The modeled effects of fire on carbon balance and vegetation abundance in Alaskan tundra

Arctic climate is warming at a rate disproportionately faster than the rest of the world. Changes have been observed within the tundra that are attributed to this trend, including active layer thickening, shrubland expansion, and increases in fire frequency. Whether tundra remains a global net sink of carbon could depend upon the effects of fire on vegetation, specifically concerning the speed at which vegetation reestablishes, the stimulation of growth after fire, and the changes that occur in species composition during succession. While rapid regeneration of graminoid vegetation favors the spread of this functional type in early succession, late succession appears to favor shrub vegetation at abundances greater than those observed before fire. Possible reasons for this latter observation include changes in albedo, soil insulation, and soil moisture regimes. Here we investigate the course of succession after fire disturbance within tundra ecosystems, and the mechanisms involved. A literature review was conducted over previous studies on burn sites, and a series of simulated burn experiments were attempted on the burn site left by the 2007 Anaktuvuk River fire to assess the behavior of version 2.1 of the Ecosystem Demography model (ED2) in the simulation of tundra fire. Though uniquely suited for the heterogeneous landscapes found within tundra, ED2 has not yet been applied to these ecosystems. Prior to validation, we parameterize and calibrate ED for the Alaskan tundra. The land surface sub-model within ED is modified to simulate permafrost through the effects of an increased soil-column depth, a specialized peat texture class, and the simulated effects of wind compaction and depth hoar on snow density. Parameterization was conducted through Bayesian techniques used to constrain parameter distributions based upon data from a literature survey, field measurements at Toolik Lake, Alaska, and an assimilation of three datasets. At each step, priority was assigned to measurements that could constrain parameters that account for the greatest explained variance in model output as determined through sensitivity analysis. Results of variance decomposition were compared to judge the contribution of each effort to model uncertainty. Among the datasets considered by assimilation were estimates of net ecosystem exchange at Barrow and Atqasuk, Alaska, and height-based growth data collected from bud-scar measurements in the same vegetation survey used to derive field measurements. Variance decomposition of the model following the literature survey and field campaign revealed that reproductive allocation, growth respiration, and a leaf allometry parameter explained the largest fraction of model uncertainty among any parameter across all plant functional types. Evergreens were ranked as the plant functional type whose parameters were most responsible for model uncertainty, followed by deciduous and then graminoids. This was attributed to a steady increase in evergreen biomass throughout simulations. The literature survey provided the greatest constraint to model parameters; however model uncertainty increased following the literature survey due to changes in model sensitivity towards parameters. Data assimilation was found to provide the second greatest constraint of any method of constraint considered here. Height-based growth was most effective in constraining parameters among data assimilation sources, and carbon flux at Atqasuk was least effective. The limited field measurements we conducted were least effective in constraining parameters. Changes made to the soil sub-model were successful in simulating permafrost. Height and bud-scar measurements used during data assimlation against modeled growth were also utilized in a statistical analysis with the purpose of investigating whether height-based growth has increased throughout the observational record, as well as whether growth has any relation to meteorological drivers including temperature and precipitation. From our dataset of bud-scar measurements, we found plant height was a significant factor in determining annual height growth. Plot, vegetation class, and species were not effective predictors of growth. Year was found to have an effect on growth, but no significant trends were found in growth over time. No significant trends were found between growth and precipitation; however growth is significantly related to the mean annual temperature of the year prior to growth. This relationship was strongest when considering only the effects of temperature during summer months. Following parameterization, a series of simulations were performed to gauge the suitability of ED in predicting carbon balance and vegetation composition following fire within tundra. Ensembles of model runs were conducted within burned and unburned sites along the Anaktuvuk River fire scar. Modeled net ecosystem exchange at these sites was compared to the observations from flux towers. In addition, a series of simulations were performed at these sites to access the suitability of the model in simulating post-fire vegetation succession over a time scale of 20 years. Two simulations were performed on burned and unburned tundra, as was done in the ensemble analysis. An additional set of three simulations was also performed on unburned tundra in which one of three alterations was applied that were simulated in burned tundra. Alterations reflected observations made in past studies within the Anaktuvuk River burn scar, and consisted of a reduction of aboveground biomass, a temporary reduction in surface albedo, and a reduction in the depth of the organic soil layer. Predictions made by ED differed significantly from observations within both burned and unburned tundra. Ensemble runs on burned and unburned tundra were both able to encompass a majority of observed carbon flux measurements within their 95% credible intervals. Modeled NEE was found capable of recovering to pre-fire values within 7 years after the fire, and carbon lost during the fire was restored within 3 years. Results of simulations considering separate alterations from fire suggest the nature of plant composition and carbon balance within the model is driven heavily by the combustion of vegetation, with alterations to surface albedo providing an effect to a lesser degree. Several issues were observed within long-term simulation, including a rapid growth within evergreen and deciduous vegetation, the die-off of graminoid vegetation, and a failure to reestablish evergreen vegetation following their removal by fire. Recovery within the model was dominated by a rapid succession of the deciduous PFT, which conflicted with observations from literature that stated early succession was dominated by graminoids. Further work will be needed for ED to make reliable predictions within tundra ecosystems