.On this basis,we uses the data assimilation algorithm to integrate remote sensing product information with optimized LPJ-DGVM.Then,the original model simulation track is constantly corrected in the simulation process,so as to improve the model's applicability.The above results were verified by the sites and then they are promoted to the Chinese region to simulate and analyze the spatial pattern of GPP and evapotranspiration ET of the total primary productivity of China from 2000 to 2015.The main conclusions are as follows:?1?22 adjustable parameters?including 7 function types:photosynthesis,respiration,water balance,allometry,mortality and establishment?selected from LPJ-DGVM are used to randomly obtain different parameter combinations within their respective value scope,and the results show that 22 parameters can cause a greater uncertainty of GPP and ET simulation results,especially focus on the growing season.The relative uncertainty?RU?of GPP in all sites is between 0.9-1.25 and there is no obvious inter-annual variability,while the change trend of ET relative uncertainty among months is obvious,and is under 0.5,which is significantly lower than GPP,indicating that the 22 parameters selected have a more significant impact on GPP simulation.?2?LPJ-DGVM is highly nonlinear,and there are many parameters in the internal and the parameters affect with each other.So there is a large calculated amount if they are demarcated one by one,and the simulation accuracy can be improved very little.Therefore,this study adopts the global sensitivity analysis method EFAST to respectively conduct a sensitivity analysis on five parameters related to carbon and water in the model?GPP,NPP,R_h,ET and Runoff?,so as to obtain the parameters'first order and global sensitivity index.The results show that the sensitivity parameter category caused by the changes of carbon flux?GPP,NPP and R_h?is very concentrated and basically similar.The influential parameters are?_C33 and?_a,and the parameters'function is relatively independent,and woodland and grassland have no significant difference for the selected parameter sensitivity;The sensitivity parameters that cause the change of water flux?ET and Runoff?are mainly concentrated in the photosynthetic and water balance module,and the parameters which has a larger influence include:?_C3,g_m,?_m,?_max,c3,?and so on.The total sensitivity index of some parameters(such as?,?_a,?_max,c3)is higher than the first-order sensitivity index,indicating that they mainly influence the annual average of ET and Runoff mainly through interacting with other parameters.?3?According to the sensitivity analysis results,it can be known that the selected carbon and water sensitivity parameters are similar.Based on the sensitivity index,ten parameters to be optimized are selected as follows:?,?_a,?_max,c3,?_C3,a_C3,g_m,?_m,k_allom3,k_mort1,f_air.Based on the situation that GPP and ET are treated as the constraint optimization parameters simultaneously,and the average error accumulated between the observation values obtained from flux observation and the model's simulation values are set as the objective function,and then with the aid of simulated annealing method,the searching is conducted within the scope of 10 feasible regions to seek the parameterization scheme with a minimum objective function.The results show that,after the parameters are optimized,the simulation performance of the model is improved obviously.In view of GPP,the overvaluation of the peak value of growth season in CN-Dan site has been greatly improved.After parameter optimization,the correlation of GPP simulated values of all sites increased by 0.24 and RMSD decreased by nearly 40%.As far as ET is concerned,the overall improvement effect of all sites is promoted,and the correlation increased by 0.19,and RMSD increased by about 26%.All the above results show that the parameterization scheme which adopts GPP and ET restriction and optimization is more suitable for LPJ model to simulate GPP and ET of the sites in China.?4?In order to further optimize the model's state variables,EnKF is adopted to assimilate GLASS LAI product into LPJ-DGVM whose parameters have been optimized,and the vorticity flux site is adopted to verify and analyze the assimilation results.The results show that GLASS LAI assimilation into LPJ model can effectively improve the simulation precision of GPP,and some sites?e.g.,CN-Qia and CN-Din?after assimilation present a higher consistency in terms of sequential variation trend,numerical size and observation value.Compared with the simulation results after optimization of parameters,the overall accuracy increased by 0.04 and RMSD decreased by 2.19 gC m^-22 month^-1.When comparing the evaporation ET results after assimilation with ET simulation results which have a parameter optimization,it can be seen that some sites are slightly improved in the growing season,and the overall correlation R² only increased by 0.01,and RMSD reduced by 0.17 gC m^-2 month^-1.?5?Based on the above results,LPJ-DGVM assimilation model which adopts parameters optimization simulates GPP and evaporation ET of the total primary productivity of vegetation in China,and analyzes the seasonal and annual dynamic change of GPP and ET in the study area,and then compares them with the existing research results.The results show that the spatial distribution difference of GPP and ET is small between 2000 and 2015,and both show a trend of decreasing from southeast to northwest.Between 2000 and 2015,the fluctuation range of GPP in China is 6.02PgC yr^-1-6.76 PgC yr^-1,with an average value of 6.33 PgC yr^-1,and there is a rising trend during the 16 years.The fluctuation range of ET is 472.48mm yr^-1,-509.77mm yr^-1,,with an average of 485.98 mm yr^-1,,and the overall variation trend of the 16 years is not significant.What's more,GPP and ET in China show obvious seasonal spatial distribution differences,with the lowest in summer.This paper mainly explored the LPJ-DGVM physical and chemical parameters and remote sensing-model coupling optimization method,which has been clear about the effect of different parameters on different types of sites in China,and deepened the data assimilation in the application of remote sensing information coupled with process model.This research has great influence on further improving the accuracy of LPJ-DGVM carbon and water flux simulation,and regional extension studies.