We selected four sites to evaluate the response to fire events and fire exclusion practices, both of which are common in the region [16]: Boulder, Allenspark, Cheesman, and Fort Collins, with the intention of capturing a large range of variability in the Front Range of Colorado and maximizing the realm of inference. Each site had complete weather records, fire history documentation in the scientific literature, and field observations of trace gas flux rates.

Weather data were readily available and complete for each location. At three sites [17], the daily weather records exceeded 100 years with the fourth site (Allenspark) having 58 years of record (Table 1). The long period of observation provided a wide spectrum of variability in weather and climate. The climate of the northern Front Range is dry, with the lowest precipitation occurring in the winter. Precipitation is unimodal (Fort Collins and Boulder) to weakly bimodal (Allenspark and Cheesman) with peaks in the late spring (April and May) and/or summer (July and August) (Figure 1). Even during times of increased moisture vegetation can remain stressed due to the greater temperatures that occur concurrently. Mean annual temperature ranges between 4.7 and 11.0 °C at the four sites (Table 1) while mean annual precipitation ranges from 384–520 mm (Table 1).

At each of these four sites we found estimates for fire return intervals (Table 2). Brown et al. [3] found that composite fire return intervals generally ranged between 6 and 12 years for Cheesman, though there were long periods (1723–2000, with fires in years 1723 and 1851) where fires were much less frequent. Two of the sites, Boulder and Allenspark, are located in Boulder County, Colorado where Veblen et al. [18] conducted an extensive fire history study. Boulder had an average fire return interval of 14 years with exact fire scar dates back to 1679. Sites near Allenspark had an average fire return interval of 40 years based on fire scar chronologies that date back to 1541. For the DAYCENT modeling we constrained the number of fires at these sites by selecting only the fire years during which at least 10% of the examined trees had recorded a fire scar. In order to make a good estimate of fire return interval for the Fort Collins site, we interpolated the fire return interval from an equation developed by Brown and Shepperd [19] for the region that relates fire return interval with latitude. We then pooled that with information from Rocky Mountain National Park [20]. These patterns of fire return interval are dictated in large part by the climate of the region and the individual sites that vary both in latitude and altitude.

A dynamic ecotone is formed as the grasslands of the Great Plains merge into the forests of the foothills of the Colorado Front Range. We focused on two forest types within this ecotone, lower elevation ponderosa pine stands (that tend to be open with substantial grass, 1800–2100 meters) and mid-elevation ponderosa pine (that have some canopy closure, mixed with Douglas fir, 2100–2400 meters) [21,22,23,24].

We used site-specific soil data from the online NRCS Soil Survey. The Northern Front Range and Larimer County, Colorado are dominated by soils with a sandy loam texture. A typical soil profile is 0.90 m thick with bedrock at 1 m depth. Soils of Boulder County, Colorado are gravelly loamy sand with soil depths of approximately 1.5 m. At Allenspark, soils profiles are 1.5 m deep with a cobbly and stony sandy loam soil texture. Soils that surround the Cheesman Reservoir are shallow with weathered bedbrock at 0.75 m depth and the soil texture is gravelly throughout the profile. Using the soil texture triangle we converted texture descriptions from the NRCS Soil Survey to % sand, silt, and clay (Table 3).

Model simulations were initiated with the site-specific soil and weather data described above. We ran a total of four separate simulations, one per site. We initialized the DAYCENT model as a forest ecosystem with a grass component, and simulated plant production with two subroutines: one for the grass understory (CROP.100), and a second for the tree component (TREE.100). We used literature values for other ponderosa pine forests as estimates for initial conditions (pool sizes) for biomass and soil carbon, and ran simulations for each site for 2000 years. Between year 1 and 1500, we scheduled regular fires that reflected the fire return intervals at each site (Table 2). The length of these simulation runs has been found to bring other modeled ecosystems into equilibrium [28]. We examined total soil organic matter pools at each site to verify steady state that we defined as having less than a 5% change in total soil organic matter from year to year. Each of the four simulated sites met this criterion.

At the time of each fire, DAYCENT ran two subroutines, TREM.100 (Table 4) and FIRE.100 (Table 5). The TREM routine removed biomass from trees (and constituent pools) for either a low or high severity fire [29,30]. The FIRE routine worked in a similar fashion except that rather than removing biomass from tree pools, it removed the grass and litter components at rates that correlated to the fire severity. At year 1500, we began to analyze the simulated fluxes (g m⁻²·yr⁻¹) of CH₄, N₂O, and NO as well as gross nitrification rates (g N m⁻²·yr⁻¹) with the intent of observing how they responded to simulated fire events and more recent (circa 1920) fire suppression practices common to the region [16].

Simulated methane uptake was minimally affected by fire. All four of the sites we modeled exhibited a slightly increasing trend over the three-year period that included the year prior to, the year of, and the year after a fire (Figure 2). Allenspark departed the most from pre-fire years, when methane uptake increased by a mean of 0.05 g·CH₄·m⁻² during the year of a fire. At one-year post-fire, the mean CH₄ uptake had decreased though it was still slightly greater than the mean for one year pre-fire. Allenspark also exhibited the greatest variability of the four sites we modeled. This is best explained by the longer fire return interval (40 years), greater fire severity (high), and climate variability. In contrast, each of the other three sites showed much less variability, suggesting that fire has only a minor effect on CH₄ oxidation. None of the factors that control the rate CH₄ oxidation in the DAYCENT submodel, field capacity, soil texture and bulk density, are directly impacted by the modeled fire events. If we extrapolate our estimate to incorporate our knowledge of a fire’s burn area we may use these data to estimate the change in sink strength as a result of a single fire event. In 2002 the Hayman fire burned ~55,000 ha around which the Cheesman site was selected. Using these data, we estimate that the net reduction in sink strength between one year before and one-year after that fire would be approximately 55,000 kg CH₄ (or 1.0 kg ha⁻¹).

Each of the simulated sites exhibited a large increase in gross nitrification during a fire year relative to the year prior. Simulated N availability following the fires was increased at two of the sites, Fort Collins and Allenspark (Figure 2). This increase persisted into the year after the fire for Allenspark only, with Fort Collins returning to pre-fire levels within a year. Gross nitrification rates were relatively unaffected by fire at both Boulder and Cheesman (Figure 2). We attribute this primarily to the lower severity of the fires (less N lost, Table 4, Table 5) that were simulated at these sites. This is as compared to the Allenspark and Fort Collins sites, which generally had longer fire return intervals and consequentially greater fire severity (greater N lost).

Generally, greater fire severity leads to an increase in gaseous N loss from the forest during the fire. However, this effect of reduced substrate availability decreased as competition for nutrients with plants was also greatly reduced as the high severity fires at Allenspark and Fort Collins removed a large portion of the live vegetation. The model compares well with field data; Hamman et al. [50,51] found that fire severity and altered fire regime can directly influence the soil microbial community structure and biogeochemistry in ponderosa pine forests of the Colorado Front Range. Carreira et al. [52] investigated the effect of a single fire and found that it significantly increased N availability and net nitrification rates. Similar findings have been documented in southwestern U.S. ponderosa pine [53,54] and a variety of other ecosystems [55].

We observed three patterns for gross nitrification means (Figure 2): no change (Boulder), response and return (Cheesman and Fort Collins), and persistent change (Allenspark) over the three-year time period for each fire event. Nitrogenous gas (N₂O and NO) fluxes are the result of incomplete oxidation during nitrification or reduction during denitrification. Of the four modeled sites, none had average rainfall greater than 530 mm yr⁻¹. The N-gas submodel is driven by soil moisture and no denitrification occurs when % WFPS < 55% [11]. This was supported by model observation of very low to absent N₂ flux, the end product of denitrification [11].

Nitrous and nitric oxide fluxes (Figure 2) followed a nearly identical pattern to simulated gross nitrification rates. Both are by-products of nitrification, and together, they accounted for up to 25% of the gross nitrification N. In a related study [37] we measured in situ N₂O fluxes along a chronosequence of fires in the Colorado Front Range. We scaled these data up and calculated a mean flux of 0.002 g N m⁻²·yr⁻¹. Other sites in the Front Range were estimated to have N₂O flux rates between 0.164 and 0.937 g·N·m⁻²·yr⁻¹. Averaging across all four sites, DAYCENT calculated annual N₂O fluxes of 0.016 g·N·m⁻²·yr⁻¹, which is an order of magnitude above and below these field measurements. These field based studies were limited in their spatial and temporal coverage, whereas DAYCENT appears to have integrated the variability observed in those studies producing an intermediate estimate of N₂O fluxes from these systems.

Fire suppression practices did not alter methane uptake rates in the DAYCENT simulations (Figure 3). Within site variability was low for these two time periods. We calculate percent difference for pre- and post-fire suppression to be less than 1.2% at all 4 sites. Where methane uptake showed a short term increase immediately following a fire the response appears to be insignificant when considering longer time scales such as a single (hypothetical) fire return interval. While we saw a minor decrease in CH₄ uptake, we had expected greater differences to become apparent as fire suppression allowed tree biomass (or tree densities) to increase, which would consequently increase nutrient competition. The simulation data suggest that either methanotrophic bacteria are not in direct competition with plants for nutrients or that the model fails to capture this interaction.

Nitrous oxide fluxes were greatest at Allenspark, which had nearly 3-fold the rates observed at Cheesman, Boulder, and Fort Collins. Based on the model, this outcome is the result of differences in precipitation and soil properties between Allenspark and the other three simulated sites. DAYCENT first partitions N₂O fluxes from nitrification and denitrification processes and then sums these values into the single value reported as the N₂O flux. At Allenspark, precipitation events stimulated pulses of N₂O flux denitrification, whereas similar events at the other three sites were only great enough to stimulate N₂O flux from nitrification. Changes in N₂O fluxes in response to fire suppression (Figure 3) were greatest at Fort Collins (−36.7%), followed by Cheesman (−25.2%), Boulder (−14.5%), and Allenspark (10.3%). N₂O flux rates at these four sites accounted for <1% of gross nitrification. Nitric oxide flux rates (Figure 3) followed the same pattern, although the fluxes were an order of magnitude greater than N₂O fluxes. Fire suppression reduced NO fluxes by 39.7% at Fort Collins, 28.6% at Cheesman, 19.0% at Boulder, and 15% at Allenspark. Since the climate wasn’t different during the fire suppression period relative to prior, the changes in N₂O and NO flux reflect a change in substrate availability as continued plant uptake during this the fire suppression period led to N immobilization.

We found Allenspark to have the greatest rate of nitrification both before and after fire suppression that was simulated to begin in 1920. Fire suppression decreased gross nitrification rates there the least, by 13.5%, compared to the other three sites: Boulder (−15.2%), Cheesman (−25.3%) and Fort Collins (−37.1%). Stark and Hart [49] measured gross nitrification rates of 25 to 79 mg·N·m⁻²·d⁻¹ (9.1 g·N·m⁻² yr⁻¹ to 28.8 g N·m⁻²·yr⁻¹) in New Mexican and Oregon ponderosa pine, respectively. Stark and Hart [49] suggest that C and N supplies are the primary controls on gross nitrification rates. They suggest that internal cycling of NO₃ ⁻ is strongly controlled by microbial uptake which can be rapid relative to plant uptake in forested ecosystems. The DAYCENT simulations seem to capture this short-term dynamic as gross nitrification rates increase during the year of the fire (Figure 2). Over longer time scales, the DAYCENT simulations predict that fire suppression has led to a decrease in gross nitrification rates relative to pre-1920 rates (Figure 3).

The DAYCENT simulations indicate that fire is important for maintaining N availability in ponderosa pine ecosystems. Covington and Sackett [53,54] found fire to increase NH₄ ⁺-N immediately in ponderosa pine forests of varying ages in Arizona. Within a year that NH₄ ⁺ pulse had dissipated through nitrification processes leading to a pulse in NO₃ ⁻-N that was detected one year after fire. A review of N response to fire [55] documented a similar pattern across multiple ecosystems for NH₄ ⁺ and NO₃ ⁻. Long-term decreases in gross nitrification observed in DAYCENT may be indicative of leaching losses of NO₃ ^‑ from these forested ecosystems that may have consequences for production in the future.