W16: Model-data fusion in the studies of carbon-climate-human system

organised by Georgii Alexandrov


Title: Model-data fusion in the studies of terrestrial carbon sink

Authors: G. A. Alexandrov et al.

Abstract: Current uncertainty in quantifying the global carbon budget remains a major contributing source of uncertainty in reliably projecting future climate change. Furthermore, quantifying the global carbon budget and characterizing uncertainties have emerged as critical to a successful implementation of United National Framework Convention on Climate Change and its Kyoto Protocol. Beyond fundamental quantification, attribution of the processes responsible for the so-called Ôresidual terrestrial uptakeÕ is important to the carbon cycle communitiesÕ ability to simulated the future response of the terrestrial biosphere to climate change and intentional sequestration activities. This paperÕs objective is to describe the efforts of the workshop participants and their approaches to model-data fusion enabling continued advances in the solution of quantifying carbon cycling and the terrestrial mechanisms at work.

Title: On-Line Climate Model Simulation of The Global Carbon Cycle And Verification Using the in Situ Observation Data

Authors: Kazuo Mabuchi, Hideji Kida

Abstract: A new version of BAIM that is Biosphere-Atmosphere Interaction Model Ver.2 (BAIM2) was developed. BAIM2 can estimate not only the energy fluxes, but also the carbon dioxide flux between terrestrial ecosystems and the atmosphere. The photosynthesis processes for C3 and C4 plants are adopted in the model. The carbon storage of vegetation is divided into five components (leaves, trunk, root, litter, and soil), and the carbon exchanges among the components of vegetation and the atmosphere are estimated in each time step of the on-line model integration. The values of morphological parameters using in the model are derived from the carbon storage values of the components, and the phenological changes of vegetation are reproduced by the model. The BAIM2 was incorporated into a spectral general circulation model, and was connected on-line to the atmospheric model. This general circulation model had a triangular truncation at wave number 63 (T63) and had 21 vertical levels. The horizontal grid interval was 1.875? (192 ? 96 grids). The atmospheric prognostic variables were the temperature, specific humidity, divergence and vorticity of the wind, the carbon dioxide concentration in each atmospheric layer, and surface pressure. The vegetation type of each grid point was specified and the interactions between the land surface vegetation and the atmosphere were estimated by the BAIM2 at each grid point. The time step interval of the integration was about 20 minutes. Using this climate model, a control time integration was performed under the actual global vegetation condition. The control integration was continued for 30 years, and an evaluation of the results was performed. The amplitudes and characteristics of seasonal cycle of the carbon dioxide concentration simulated by the model were generally consistent with the observed data at the in situ stations. The results simulated by the model can be evaluated partially by the observation data. There is a necessity, however, of the further cooperation between the researchers working in the different disciplines. And also, we should consider the utilization of the remote sensing data still more for the verification of model results.

Title: State-Parameter Estimation of Ecosystem Models Using a Smoothed Ensemble Kalman Filter

Authors: Mingshi Chen

Abstract: Much of the efforts of simulation-based methods of ecosystem analyses have been focused on either improved methods for parameter estimate wherein state variable uncertainties were not explicitly taken into account or improved procedures for estimating time-varying state variables wherein the parameters were assumed to be known in advance. The main weakness of the approaches only focused on parameter estimate is that they attribute all errors from input, output and model structure to model parameter uncertainties. The approaches only applied to state variable estimate may reduce the prediction accuracy of models due to ignoring time variation of parameters. This research tries to find a procedure that can provide the simultaneous estimate of states and parameters through using a smoothed ensemble Kalman filter (SEnKF) to assimilate a carbon cycle model with observed fluxes of carbon (C), weather, hydrology, energy, and remote sensing data at three forest sites: Howland (Maine, USA), Boreas (Alberta, Canada) and Niwot Ridge Forest (Colorado, USA) from 2000 to 2004. Here the kernel smoothing of parameters employed on ensemble Kalman filter aims at overcoming the over-dispersion of parameter samples and loss of information between time points when the parameters are considered to be fixed. The analyses demonstrate that the model parameters, such as light use efficiency, respiration rates, minimum and optimum temperature and so on, are most highly constrained by eddy flux data at daily to seasonal time scales and especially light use efficiency takes on a strong seasonality. To validate that simultaneous parameter estimate can improve prediction of the models, we assimilate only each five- or ten-day eddy flux data into the model although data are almost daily available in the three tower sites. Results show that state variables including GPP, respiration and NEE in modified model much better match data at those time points when data are not intentionally assimilated than using the original model in which the parameters are fixed. In addition, the results also show that the SEnKF can reduce dramatically variance of the state variables stemming from uncertainties of parameters and driving variables. Therefore, SEnKF is a robust and effective algorithm in evaluating and developing ecosystem models and improving understanding and quantification of the C cycle parameters and processes.

Title: Getting global pattern of plant productivity through combining observations and a process model

Authors: Georgii Alexandrov

Abstract: This study is addressed to the problem of getting global pattern of plant productivity through combining observations and process models. It is based on the assumption that plant productivity at large is determined by climate and therefore its global pattern can be derived from the global pattern of temperature, precipitation, incoming solar radiation and Earth surface reflectance in some wave ranges. The general principle of model-data fusion is to find ‘optimal match’ between diverse observations and model. In this study we use a specific terrestrial biosphere model, known as TsuBiMo, and develop numerical methods for matching TsuBiMo, biometric measurements of plant productivity, such as Osnabrueck database of terrestrial Net Primary Production (NPP), and the measurements of Net Ecosystem Exchange (NEE) that coming from the global network of CO2 flux observations, known as FLUXNET. The fusion of TsuBiMo and NEE measurements gives temporal pattern of plant productivity, the relation of annual NPP to seasonal course of temperature and precipitation. The biometric measurements of NPP taken in many locations during 1982-2002 make it possible to characterize spatial pattern NPP, the relation of average annual NPP to the average seasonal course of temperature and precipitation. The ‘optimal match’ between the model and two different types of observations gives empirically justified spatio-temporal pattern of plant productivity.

Title: A receptor-oriented modeling approach to estimate regional carbon exchange in New England and Quebec by combining atmospheric, ground-based, and satellite data

Authors: Daniel Matross, Arlyn Andrews, Christoph Gerbig, Steven Wofsy, Pathmathevan Mahadevan

Abstract: We develop a data-driven diagnostic tool with a minimum number of parameters to estimate terrestrial carbon flux on a regional to continental (104 to 106 km2) scale. Our approach is a receptor-oriented modeling framework, consisting of a time-reversed Lagrangian adjoint model (STILT) [Gerbig et al. 2003a,b; Lin et al. 2003] coupled to a vegetation CO2 flux model that calculates gross primary production and respiration by partitioning photosynthetic and non-photosynthetic respiration, the Vegetation Photosynthesis and Respiration Model (VPRM) [Mahadevan et al., 2006]. To drive atmospheric transport, the adjoint transport model utilizes assimilated wind fields from RAMS, EDAS-40, or WRF. The biosphere model incorporates MODIS-derived enhanced vegetation index (EVI) and land surface water index (LSWI) together with GOES-derived shortwave radiation [Diak et al. 2004] to capture surface spatial heterogeneity and variations in soil moisture, canopy nutrition, solar input and phenology. To integrate functional dependence of CO2 flux, we fit parameters within VPRM using Ameriflux eddy covariance data. This coupled, fundamentally bottom-up biospheric model is successfully validated using tall tower CO2 concentration data, indicative of integrated regional-scale carbon exchange, from the new NOAA CMDL Argyle tall tower in central Maine. Results from the model show remarkable agreement with tall tower observations without any parameter adjustment. We produce representative regional flux estimates for the greater Maine and southern Quebec region in summer 2004 which are fully consistent with tower and aircraft observations of CO2 concentrations. We characterize potential sources of uncertainty including parameterization for vertical mixing within the planetary boundary layer and the accuracy of the modeled upstream boundary condition. Through a linear least-squares analysis, we evaluate the potential for this tower data to constrain parameters in more complex inversion schemes. This work demonstrates initial validation of a bottom-up terrestrial CO2 flux model using regionally representative, rather than just locally representative, atmospheric concentration data and produces a reasonable a priori condition for future inversion studies.

Title: Development of an ecosystem model using observational data for making semi-real-time prediction of forest fires

Authors: Akihiko Ito

Abstract: In the JST-SORST project, we are developing a semi-real-time prediction system of forest fires in Asian region, using a forest biomass-burning model, satellite-based active fire map, and model-based weather forecast data. Active fires in Asia are detected from the MODIS data within half day from actual satellite observation, and 96-hour weather forecast data are obtained from Japan Meteorological Agency every day. Then, a coupled carbon cycle and fire regime model is driven by these data to make a prediction with respect to fire expansion and carbon emission by biomass burning during the next 72 hours. This data-assimilation system is under development by the IIS, University of Tokyo, and FRCGC-JAMSTEC, and is to be useful for disaster protection and carbon budgeting.

Title: The ratio of anthropogenic carbon emissions to net ecosystem carbon uptake: A hot issue for emission traders?

Authors: Frank Veroustraete, Willem Verstraeten

Abstract: Carbon emissions and terrestrial ecosystem net uptake fluxes are key variables to guide climate change decision makers and emission traders. A common method for the estimation of vegetation carbon fluxes is inventory-based (forests, agriculture). However, harvest and logging inventories are limited in time and space and prone to non-comparable methodology. Moreover, carbon inventories are limited to net primary productivity (NPP). Additionally, no information is available when applying inventory based methods, on the influence of water limitation nor are natural ecosystems included in classical carbon inventories. To develop carbon emission trading support tools, a sound application perspective is offered by expert systems based on earth observation (EO). They allow estimates of net carbon fixation using a minimum of meteorological data and overcome the methodological limitations of inventory based approaches. The core of this type of expert systems is a Monteith type production efficiency model (PEM). For example C-Fix, which is a known PEM, produces estimates of carbon mass fluxes e.g., gross primary productivity (GPP), NPP and net ecosystem productivity (NEP) at different spatial scales. Global carbon budget studies are, currently, still dominated by temperature driven approaches. Nevertheless, there is typically a strong coupling between the carbon and hydrological cycle. Hence, to take water limitation in carbon studies into account, water availability for vegetation must be estimated, preferably with EO. Applications of this type of EO based expert systems are manifold. Mapping of spatial and temporal patterns of GPP, NPP, NEP and soil respiration as well as mapping of water use efficiency in water stressed areas and ultimately the determination of the ratio of anthropogenic carbon emissions to net ecosystem carbon uptake at different spatial scales. Emission traders are hereby offered a tool to identify the areas on Earth where the demand for carbon uptake areas is the highest and moreover they can quantify the capacity of carbon uptake of a certain area as well. Two important issues indeed in view of a new type of trade, which is coming into being. Keywords: Kyoto protocol, emission trading, earth observation, expert system, soil respiration, water limitation.

Title: State-Parameter Estimation of Ecosystem Models Using a Smoothed Ensemble Kalman Filter

Authors: Mingshi Chen, Shuguang Liu, Larry Tieszen

Abstract: Much of the effort in modeling carbon dynamics has been on estimating optimal values for either model parameters or state variables. The main weakness of estimating parameter values alone (i.e., without considering state variables) is that all errors from input, output and model structure are attributed to model parameter uncertainties. On the other hand, the accuracy of estimating state variables may be reduced if the temporal evolution of parameter values is not incorporated. This research develops a smoothed ensemble Kalman filter (SEnKF) to simultaneously estimate system states and model parameters of a simple carbon cycle model. The approach is applied to assimilate observed fluxes of carbon (C) and major driving forces at three Ameriflux forest sites: Howland (Maine, USA), Boreas (Alberta, Canada) and Niwot Ridge Forest (Colorado, USA). The aim of applying a kernel smoothing algorithm on ensemble Kalman filter is to overcome the dramatic sudden changes of parameter values in time and loss of continuity between two consecutive points in time. Our analyses demonstrate that the model parameters, such as light use efficiency, respiration coefficients, minimum and optimum temperature for photosynthetic activities and so on, are highly constrained by eddy flux data at daily to seasonal time scales. The SEnKF stabilizes parameter values quickly regardless of initial values of the parameters. Potential ecosystem light use efficiency demonstrates a strong seasonality. Results show that the simultaneous parameter estimation procedure dramatically improves model predictions. In addition, the results also show that the SEnKF can dramatically reduce the variance of the state variables stemming from uncertainties of parameters and driving variables. SEnKF is a robust and effective algorithm in evaluating and developing ecosystem models and improving understanding and quantification of the C cycle parameters and processes.

Title: Using Regional Biospheric Model to Constrain CO2 Inversion

Authors: Douglas Chan, Misa Ishizawa, Kaz Higuchi

Abstract: The typical global Bayesian inversion of CO2 fluxes requires prior information in the form of the background CO2 concentration field generated by the global biospheric fluxes. We examined the sensitivity of the inversion results to different global biospheric fluxes. By comparing inversion results using global biospheric fluxes with and without synoptic variations, we estimated the sensitivity of inversion results to the additional synoptic biosphere atmosphere interaction. The land and ocean fluxes could change by 0.2 to 0.3 GtC/year. Therefore more accurate spatiotemporal biospheric fluxes may help constrain inversion estimates. We are exploring using regional biospheric models with more flux measurement validations (upscaling) to better constrain global/regional inversion results.

Title: Nonlinear Effects In Ecosystem Models Of Elements Dynamics At Local Level

Authors: Alexander Komarov

Abstract: Local level in the models of elements cycling in the ecosystems means a population approach to the modeling of the elements dynamics within the stand and respective soil parcel. Stand is considered as a community of tree populations located in common environment conditions. Complexity of the ecosystem processes and their interactions leads to nonlinearity of the models? structure and its mathematical description. Nonlinearity has properties which can hide the result of a model?s prognosis: 1. nonlinear systems demonstrate oscillatory or quasi-stochastic dynamics; 2. characteristics of a stationary state depends on the model?s structure and initial values, in multi-stable system the final state depends on initial values; 3. nonlinear system can be unstable in relation to the small changes of the parameters or initial values of leading variables; moreover, such a small change can move the stability state to another one etc. Thus, nonlinearity together with uncertainties in the system parameters and initial values proves to be an important problem at understanding of the model?s results. Can we find among ecosystem processes any that compensate nonlinearity effects and decrease the area of indistinctness? First, soil decomposing processes as a whole can be treated as a smoothing transformation of the leading variables at possible strong external impacts: draughts, insect outbreaks, significant weather oscillations etc. Second, some ecophysiological processes, which depends, for example, on tree age can decrease the dispersion of variables. Third, perhaps, the most important and most obscure, the structure of the interspecific and intraspecific interactions can lead to the relative stability and quasi equifinal results. Nonlinear effects can be demonstrated using the system of models EFIMOD (Chertov et al., 2003; Komarov et al., 2003), which is a simple stand simulator at population level linked with soil sub model and soil climate generator.

Title: Forest and soil dynamics at different silvicultural regimes and forest fires: simulation modelling

Authors: Alexey Mikhaylov, Andrey Martynkin, Alexander Komarov

Abstract: The forest simulation model of the stand/soil system uniting population and balance modelling approaches, EFIMOD, has been used for a long-term forest simulation of different silvicultural regimes and sporadic forest fires. It allows for simulation of the carbon and nitrogen pools dynamics in forest ecosystems in wide range of site and climatic conditions. Structure of the model system allows for a use of standard forest inventory data. A Rothermel’s surface fire model has been inserted into the EFIMOD. Output variables are the inventory stand data, pools of carbon and nitrogen in the stand and soil, the dynamics of CO2 emission and some other characteristics. It can be used for the account of carbon balance on local (population) and regional levels. The case study was conducted on a 300-hectare forest area with 108 stands in the experimental forest “Russky Les” 100 km south of Moscow, Russia. Four strategies of silvicultural regimes were simulated for a 200 year time span: natural development, selective forest harvest, authorised Russian forestry practices according to the forest laws, and unauthorised forest practices that have increased over the last decade in the country. The naturally developed forest has maximal increase of carbon in soil and forest. At selective cuttings carbon pools also increase, at authorized practice they are almost stable. Unauthorised practice scenario demonstrates gradual decrease of carbon. Forest fires simulations demonstrated that at peak emission CO2 at the fire was notably compensated by forest growth of young trees after this catastrophic event. But high frequency of forest fires leads to soil carbon decrease and consequently decreasing of carbon in the ecosystem. Frequent fires do not allow for the trees to reach productivity maximum. Crowning fire is more dangerous than ground fire because trees dying. Some comparative case-studies were analyzed in different climatic and site conditions.