A11F-01 INVITED 08:00h
A Global Carbon Cycle Data Assimilation System (CCDAS) to Infer Atmosphere-Biosphere CO2 Exchanges and Their Uncertainties
Atmospheric inversion studies have become an important tool for identifying sources and sinks of CO2 at the interannual time scale. For determining detailed patterns they suffer from the inverse problem being seriously under-constrained. Such methods are usually contrasted with process-based models of the terrestrial or oceanic carbon cycle. These models, however, cannot take into account information gained from CO2 measurements such as the extensive flask-sampling network. Here, we present results of a two-stage assimilation study of satellite radiances (identifying vegetation activity) and atmospheric CO2 concentration data using the terrestrial biosphere model BETHY. The controlling parameters for the second stage in this model are inferred by nonlinear optimization based on the model's adjoint. Uncertainties in these parameters are calculated from observational and model uncertainties via the model's Hessian and then mapped forward on predicted quantities such as net CO2 fluxes to the atmosphere via the model's Jacobian. The adjoint, Hessian, and Jacobian are generated by automatic differentiation of the model's source code. The model is able to fit the observations moderately well on a seasonal time scale and very well on an interannual time scale. It appears that the requirement to fit both the seasonal and interannual dynamics in the CO2 record is a strong constraint on model formulation. We will report on progress in model development and also on new experiments such as including ocean basis function in the optimization procedure.
A11F-02 08:15h
Parameter Optimization Using the Adjoint of a Biospheric Model
Regional budgets are best tackled by a combination of top-down and bottom-up estimation. In such approaches, process-based models are simultaneously constrained by local and integrated observations. This paper presents the local part of such an approach. We use an adjoint model generated from a biospheric model. The adjoint model, once generated, is a powerful tool in data assimilation. Our intention is to couple it to the adjoint of an atmospheric model in the next step to provide an interactive land-surface flux at much finer resolution. Running the adjoint of such a coupled model can solve the inversion problem efficiently. On top of the improved flux estimates and uncertainties, it would also optimize the parameters in the biospheric model for individual regions/biomes to shine light on the underlying processes. Here, we present the results from the adjoint of the CSIRO Biospheric Model. Using the eddy covariance measurements from Tumbarumba (an evergreen forest of eucalyptus trees) and Harvard Forest (deciduous), the modelled latent heat, sensible heat and carbon fluxes are compared to observations. The parameters in the biospheric model are optimized via the gradient descent method guided by a cost function based on the least squares of the misfits. The optimized parameters for both annual and seasonal datasets already show large discrepancies in the prescribed leaf area index which has been a product from remote sensing instruments. There is also a general insensitivity to parameters for soil and plant respirations; however, they are best optimized by the CO2 concentration data which will be introduced in the next step when coupled to an atmospheric model.
A11F-03 08:30h
A Regional Scale Coupled Atmosphere-Ecosystem Model: Formulation and Results
The formulation of self-consistent and computationally efficient atmosphere-ecosystem models requires the bridging of a wide range of spatial and temporal scales. Disturbance events such wind-throw, fire and land-use change give rise to significant sub-grid scale heterogeneity in ecosystem structure and function at a variety scales ranging down to the the size of an individual canopy tree, far below the resolution of both climate and numerical weather prediction models. Moreover, over decadal timescales, the spatial distribution of this heterogeneity is dynamic due to the successional dynamics that follow disturbance events within ecosystems. To address this problem, we have developed the Ecosystem Demography Land Surface Model (ED-LSM), an integrated biosphere model that incorporates plant community dynamics, soil carbon and nitrogen biogeochemistry and land surface biophysics. The fast timescale fluxes of carbon, water and energy between the ecosystem and the atmosphere are captured using the leaf photosynthesis and soil decomposition modules of Ecosystem Demography (ED) model coupled to a multi-leaf layer, multi-soil layer implementation of the LEAF-2 biophysical scheme. Long term changes in the biophsyical, ecological and biogeochemical structure of the ecosystem are captured using the ED model's system of size- and age-structured partial differential equations that track the changes in the vertical and horizontal heterogeneity of above and below ground ecosystem structure that result from ecosystem responses to the atmosphere that play out over years, decades and centuries. The model can be run both off-line and coupled to the Regional Atmospheric Modeling System (RAMS), which simulates both atmospheric dynamics and tracer transport of carbon dioxide. We have carried out coupled simulations of the model in temperate, tropical and boreal regions. Comparison of our results with observations from eddy-flux towers and meteorological stations highlights the models ability to capture the influence of the heterogeneous land surface on the dynamics of the land-surface interaction in these different regions on time scales ranging from the synoptic to the decadal.
A11F-04 08:45h
Estimations Of Regional CO2 Fluxes - Development Of A Modeling Framework Designed For The Ring Of Towers
Whereas inverse modeling of the sources of atmospheric trace gases on global scale with low spatial and temporal resolution is an established technique, there are many fewer applications to smaller scales. Additional challenges are created for regional modelers because they must deal with limited domains. The inflow fluxes across lateral boundaries are usually several orders of magnitude higher than the surface fluxes from the regional domains. Another difficulty appears when the trace gas emission has a strong diurnal cycle that cannot be neglected at these scales, as in the case of CO2. We have been developing a set of techniques to combine the use of numerical models with regional CO2 measurements. The regional inversion framework is built around CSU RAMS (Regional Atmospheric Modeling System) and the Lagrangian Particle Dispersion (LPD) model. The LPD model is used in adjoint mode to trace particles backward in time to derive influence functions for each concentration sample. The influence function provides information on potential contributions both from surface sources and inflow fluxes that make their way through the modeling domain boundaries into the CO2 concentration sample. Then the Bayesian inversion technique is applied in an attempt to estimate unknown surface emissions. CO2 flux is treated as a sum of respiration flux and assimilation (uptake by vegetation). Additional constrains are formulated for these fluxes using information from RAMS output (shortwave radiation, soil temperature, vegetation type) traced by Lagrangian particles. The modeling framework is being applied to estimate CO2 fluxes within 500 km radius from WLEF TV tower in northern Wisconsin instrumented with continuous measurements of CO2 concentration at 6 levels from 11 to 396m. Additional CO2 measurements include five 76 m communication towers operating during summer season of 2004. These towers form a ring around the WLEF tower with 100-150 km radius. The preliminary tests of the modeling framework were performed with the aid of model generated concentration pseudo-data for August 2000. Different configurations of source areas and different assumptions concerning expected model-data mismatch error were investigated. The results for CO2 flux estimation using concentration data form the ring of towers are very promising as long as the inflow CO2 flux is known or if its good a-priori estimation is available. For this purpose we are going to link our regional inversion system to a global transport model based on PCTM (Parameterized Chemical Transport Model) driven by CO2 fluxes provided by SiB3 (Simple Biosphere Model). Further inversion experiments using pseudo-data are being performed in parallel with a preparation for inversions using real CO2 observations for the summer of 2004.
A11F-05 09:00h
Estimation of the regional sources and sinks of CO2 with a focus on Europe using both regional and global atmospheric models
With a specific focus on Europe as part of the AEROCARB/CARBO-EUROPE project, we use atmospheric models to assess the monthly sources and sinks of atmospheric CO2 for the year 1998. A Bayesian inverse approach is used (top-down) where the spread between modelled and measured CO2 concentrations in the atmosphere is minimised. We divide Europe into as much as 26 regions to address the issue of aggregation error and determine what degree of refinement we can reach using this inverse method and the latest network of CO2 measuring stations over Europe. One of the specificity of this work is to use both mesoscale and global models with the same protocol to do these inversions. We evaluate the spread in the results arising when using these different models. The west (sink) to east (source) gradient on the fluxes found over Europe is analysed. Several sensitivity studies are performed and classified according to the importance of their effect on the European fluxes. Specificities attached to the use of mesoscale models are disccussed.
A11F-06 09:15h
Statistical Diagnostics of CO$_{2}$ Inversions
Over the last decade, Bayesian synthesis inversion has become the most common technique for interpreting global-scale spatial distributions of CO${}_{2}$. While the technique is formally based on statistical estimation, in few of the studies has there been any testing of the statistical model. The development of techniques has been a sequence of {\it ad hoc} adjustments in the light of problems experienced. This presentation describes a series of tests that revisit earlier calculations and investigate the extent to which problems could have been avoided if more comprehensive statistical testing had been adopted. These statistical tests can also be used in conjunction with current inversion studies, to evaluate whether results violate the assumptions inherent in the statistical model implemented. Problems such as biased priors, unrealistic covariance parameters, and residuals not following assumed distributions can be diagnosed. The test case uses 12 ocean regions, 4 regions of deforestation, and 8 regions each of CO${}_{2}$ fertilisation and seasonal CO${}_{2}$ exchange. Fossil CO${}_{2}$ and oxidation of CO are treated separately. In the standard case, the distribution of normalised residuals is found to be consistent with the Gaussian distribution assumed in the statistical model. The distribution of normalised deviations of flux estimates from priors is found to have a smaller spread than expected from the notional statistical model. This is interpreted as a reflection of the common practice of using weak (minimally-informative) priors, informally regarding them as a regularization constraint rather than an equivalent source of information.
http://ms.unimelb.edu.au/~enting/invstats.html
A11F-07 09:30h
Characterizing Model Errors for Inverse Modelling of Atmospheric Trace Gases
Inverse modelling has become a powerful tool for improving estimates of surface fluxes of environmentally important trace gases. However, the a posteriori flux estimates depend critically on properly characterizing forward model errors, which are typically specified on an ad hoc basis. We present a new approach for quantifying model error for the inverse modelling of CO based on the "NMC method," which has not previously been applied to the inverse modelling of atmospheric trace constituents. The model error for the GEOS-CHEM simulation of CO is estimated using the differences between successive chemical forecasts of CO (48-hours vs. 24-hours), generated during February-March, 2001. We examine the dependence of the error correlation structure on region and local meteorology. We determine the consistency of the error statistics from the NMC method with those calculated by comparison of the CO simulation from GEOS-CHEM with that from the UCI/FRSGC model, and those based on the differences between GEOS-CHEM and observations of CO from MOPITT.
A11F-08 09:45h
Atmospheric Tracer Inverse Modeling Using Markov Chain Monte Carlo (MCMC)
In recent years, there has been an increasing emphasis on the use of Bayesian statistical estimation techniques to characterize the temporal and spatial variability of atmospheric trace gas sources and sinks. The applications have been varied in terms of the particular species of interest, as well as in terms of the spatial and temporal resolution of the estimated fluxes. However, one common characteristic has been the use of relatively simple statistical models for describing the measurement and chemical transport model error statistics and prior source statistics. For example, multivariate normal probability distribution functions (pdfs) are commonly used to model these quantities and inverse source estimates are derived for fixed values of pdf paramaters. While the advantage of this approach is that closed form analytical solutions for the a posteriori pdfs of interest are available, it is worth exploring Bayesian analysis approaches which allow for a more general treatment of error and prior source statistics. Here, we present an application of the Markov Chain Monte Carlo (MCMC) methodology to an atmospheric tracer inversion problem to demonstrate how more gereral statistical models for errors can be incorporated into the analysis in a relatively straightforward manner. The MCMC approach to Bayesian analysis, which has found wide application in a variety of fields, is a statistical simulation approach that involves computing moments of interest of the a posteriori pdf by efficiently sampling this pdf. The specific inverse problem that we focus on is the annual mean $CO_2$ source/sink estimation problem considered by the TransCom3 project. TransCom3 was a collaborative effort involving various modeling groups and followed a common modeling and analysis protocoal. As such, this problem provides a convenient case study to demonstrate the applicability of the MCMC methodology to atmospheric tracer source/sink estimation problems.