Assessing NEE and Carbon Dynamics among 5 European Forest types: Development and Validation of a new Phenology and Soil Carbon routines within the process oriented 3D-CMCC-Forest-Ecosystem Model

The two main processes involved in forest ecosystems carbon balance are photosynthesis (GPP) and respiration. Ecosystem respiration (Reco) is determined by heterotrophic and autotrophic respiration, the former driven by microbial decomposition of soil organic matter (SOM), the latter by growth and maintenance of plant tissues. By differencing photosynthesis and respiration we have an estimate of the global carbon budget of a forest ecosystem, namely the Net Ecosystem Exchange (NEE). In this purpose, the aim of this work was to implement the 3D-CMCC-Forest Ecosystem Model (6.1.) to better estimate GPP and assess the C cycle in European forests. We included a new soil Carbon dynamics routine and several modifications in phenology, respiration and littering. Bud burst phenology has been improved with a new "Nonstructural Carbon injection function" representing the quantity of Carbon daily destined to new leaves and fine roots development. Fall phenology has been improved with a novel semi empirical logistic function to simulate leaf falling. Evergreen leaves turnover has been completely redesigned following an intra-crown competition logic. Soil carbon dynamics through the Residues, Microbial and Humus pools have been developed following a zero order kinetics equation, representing microbial decomposition activity. Autotrophic respiration has been implemented with a soil water potential factor to represent stomata closure when drought occurs. A new canopy vertical structure initialization rationale has been developed using the Perfect Plasticity Approximation algorithm; unfortunately it could be tested only on sites where dendrometric data were available. 3D-CMCC-FEM 6.1.v was validated against 6 EddyCovariance CarboEurope towers, representing 5 of the most diffuse forest ecosystems in Europe. The sites have been chosen to represent a climatic and longitudinal transect trough the European continent, so that the model could be tested on different critical boundary conditions. The GPP, Reco and NEE fluxes were validated for about 10 years at each site. To evaluate the model efficiency we tested daily and monthly correlation, Nash- Sutcliffe Model Efficiency, Goodness of Fit to a mono parametric linear regression. The model's plasticity and ability in representing observed anomalies was determined by analyzing inter annual, month and seasonal variability following published methods. We then statistically inferred the relationships between expected and observed frequency distributions of the anomalies. The results were quite encouraging; GPP r 2 was averagely 0.74 (daily) and 0.89 (monthly), the RMSE of about 1gC m -2 d -1 , the NSE greater than 0.7. Anomalies results were very good too; the NRMSE was averagely of 1.2 gC m -2 d -1 and their distribution were always significantly consistent with the observed ones. Reco r 2 was averagely 0.59 (daily) and 0.69 (monthly), the RMSE of about 0.83 gC m -2 d -1 , the NSE greater than 0.54 (daily) and .75 (monthly). Anomalies results were very good too; the NRMSE was averagely of 1.2 gC m -2 d -1 , their sign was captured for about 70% of the times and their distribution were always significantly consistent with the observed ones. The propagation of uncertainties resulted in NEE r 2 averagely of 0.56 6(daily) and 0.89 (monthly) (0.66 and 0.82 excluding the Mediterranean forest), the RMSE of about 1.5 gC m -2 d -1 , the NSE greater than 0.51 excluding the negative value of the Q. ilex stand. Anomalies results were acceptable and in line with the other PBMs in literature. Even though the NRMSE was averagely of 1.3 gC m -2 d -1 the frequency distribution of the anomalies distribution coincided with the observed ones just for half the sites. The model showed interesting improvements from the 5.1. version (in prep.), even more from the published 4.0 version. The model showed its weakness in representing the Mediterranean Forests, probably because of the over simplistic way to represent soil water dynamics and stresses. The use of the water potential RA liming factor apparently confirmed this hypothesis, since Reco was significantly improved and gave even better results than GPP after its implementation. In conclusion this work positively achieved its objectives. The model now reliably estimate all the components of the C cycle for the main European forest ecosystems. The new functions resulted in better GPP and RA estimation, finally allowed the model to simulate RH, Reco and NEE, and introduced new ideas to the forest modeling international panorama. The 6.1. version thus has wider perspectives and applicability and may be taken into account for several different applications; from predicting the net C cycle on regional scale, to assisting future forest management on finer scales up to 1 hectare.