Uncertainty Analysis Of Simulated Terrestrial Ecosystem Carbon Storage From Additional Phosphorus Process With A Model-data Fusion Approach

Phosphorus(P)is not only an important chemical element of living organisms,but also is a key factor in affecting terrestrial ecosystem structure and function.To improve the accuracy of ecosystem models,some studies have integrated the P processes into terrestrial ecosystem models.However,the model uncertainties caused by the structure and data sets of P cycle were still unclear.In this study,we used a model-data fusion approach to quantify the uncertainties of parameters estimation and simulated carbon(C)storages from additional P process(including structure and data sets).Based on a terrestrial ecosystem coupling model(TECO-CNP),we set up three experiments: CNCN(TECO-CN model with C and nitrogen(N)observation);CNP-CN(TECO-CNP model with C and N observation);CNP-CNP(TECO-CNP model with C,N and P observation)The main results were as follows:(1)For the parameter estimation,13,13 and 20 parameters were well constrained in CN-CN,CNP-CN and CNP-CNP experiments,respectively.The results showed that the structure of P cycle introduced more unidentifiable parameters,while additional P observation provided more information for constraining parameters associated with both P and C cycles.(2)By comparing the simulation and observation,we found that the deviation of CNP-CNP experiment was lower than that of the other two experiments.This indicated that adding the P process could reduce the model uncertainty.However,the soil organic C,soil total N and litter P remained largely deviation.(3)The relative information index was used to quantify uncertainty of simulated C storages caused by the P process.Our results showed that,in both short-term simulation and long-term prediction,the relative information index of P cycle structure in most C pools was negative,which would increase the uncertainty.However,the relative information index of P observation in most C pools was positive,which would reduce the uncertainty.In general,except for microbes and passive SOM,the relative information index of P process in other carbon pools was positive,indicating that P process could reduce the uncertainty of model simulation in these C pools.(4)The relative information gain was used to quantify the effect of P process on posterior simulated C storages.Our study showed that,in both short-term simulation and long-term prediction,the P process led to changes in the posterior C storages.In most carbon pools,relative information gain of P cycle structure was larger than that of P observation,indicating that the P cycle structure was the main factor affecting the posterior C storages.In conclusion,the P process could provide additional information to model parameterization,thus constraining more parameters and improving the performance of the model.This information could also change posterior simulated C storages and reduce its uncertainties.Quantifying the uncertainty form additional P process could help us in understanding the terrestrial C-N-P interaction,improving the simulation of C cycle in response to climate change and providing theoretical basis to future observation.