The National Ecological Observatory Network (NEON) is a continental-scale facility that will collect ecological data, including eddy covariance flux observations, from 60 sites in the continental US, Alaska, Hawaii, and Puerto Rico over 30 years. We are vigorously quantifying uncertainty in all NEON data products and are investigating the propagation of uncertainty from basic observations when generating high-level data products, such as continental-scale, gridded maps of carbon and energy fluxes.
One approach we are using to integrate many observational data streams into high-level data products is to use model-data fusion. In this approach a process model is used to provide an analytical framework for data interpretation, synthesis, interpolation and extrapolation. In theory, uncertainty from many sources can be accounted for including: (i) observational data, due to incomplete and noisy observations and biases; (ii) process model structure; (iii) process model parameters; (iv) boundary conditions, including meteorological drivers; and (v) the statistical method used to combine data and process model.
In this study we investigate how realistic assessments of uncertainty in a range of ecosystem observations and boundary conditions interact to affect our ability to quantify and reduce uncertainty in ecological forecasts.
Results/Conclusions
We have developed a Data Assimilation System (DAS), which couples the Community Land Model (CLM) with the Data Assimilation Research Testbed (DART), an advanced system for ensemble data assimilation (DA). We have used this new tool to undertake a number of experiments using a suite of synthetic and real data for which we varied uncertainties around ranges similar those anticipated for observations from the NEON platform.
We show how the amount of uncertainty, the type of uncertainty and the way this is treated in the DAS can strongly affect the confidence intervals on the high-level data products that are generated. Without a proper treatment of this uncertainty in observational data the DAS often produces over-confident, and potentially misleading, results. We suggest that whilst the uncertainties associated with high-level data products might be large, this does not reduce their value. Rather, this proper quantification of uncertainty increases their usefulness, allowing investigators to fully exploit the information that these data contain, but also highlighting their limitations. It also provides insights into areas where we lack knowledge and should concentrate our investigations in ecological science.