Parameter Estimation Of Terrestrial Ecosystem Process Model And Its Application In Carbon And Water Fluxes Simulation

It is of great significance to accurately understand the changes of carbon and water fluxes in terrestrial ecosystems and deeply understand the process of carbon and water cycle and its coupling effect,which can promote the sustainable utilization of carbon and water resources.Flux observation and model simulation are two common methods to study carbon and water fluxes.Just using flux observation cannot reflect the internal process and control mechanism of the ecosystem,and it is difficult to quantitatively express the spatiotemporal evolution characteristics of carbon and water fluxes.Model simulation based on physiological and ecological processes is an effective method to quantitatively express the spatiotemporal characteristics of carbon and water fluxes and analyze its internal mechanism.The biogeochemical model represented by CEVSA have relatively mature mechanical and simulation capabilities,which have been widely used in the field of terrestrial ecosystem carbon and water research.However,the equations used to describe the structure of the CEVSA model contain a large number of parameters related to the physiological and ecological information process of vegetation,and most of the parameters are derived from experimental observations or expert experience.So,it results in low simulation accuracy of the model in practical application and limits the application of the model.It is necessary to combine the observation data to optimize the key parameters of the model,so that the model can have higher accuracy and practical application.And it deepens people's further understanding of the model structure and lays a foundation for the parameterized scheme of CEVSA model.In this paper,we used the One-At-A-Time?OAT?parameter sensitivity analysis method to analyze the sensitivity parameters of even typical vegetation types based on CEVSA model,namely Qianyanzhou evergreen coniferous forest,Dinghushan evergreen broad-leaved forest,Changbaishan mixed forest,Inner Mongolia grassland,Yucheng farmland,Haibei shrubland and Dangxiong meadow.Secondly,according to three different optimization experiments,we used the Differential Evolution Markov Chain?DEMC?optimization algorithm to optimize the 10-12 key parameters,which were screened out under seven vegetation types based on the existing observation data.Finally,we used Coefficient of Determination?R²?,Root Mean Square Error?RMSE?and Nash-Sutcliffe?NSE?to comprehensively evaluate the simulation performance of the model after parameter optimization.The main conclusions are as follows:?1?The sequence of seven typical vegetation sensitivity parameters is basically the same.The output of the carbon and water fluxes by the model is the most sensitive to the four photosynthesis parameters:namely nitrogen absorption parameters of vegetation(N_s,N_m,N_c1)and the intermediate calculation parameter of vegetation nitrogen absorption(d_enom).In this paper,we used 2.5%as the sensitivity discrimination index.The sensitivity of the four parameters were as high as 74.81%to152.88%,which were much higher than the sensitivity index of the others.And it indicated that these four photosynthesis parameters were the most critical parameters affecting the simulation of carbon and water fluxes in CEVSA model.Secondly,the sequence of the sensitivity of the seven kinds of typical vegetation under the NEP and the Gross Primary Productivity?GPP?and the Ecosystem Respiration?RE?were relatively similar,which are certain differences in numerical.We found that key parameters on the impact of the NEP mainly come from the change of the GPP or mainly come from the change of the RE?NEP=GPP-RE?.So we can't just rely on the parameter sensitivity analysis result to decide the change relationship between the carbon flux.In addition,the photosynthesis parameters affect the change of NEP and ET at the same time,and the sequence of which are basic same.It's indicating that the stomatal behavior of plants in the CEVSA model couple the photosynthesis and the transpiration.?2?The 10-12 key parameters selected under seven typical vegetation types were optimized,and there were 3-8 parameters that could be constrained by NEP observation data.The four most sensitive photosynthetic parameters,vegetation nitrogen absorption(N_s,N_c1,N_m)and the intermediate calculation parameters of vegetation nitrogen absorption(d_enom),have the largest changes in the posterior median relative to the original model values under different vegetation types.Therefore,we should pay more attention to the optimization of these parameters in the promotion and application of the model.Moreover,the parameter,decomposition of soil moisture influence factor coefficient(f_moi)was lower than the original model value and the posterior 95%confidence interval was more concentrated under all vegetation types.Indicating that the value of this parameter in the original model may be high in the soil moisture balance expression module.In addition,under the three typical forests,the number of parameters that can be effectively estimated by experiment 3?NEP&ET?constraint is the most,the number of parameters estimated by the experiment 1?NEP?constraint is the second,and the number of parameters estimated by the experiment 2?ET?constraint is the least.Moreover,comparing with the experiment 1 and the experiment 2,the constraint scheme of the experiment 3 has the narrowest posterior 95%confidence interval for most parameters.Indicating that using the NEP and ET observation data constrain model can effectively reduce the uncertainty of parameter estimation.Besides,specific leaf area?SLA?,nitrogen absorption parameters of vegetation(N_s,N_c1,N_m,N_c2)at the three forest sites have the highest variability,leading to the highest uncertainty compared with other parameters.But the stomatal conductance response parameter(g_s1),sapwood maintain respiration(S_re)and Rubisco response to CO₂ concentration??₂?have the minimal variability.Comparing to other parameters,the uncertainty is minimal.?3?Under the seven typical vegetation,using the experiment 1 constraint model,the optimal parameter set has a good consistency with the NEP observation value,but the simulated water flux ET deviates from the observation value.Under the three typical forest sites,the optimal parameter set obtained by the experiment 2 has a good consistency with the ET observation value,but the simulated carbon flux NEP has a big error with the observation value;the optimal parameter set obtained by the experiment 3 is the best to simulate the carbon and water fluxes.?4?Whether in the adjust year?2003-2007?or the validation year?2008-2010?,the simulation capability of the model for carbon and water fluxes after parameter optimization is much higher than that before parameter optimization.Indicating that the optimized parameter set can better simulate the carbon and water exchange characteristics of three typical forest sites.After parameter optimization,RMSE_NEP and RMSE_ET were reduced by 39%-45%and 15%-18%respectively,which significantly improved the deviation between simulated and observed values of carbon and water fluxes.Besides,all NSE is greater than zero,indicating that CEVSA model is more accurate and reliable in simulating carbon and water fluxes after parameter optimization.For the simulation of ET in Qianyanzhou,the simulation ability after the optimized model decreased slightly.Possibly,among of the twelve selected key parameters,except for the nine photosynthesis parameters,two autotrophic respiration parameters and one parameter affecting the soil carbon decomposition were also included.The optimized parameter set improve the simulation of NEP obviously.But some poorly constrained respiration parameters make the simulation of ET has certain deviation.