Quantifying uncertainty in projections of continental fluxes of carbon and energy using the NEON platform

Background/Question/Methods

As we enter a new “data rich” period for ecological science new possibilities become available which allow for better forecasting of many ecosystem processes. We are utilizing data from the continental-scale NEON platform and other monitoring networks (FLUXNET, ICOS, LTER etc.), in conjunction with ever-increasing computing power, land surface model sophistication and new statistical and optimization methodologies to address a pressing societal need for improved quantification and reduction of uncertainty of projections of carbon and energy fluxes across the US.

We have developed a data assimilation system coupling the Community Land Model (CLM) with the Data Assimilation Research Testbed (DART), an advanced facility for ensemble data assimilation (DA). Using this new tool we are able to constrain the model with data to give predictions that best approximate the observations. Using this approach, in theory we able to account for multiple sources of uncertainty in observational data, model structure and parameters, boundary conditions and the statistical methods used in the DA in a rigorous way. 

Results/Conclusions

We have investigated if we can use CLM-DART with synthetic and real data (and their associated uncertainties) about the past and current states of an ecosystem to improve our projections of carbon and energy fluxes. We have used the DA system to (i) provide estimates of model states; (ii) quantify uncertainties in these model states; (iii) propagate and quantify uncertainties in projections; and (iv) test how these resultant probability density functions can be used to interpret model structural validity. We show how it remains very challenging to correctly specify all sources of uncertainty and how this propagates into uncertainty in projections. In particular, the proper specification of model uncertainty in a complex land surface model, such as CLM, remains problematic.

As well as being a fundamental tool NEON will use in operations to produce foundational high-level data products, we believe this DA system is an example of how ecological theory, process models and observational networks can be combined to produce meaningful ecological information that can be used to forecast changes in ecosystems, in educating teachers and students and supporting decision making.