Benchmarking The Ecosystem Respiration Simulated By CLM4.5

Ecosystem respiration,after gross primary productivity,is the second largest carbon flux between terrestrial ecosystems and the atmosphere.However,it is particularly RE that is difficult to constrain at the global scale due to limited understanding of complex interactions of physical,chemical,and biological processes and resulting high spatial-temporal dynamics,thus affecting the accurate assessment of carbon budget of global terrestrial ecosystems.Model simulation is a common method for estimating RE changes in terrestrial ecosystems.At present,the accuracy of RE simulation in terrestrial ecosystem process models is still insufficiently understood and remains associated with large uncertainties.In this study,FLUXNET and ChinaFLUX were benchmarked for the results of community land model version 4.5（CLM4.5）.The results of regional and site-scale RE simulation were validated and analyzed in terms of spatial and temporal scale.The main purpose was to explore the causes of RE simulation errors in different scales.Furthermore,according to the main causes of the simulation error of RE,we improved the simulation of RE from the point of view of CLM4.5 parameters.（1）Compared with the annual RE from 47 stations of FLUXNET,it has a better correlation with the simulation of RE in spatial pattern（R=0.61,P<0.01）.Except for the overall underestimation of RE in grassland simulation（the relative error was-18.89%）,the other plant functional types were overestimated in varying degrees.CLM4.5underestimated RE at high latitude sites while overestimated it at low latitude stations.（2）Compared with CLM 4.5-RE spatial pattern simulation results,Its Inter-annual simulation correlation is relatively low.In terms of different ecosystem types,the high correlation and the low deviation of simulated RE in grassland ecosystem maked the overall simulation effect of RE better than that of forest ecosystem.The results of CLM4.5-RE were similar to those of spatial pattern simulation.that is,the Simulation RE of each site in grassland ecosystem was underestimated,and the different plant functional types in forest ecosystem were overestimated in varying degrees.（3）CLM4.5-RE simulated the overall trend of seasonal deviation was similar to inter-annual variation.From the perspective of different plant functional types,the simulated RE of evergreen forest（evergreen coniferous forest and evergreen broad-leaved forest）was overestimated throughout the year,while the other plant functional types overestimated and underestimated in different degrees in growing season（April to September）and non-growing season（October to March）,especially in grassland sites during the growing season.In order to evaluate the comprehensive simulation effect of seasonal variation of all plant functional types,we ranked the simulation results of each plant functional type according to the indexes（R,RMSE,MBE）.The simulation results of different plant functional types from the best to worst were ranked as:mixed forest,evergreen coniferous forest,grassland,farmland,deciduous broad-leaved forest and evergreen broad-leaved forest.（4）The CLM4.5 captured the seasonal variation of RE better in Changbai Mountain,Dinghu Mountain and Qianyanzhou Mountain,compared to the simulation results of the forest ecosystem in all plant functional types at regional scale.The simulation results of RE in growth season and non-growth season at different sites were the same as those at regional scale.In other word,the annual RE of evergreen forests represented by Dinghu Mountain and Qianyanzhou was overestimated,and that of deciduous broad-leaved forests and mixed forests represented by Changbai Mountain was overestimated and underestimated in varying degrees.Overall,the underestimation of growth season was more seriously than that of growth season.（5）The correlation of seasonal variation of RE at Haibei and Inner Mongolia stations was better than that of regional scale.The seasonal variation of RE simulated at site scale was similar to that of regional scale.In other words,there are different degrees of overestimation and underestimation in spatial and inter-annual and seasonal pattern.CLM4.5 overestimated the seasonal variation of RE at Haibei as a whole.Inner Mongolia Station showed that the simulated RE was overestimated in non-growing season and underestimated in growing season.（6）The parameter sensitivity analysis of Changbai Mountain and Haibei showed that Q_10-RS was the key parameter which was affecting the soil respiration and RE in forest and grassland stations.Based on this,we classified the factors which were affecting Q_10-RS0-RS according to Literature and especially the optimal Q_10-RS parameters were given for different plant functional types.The optimal Q_10-RS for evergreen coniferous forest was 1.99,evergreen broad-leaved forest was 2.17,deciduous broad-leaved forest was 2.6,mixed forest was 2.79,and grassland was 1.75.（7）Based on the results of RE simulation,we can prove that the optimal parameter set from literature is more suitable for CLM4.5 ecosystem respiration simulation.The results showed that the R value of grassland ecosystem increased significantly,the overestimation in seasonal variation of RE in forest stations and the overall underestimation in grassland stations was improved,and the RMSE and MBE of each sites in forest ecosystem and grassland ecosystem was decreased finally.