Integration and Model Evaluation

INDEPENDENT OBSERVATIONS FOR INTEGRATED MODEL EVALUATION

As part of the NGEE Arctic goal to improve model representations of Arctic ecosystem processes and thereby enhance the fidelity of climate predictions, an important objective is to quantify model bias and uncertainty and document improvement in climate prediction. Accomplishing this objective will require a diverse array of independent observations that have not been used to parameterize or initialize models. These observations may be used to evaluate not only model predictions of multiple states and processes at multiple scales, but also the effectiveness of the scaling approach itself.

M. Torn (LBNL), a principal investigator (PI) of the ARM Carbon project that includes three eddy covariance systems, will lead this team and many of the tasks.

D. Billesbach (University of Nebraska, Lincoln), a biometeorologist and is the instrument mentor for two eddy covariance systems for the ARM Carbon project in the Southern Great Plains and who has long experience with laser-based instrumentation for methane concentrations.

This section specifies the primary independent data streams that we will generate and the evaluation methods that we will employ to assess model performance at fine, intermediate, and climate modeling scales. We will focus on predictions of climate forcing by the ecosystem, namely the ecosystem-atmosphere exchange of energy, greenhouse gases, and water. These fluxes are controlled by (or emerge from) the interaction among vegetation, hydrologic, and biogeochemical processes and thus integrate over model simulations of the three process areas of NGEE Arctic.

Integrated Model-Data Evaluation Goal: Quantify the integrated climate forcing from ecosystem greenhouse gas, energy, and water fluxes across a range of permafrost conditions and spatial scales and document improvement in model predictive skill of this forcing.

Although a number of excellent studies have been carried out in recent years that examine climate feedbacks related to either energy-albedo (e.g., Sturm et al. 2001b) or biogeochemistry (e.g., Schuur et al. 2008), few have included observations of both biogeochemistry and energy feedbacks. Thus, there have not been adequate datasets for testing simulations of the full suite of climate feedbacks—particularly across the range of scales targeted by NGEE Arctic. We will generate observations needed to construct the full surface energy budget and GHG budget of the ecosystem. We will evaluate the ability of the models to represent the integrated system response and the validity of using these models to evaluate future changes in energy, water, and biogeochemical influences on climate.

The influence of Arctic ecosystems on climate, and in particular on fluxes of GHGs, energy, and mass, will be evaluated at scales corresponding to our three modeling scales. It will not be possible to make direct observations of all quantities at all scales, but we have identified important, observable quantities at each scale. We will also make a suite of isotopic measurements that are diagnostic of model performance in simulating these integrated fluxes.

Independent Observations of Land-Atmosphere Exchange. We will collect observations that integrate land-atmosphere exchange processes at scales useful for testing the multiscale models. A few of these observations were described in the process research sections above. We propose to do intensive testing of scaling from fine to intermediate scale (10 cm to footprint of eddy flux tower), and evaluation with remote-sensing products at the climate scale. In each case, we aim to compare models with the native scale of observations for each quantity of concern. The tasks for this section are to measure net fluxes of GHGs (CO₂, CH₄, N₂O), sensible and latent fluxes, spectrally binned reflectances, surface temperature, and ground heat flux. To achieve fine-scale reflectance observations during the shoulder season, we will evaluate the need for tram-mounted instruments and will deploy a tram system as appropriate. Hydrologic output will be measured as part of the Hydrology tasks. These observational tasks, approaches, and scales are outlined in Table 1.

Isotopic Observations for Land-Atmosphere Exchange. We will analyze the isotopic composition of key stocks and flows that will be useful in testing or constraining integrated predictions. For example, the ¹⁴CO₂ composition of soil respiration is predicted by CLM4.5, and an independent measurement of this value can indicate how well the model simulates the age of carbon being respired as thaw deepens. Some of the observed isotopic values are not yet predicted by the models, but they will be used to develop other validation variables. For example, the hydrology models will not predict the isotopic composition of water, but water isotopes can be used to partition the water flows into constituent flows that are predicted by the model (e.g., sources of subsurface lateral flows or evaporation vs transpiration). The isotope observation tasks, approaches, and applications are shown in Table 2. These tasks are (1) Carbon isotopic composition of CO₂ in soil respiration and soil gas (¹⁴CO₂ and ¹³CO₂); (2) isotopic composition (¹³C-CH₄, H/D-CH_4, ¹⁴C) of CH₄ in soil respiration and soil gas; and (3) Isotopic composition of water in outflow, inundated areas, and plant tissue. An important sub-task will be design, construction, and installation of gas sampling wells in the

different landscape functional units. Vacuum line capabilities for efficient extraction of small-volume samples will be implemented at Lawrence Berkeley National Laboratory (LBNL). The ¹⁴C of roots and SOM are useful diagnostics for modeling of more specific processes and are included in the vegetation and biogeochemistry sections above.

Evaluate model performance and assess improvement in prediction skill. We will estimate prediction errors by comparing output with observations, with a focus on comparisons at the native scale of observations as much as possible. In terms of error estimation, initially we will focus on simple comparisons, expressing differences as root mean square error and evaluating the correlations between modeled and observed quantities. Critically, we will compute these error estimations after completing each phase of model development—for example as observations are applied for parameter inversion or process understanding leads to changes to the model structure—to provide an objective test of model improvement and to quantify improvement in prediction skill.

For time series data, we will conduct evaluations over different averaging periods, such as daily, monthly, and seasonally, to test representation of diurnal and seasonal dynamics. For spatially distributed data, we will conduct evaluations at different spatial scales, testing the up-scaling and down-scaling components of our scaling approach. We will use the observations listed in Table 1 as well as other metrics for model performance that we derive from these datasets, such as light use efficiency and the short-term temperature response of ecosystem respiration. Metrics appropriate for the range of processes integrated in the model (e.g., biogeochemistry, hydrology), spatial scales (fine to climate scale), and temporal scales (hourly to centuries) will be designed to facilitate these comparisons.

At the fine scale, model predictions of biogeochemical dynamics will be evaluated against chamber-based fluxes and observed vertical distributions of carbon and nitrogen compounds in soil organic matter (see BGC tasks). Energy and temperature predictions will be compared using observations from hand-held instruments. A subset of plant community composition observations will be reserved for evaluation of dynamic vegetation model predictions in response to permafrost degradation and resource redistribution (see vegetation tasks).

At the intermediate scale, predicted seasonal hydrology dynamics will be evaluated against observations of surface runoff and fractional inundation area (see Hydrology tasks). Eddy covariance measurements of energy and greenhouse gas fluxes will be used to evaluate landscape-scale predictions, based on dynamic estimates of the measurement footprints. Predictions of landscape-scale LAI and canopy N content will be evaluated against ground and airborne spectroscopy and leaf level measurements (see Vegetation tasks). Emergent relationships between LAI and canopy N content might also provide an independent evaluation of model predictions of landscape-scale GPP.

There are fewer opportunities for evaluation at the climate modeling scale because of challenges in making robust observations at large spatial scales. Albedo and other spectral reflectance measurements present a good opportunity for climate-scale model evaluation, since remote-sensing observations under clear-sky conditions can provide excellent climate-relevant estimates of landscape-scale albedos, including estimates of reflectances in visible and near infrared wavebands.

Data-model integration to examine changes in energy budget and greenhouse gas forcing associated with permafrost degradation. We will generate model-independent estimates of the integrated climate forcing to compare with model prediction. For example, total GHG radiative forcing will be estimated as the sum of all GHG fluxes (on instantaneous molecular-forcing or global-warming-potential basis). We will use these integrated climate forcing estimates to compare with model predictions at different spatial scales. The models will be used to generate hypotheses about the ecosystem processes that will determine future magnitude and rates of feedback. These hypotheses will be explored with Phase 1 data but will also be used to help prioritize efforts in Phase 2. In other words, Phase 2 will be informed by the uncertainty analysis conducted by comparing model output to observations and also by model experiments (e.g., sensitivity analyses).

To generate the independent estimate of forcing, we will construct site-level energy and GHG budgets directly from the observations. We will, for example, apply the simple radiative forcing approach employed by Randerson et al. (2006) in his comparison of energy and GHG effects from a sub-Arctic forest wildfire. Briefly, the forcing from each variable will be expressed as W m^–2 integrated over the time frame in which it operates. We will combine plot measurements with remote sensing to evaluate climate forcing associated with permafrost degradation at Barrow and potentially other sites as well.

As an initial model experiment to generate hypotheses and priorities for Phase 2, we will explore the hypothesis that that Arctic landscapes contain critical thresholds across which small perturbations can qualitatively alter the state of the system. We will begin to investigate whether small amounts of permafrost degradation are amplified by changes in soil structure, hydrology, and insulation, leading to degradation that is practically irreversible. For example, shrub expansion causes changes in albedo and surface energy fluxes that reinforce warming and shrub establishment, and put the system into a new energy-balance state (with climate consequences). This task will be carried into Phase 2 as a primary objective of NGEE Arctic: assessing the potential for arctic ecosystems to undergo irreversible change and/or contribute to abrupt climate change. There are many different components of Arctic ecosystems that could have large, local-to-global climate impacts, such as permafrost degradation and thermokarst, shrub emergence or encroachment and associated effect on albedo, and CO₂ and CH₄ release. In Phase 1, this activity will be initiated with observations and modeling at fine-scale to eddy-flux scale.

Phase 1 Deliverables

• Integrated measurements of land-atmosphere exchange processes for model evaluation: GHG fluxes, heat fluxes and surface reflectance.
• Estimations of SOM turnover in field samples from isotopic composition measurements of soil gases.
• Evaluation of predictive model performance using independent estimates of radiative forcing.