Abstract

Fire regimes are influenced by both exogenous drivers (e.g., increases in atmospheric CO₂ and climate change) and endogenous drivers (e.g., vegetation and soil/litter moisture), which constrain fuel loads and fuel aridity. Herein, we identified how exogenous and endogenous drivers can interact to affect fuels and fire regimes in a semiarid watershed in the inland northwestern United States throughout the 21st century. We used a coupled ecohydrologic and fire regime model to examine how climate change and CO₂ scenarios influence fire regimes. In this semiarid watershed, we found an increase in burned area and burn probability in the mid‐21st century (2040s) as the CO₂ fertilization effect on vegetation productivity outstripped the effects of climate change‐induced fuel decreases, resulting in greater fuel loading. However, by the late‐21st century (2070s), climatic warming dominated over CO₂ fertilization, thus reducing fuel loading and burned area. Fire regimes were shown to shift from flammability‐ to fuel‐limited or become increasingly fuel‐limited in response to climate change. We identified a metric to identify when fire regimes shift from flammability‐ to fuel‐limited: the ratio of the change in fuel loading to the change in its aridity. The threshold value for which this metric indicates a flammability versus fuel‐limited regime differed between grasses and woody species but remained stationary over time. Our results suggest that identifying these thresholds in other systems requires narrowing uncertainty in exogenous drivers, such as future precipitation patterns and CO₂ effects on vegetation.

Full text

Turn on search term navigation

Introduction

Frequent low-intensity fires are an important component of many forest ecosystems, for example, contributing to the regulation of energy, water, and carbon cycling (Agee, 1996; Flitcroft et al., 2016; Hessburg et al., 2005). However, in recent decades, large and high severity wildfires have become more common in locations that historically burned at low intensity (Deb et al., 2020; Williams et al., 2019; Xu et al., 2021). These fire regime shifts can transform ecosystem dynamics and structure, increase air and water pollution, cause flood and landslide hazards, and threaten human property and lives (Biswas et al., 2021; Bowman et al., 2009; Kemter et al., 2021; Smith et al., 2016). To mitigate future wildfire risk in the wildland urban interface, and to predict fire hazard across wildlands, forest managers need to better understand how climate change and wildfire interact at scales where management actions are implemented. However, this can be challenging because climate-wildfire interactions can vary at fine scales in response to local environmental conditions, such as topography and vegetation cover (Hanan et al., 2021; Pausas & Paula, 2012; Qi et al., 2012).

While climate change is playing an essential role in facilitating large forest fires in the western U.S. (Abatzoglou & Williams, 2016; Brey et al., 2021; Pechony & Shindell, 2010; Westerling et al., 2014), the effects of climate change are likely to vary across a continuum of climate, vegetation, and anthropogenic factors. For example, climate change (i.e., warming) can increase the frequency, duration, and intensity of drought, which can in turn increase fuel aridity and fire hazard in relatively mesic, fuel-rich environments (Abatzoglou & Kolden, 2013). However, it can simultaneously decrease fire potential in relatively arid, fuel-limited environments by decreasing net primary productivity (NPP) and, therefore, fuel loads (Hanan et al., 2021; Kennedy et al., 2021; Littell et al., 2016, 2018).

Understanding how climate change influences vegetation, fuel aridity, and fuel loading is further complicated by the role of rising atmospheric CO₂ concentrations, which can modify plant (and therefore fuel) responses to drought (Becklin et al., 2017; Warren et al., 2011). For example, increasing CO₂ concentrations can increase plant productivity by increasing photosynthetic and water-use efficiency, thereby partially offsetting the suppressive effects of drought (Becklin et al., 2017; Lewis et al., 2009; Sullivan et al., 2020). Because rising CO₂ concentrations and warming influence vegetation, fuel loading, and fuel moisture in opposite ways, it is important to disentangle which drivers dominate in different locations and under different climate conditions.

Many recent studies have focused on understanding how climate, ecosystem structure, and fuel conditions interact to drive wildfire regimes at different scales (Abatzoglou & Kolden, 2013; Halofsky et al., 2020; Hicke et al., 2012; McCarley et al., 2017; Williams et al., 2019). However, many of these observational studies are limited in their ability to isolate specific drivers or project into the future (Hicke et al., 2012; McCarley et al., 2017). Alternatively, empirical models and conceptual models have been used to predict different characteristics of future fire activity (e.g., fire frequency, Bradstock, 2010; burned area, Littell et al., 2018; fire potential, Liu et al., 2013). However, a limitation to these models is that they do not consider dynamic vegetation and changes in fuel loading, which are important for projecting future fire regimes (McKenzie et al., 2004; Pausas & Paula, 2012). Process-based models can complement field and empirical modeling studies by bridging spatial and temporal scales while also accounting for feedbacks among climate change, rising CO₂, vegetation productivity, and fire. They also enable researchers to manipulate drivers to isolate their individual and combined effects (Hanan et al., 2021).

In this study, we addressed the overarching question: How does vegetation modulate the effects of climate change on fire regimes in a semiarid watershed? We applied the coupled ecohydrological, fire regime modeling platform RHESSys-WMFire (Bart et al., 2020; Kennedy et al., 2017; Tague & Band, 2004) in a semiarid watershed in the US Inland Northwest. We used a range of possible future scenarios to assess how climate change and increasing atmospheric CO₂ interact to influence vegetation and, thus, the roles of fuel loading and fuel aridity in determining future fire regimes. Specifically, we addressed the following questions:

What are the relative and opposing roles of two key exogenous drivers in driving fire regimes: climate change (i.e., warming and changes in precipitation) and increasing CO₂ (Figure 1)?
What are the relative and opposing roles of two key endogenous drivers in driving fire regimes: fuel load and fuel aridity (Figure 1)?

Climate change and increasing CO₂ can either counteract or compound one another to influence fuel aridity, fuel loading, and resultant fire regimes. The extent to which a particular ecosystem is affected by either mechanism depends on whether or not fire is limited by fuel (biomass available for burning) or flammability (environmental conditions that enable the fuel to burn), which are partially controlled by climate (Bradstock, 2010; Krawchuk et al., 2009; Littell et al., 2009). We hypothesize that climate change can affect fuel loading and fuel aridity through changes in vegetation productivity, evapotranspiration (ET), and litter decomposition. At finer scales, historical aridity gradients across a watershed [potential evapotranspiration (PET)/precipitation (P)] also play an important role in determining the spatial distribution of how fire regimes respond to climate change (Hanan et al., 2021). Thus, it is essential to account for these biophysical and biogeochemical processes in projections of future fire regimes in order to improve our understanding of how they are changing in semiarid landscapes and to support decision-making processes at management-relevant scales.

Methods Study Area

Trail Creek is a 167-km² sub-catchment of the Big Wood River Basin located in Blaine County, Idaho, U.S., between the Salmon-Challis National Forest and Sawtooth National Forest (43.44 N, 114.19 W; Figure 2). The mean annual precipitation in the area is around 980 mm, of which 60% is snow (Frenzel, 1989). Trail Creek is characterized by cold, wet winters and warm, dry summers. Elevations range from 1,760 to 3,478 m, and there is a vegetation and aridity gradient following changes in elevation. We used an aridity index, calculated as the ratio of 38-year (water years 1980–2018) average annual PET to average annual P, to classify the spatial variation in aridity. We define areas with PET/P > 2 as water-limited, PET/P < 0.8 as energy-limited, and PET/P between 0.8 and 2 as balanced (McVicar et al., 2012). As Figure 2 depicts, lower to middle elevation slopes are water-limited and covered by sagebrush (Artemisia tridentata ssp.), mixed riparian species, and grasslands; middle to higher elevation areas are water-energy balanced and are covered by Douglas-fir (Pseudotsuga menziesii), lodgepole pine (Pinus contorta varlatifolia), subalpine fir (Abies lasiocarpa), and mixed shrub and herbaceous vegetation (Buhidar, 2001). Soils in Trail Creek are mainly coarse, permeable alluvium (Smith, 1960). No large wildfires (>400 ha) have occurred in the last 40 years (MTBS, Eidenshink et al., 2007). The soils, vegetation, and topography, however, are comparable to several sub-catchments on the western side of the Big Wood River Basin, which were burned in the 2013 Beaver Creek Fire (total 45,036 ha burnt area, Skinner, 2013).

View Image - Figure 1. Exogenous (external) and endogenous (internal) drivers of fire regimes. Question 1 (Q1) focuses on the role of exogenous drivers while Question 2 (Q2) focuses on endogenous drivers.

Figure 1. Exogenous (external) and endogenous (internal) drivers of fire regimes. Question 1 (Q1) focuses on the role of exogenous drivers while Question 2 (Q2) focuses on endogenous drivers.

View Image - Figure 2. Study site—Trail Creek. (a) Land cover overlapped with topography (elevations range from 1,760 to 3,478 m); (b) fire regime from LANDFIRE data (Rollins, 2009, Table 1); the gridded areas show different long-term aridity indices (water years 1980–2018).

Figure 2. Study site—Trail Creek. (a) Land cover overlapped with topography (elevations range from 1,760 to 3,478 m); (b) fire regime from LANDFIRE data (Rollins, 2009, Table 1); the gridded areas show different long-term aridity indices (water years 1980–2018).

The aridity index defined above generally correlates with the fire regime classifications from LANDFIRE, which are based on vegetation cover, ecological and vegetation simulation, and successional modeling (Figure 2; Rollins, 2009). In the southern, water-limited, part of the basin, the mean LANDFIRE fire return intervals (FRI) are short (i.e., 35 years) and fires are generally low or mixed severity. LANDFIRE classifies this part of the watershed as having a fuel-limited fire regime (the fuels are dry enough to burn in most years, but there is rarely enough fuel to carry fire). The northern, balanced, part of the basin, on the other hand, has a long mean FRI (i.e., 200 years) and typically burns at high severity. LANDFIRE classifies this part of the watershed as having a flammability-limited fire regime (there is enough fuel present, but fuels are generally too moist to burn or ignitions are limited; LANDFIRE, Rollins, 2009, Figure 2 and Table 1). The central part of the basin is a transitional zone, with some fuel-limited patches located in relative mesic areas, and some flammability-limited patches located in more water-limited areas. Within the northern part of the basin, there is also a small water-limited area due to slightly lower precipitation.

Table 1 Fire Regime Groups and Corresponding Characteristics (Rollins, 2009)

Fire regime group	Characteristics
Fire regime group I	≤35-Year fire return interval, low and mixed severity
Fire regime group II	≤35-Year fire return interval, replacement severity
Fire regime group III	35–200-Year fire return interval, low and mixed severity
Fire regime group IV	>200-Year fire return interval, any severity

Model Description

We used the coupled ecohydrologic-fire regime modeling platform RHESSys-WMFire (Bart et al., 2020; Kennedy et al., 2017) to model fire and vegetation responses to a range of climate and CO₂ scenarios. The Regional Hydro-ecologic Simulation System (RHESSys, Tague & Band, 2004) is a mechanistic model designed to simulate the effects of climate and land use change on ecosystem carbon (C) and nitrogen (N) cycling and hydrology. RHESSys fully couples hydrology (streamflow, ET, soil moisture), C (photosynthesis, respiration, NPP, mortality), and N fluxes (mineralization, nitrification, denitrification, plant uptake, and leaching) at a hierarchy of scales (e.g., patch, zone, sub-basin, basin). Photosynthesis is calculated based on the Farquhar model, which is a function of nitrogen, radiation, stomatal conductance, atmospheric pressure, atmospheric CO₂ concentration, and daily average temperature (Farquhar & von Caemmerer, 1982). Elevated atmospheric CO₂ concentrations can increase photosynthesis rates (i.e., CO₂ fertilization). Stomatal conductance is based on the Jarvis (1976) model of stratum conductance, which accounts for the effects of light, atmospheric CO₂ (not implemented in RHESSys), leaf water potential, and vapor pressure deficit (Running & Coughlan, 1988; Tague & Band, 2004). Recent empirical studies have shown that higher CO₂ concentrations can reduce stomatal conductance and increase plant water use efficiency; however, the magnitude of stomatal responses to CO₂ is not clear and responses can change over different time scales (Farquhar et al., 1978; Medlyn et al., 2011). Therefore, our model only simulates the CO₂ fertilization effect on photosynthesis (Farquhar & von Caemmerer, 1982), while it does not account for physiological adaptations that may reduce stomatal conductance. Litter decomposition models are based on the method developed by Thornton (1998), which is related to the C:N ratio and potential decay rate. The potential decay rate is limited by soil moisture, temperature, and nitrogen, and a higher temperature may cause a higher decay rate. RHESSys has been widely tested and applied in many mountainous watersheds (Garcia & Tague, 2015; Hanan et al., 2017, 2018, 2021; Lin et al., 2019; Son & Tague, 2019). A more detailed description of the RHESSys model can be found in Tague and Band (2004).

RHESSys-WMFire couples RHESSys with a model for fire spread (WMFire, Kennedy et al., 2017) and a model for fire effects (Bart et al., 2020), which capture fuel and climate controls on fire spread and severity. RHESSys-WMFire computes key processes at a daily time-step and partitions the landscape into patches (the smallest spatial unit, typically 30–120 m). The model therefore accounts for spatial differences in energy and precipitation conditions on soil moisture, ET, vegetation growth, and fire dynamics. Notably the model accounts for the lateral downslope redistribution of water. WMFire is a stochastic model that requires several replicate simulations (200 in the current study) to attain a representative result (Kennedy, 2019). Additional details on the model framework are provided in Text S1 in Supporting Information S1.

Input Data Selection of Storylines and CO₂ Data

Because RHESSys-WMFire is computationally intensive, we cannot conduct our simulations using data from every General Circulation Model (GCM) or Earth System Model (ESM). Thus we selected four GCMs/ESMs to bracket the range of vegetation- and fuel-relevant climates from a pool of 20 available models (Figure S1 in Supporting Information S1). We used three variables to screen GCMs: (1) a measure of annual water deficit (DEF), calculated as PET (using the “FAO56” Penman–Monteith equation for a reference grass surface, Allen et al., 1998) minus actual evapotranspiration (AET), (2) a measure of annual plant available moisture (i.e., AET), and (3) 100-hr dead fuel moisture (FM100) in summer, defined as a biomass volume that takes 100 hr to lose or gain 2/3 of the difference between the dead fuel itself and the surrounding atmosphere. Note that for the purpose of model selection, PET and AET were estimated at a monthly timescale (Abatzoglou & Rupp, 2017). The changes in DEF between the historical (1971–2000) and future (2040–2069) periods is a proxy for changes in vegetation moisture stress that promote flammability; these changes are most important for modulating burned area in flammability-limited forests (Littell & Gwozdz, 2011). The changes in AET between historical and future periods are a proxy for potential changes in plant productivity and fuel accumulation and area, therefore, more important for modulating fire occurrence in fuel-limited environments (Abatzoglou & Kolden, 2013). Furthermore, changes in standard deviation (SD) of PET and AET between the future and the historical climate are also important because climate variability can increase or temper FRI. FM100 is calculated based on the National Fire Danger Rating System using climate data from GCMs (Cohen & Deeming, 2006).

We started with 20 GCMs from the Coupled Model Intercomparison Project 5 (CMIP5; Taylor et al., 2012) that have been statistically downscaled across the contiguous United States using the Multivariate Adaptive Constructed Analogs (Abatzoglou & Brown, 2012) with 1/24-degree resolution (∼4 km) covering the time period from 1950 to 2100. Then, we used six metrics based on the assessment criteria described above to select future climate scenarios: the change in mean AET and DEF (i.e., 2040–2069 mean vs. 1971–2000 mean); the SD of monthly AET and DEF; and the number of days per year where FM100 in the future (i.e., 2040–2069) is less than the third percentile value for the historical period (i.e., 1971–2000), and the tenth percentile. Based on the ranking shown in Figure S1 in Supporting Information S1, we selected four models to cover the range of fuel- and fire-related climates (Table 2), a model representing the mean behavior (IPSL-CM5A-LR, MultiMean), one with a large increase in aridity that may promote drought (CSIRO-Mk3-6-0, ProDrought), one that promotes increased productivity but limits fire (GFDL-ESM2G, ProVeg), and one with a significant fluctuation of fire-related metrics (INMCM4, ProFire). For forcing RHESSys-WMFire, we focused on representative concentration pathway (RCP) 8.5 due to its close agreement with historical total cumulative CO₂ emissions during the historical time period through 2015 (Meinshausen et al., 2011; Schwalm et al., 2020), we also included results from RCP4.5 in Figures S13, S14 and Text S3 in Supporting Information S1 to support our discussion.

Table 2 Lists of Four Selected GCMs/ESMs Based on Fire-Related Characteristics

Model #	Storyline name	Model name	Characteristics
1	ProDrought	CSIRO-Mk3-6-0	A large increase in summer aridity that may promote drought
2	ProVeg	GFDL-ESM2G	Promotes increased productivity and limits fire
3	ProFire	INMCM4	A significant increase in fire-related metrics (FM100)
4	MultiMean	IPSL-CM5A-LR	Close to the multi-model mean

To evaluate the effect of rising CO₂ concentrations, we used observed and projected CO₂ concentrations from 1900 to 2099 produced by based on the Intergovernmental Panel on Climate Change Fifth Assessment Report (IPCC AR5). For the “with CO₂ fertilization effect,” we used transient CO₂ concentrations at an annual time step from AR5; and for the “without CO₂ fertilization effect,” we used a constant CO₂ concentration, that is, 353 ppm; the concentration in 1990.

Model Calibration and Initialization

To calibrate soil parameters, we used daily, high-resolution (1/24° or ∼4 km) gridded meteorological data from gridMET (Abatzoglou, 2013, Table 3, A), including minimum and maximum temperatures, precipitation, relative humidity, shortwave radiation, and wind speed from 1979 to 2017. To validate WMFire, we extended the gridMET records back to 1900 by interpolating ERA-20C daily reanalysis data (1900–2010, Poli et al., 2016) to the spatial resolution of gridMET. The overlapping period in these two datasets (1979–2010) were used to bias-correct the interpolated daily record (1900–1978). Further, we also bias-corrected the reconstructed record using Parameter-elevation Relationships on Independent Slopes Model data (PRISM, 1895–2017, monthly, Daly et al., 1994) at a monthly time step to ensure compatibility between the two records. Six groundwater-related parameters in RHESSys were calibrated based on streamflow observations. Fire spread in WMFire was validated against LANDIRE estimates (Rollins, 2009). A detailed description of RHESSys calibration and WMFire validation is provided in Text S2 in Supporting Information S1 (Table 4).

Table 3 Climate Forcing Data, Atmospheric CO₂ Concentration, and Their Usage for This Study

[Table omitted. See PDF]

Table 4 Summary of Model Simulation Scenarios

[Table omitted. See PDF]

Note. “Climate change” refers to the climate change effect with no CO2 fertilization (using constant CO2 concentration - ppm as the model input). “Climate change and increasing CO2” refers to their combined effects (using both RCP8.5 climate data and transient CO2 concentrations as inputs. Meinshausen et al., 2011).

After model calibration, we initialized vegetation and soil C and N pools using a target-driven spin-up approach (Hanan et al., 2018). This involved running the model for 300 years using the same observed climate data from the model calibration and WMFire validation (1900–2017), but with the anthropogenic climate change signal removed (Table 3, B) per Hanan et al. (2021), and with the fire model “on.” From this we obtained the pre-industrial and pre-suppression condition as a starting point for running different climate change scenarios. The vegetation initialization was conducted once for all scenarios, that is, they had same initial vegetation state, litter, and soil C and N storage.

Other Biophysical and Land Cover Data

We aggregated a 10-m resolution digital elevation model from the US Geological Survey National Elevation Database to 100-m resolution (USGS, 2016) to generate the topographic properties and watershed structure of Trail Creek, which include elevation, slope, aspect, patches, sub-basins, and basin boundaries. In total, we delineated 72 sub-basins and 16,705 patches. We used the National Land Cover Database (NLCD 2016; Dewitz, 2019) to classify five vegetation types. Of these, 49.6% were evergreen, 24.9% were shrub, 22.0% were grass, 0.3% were deciduous, and 3.1% were not vegetated. Soil type was assigned using a spatial continuous probability soil map (POLARIS) created by Chaney et al. (2016).

Modeling Scenarios and Climate Forcing

We conducted model simulations using RHESSys-WMFire over three major timeframes for each storyline, both with and without CO₂ fertilization: baseline, the 2040s, and the 2070s. The 2040s and 2070s timeframes were simulated with the RCP8.5 (Meinshausen et al., 2011). We also include a subset of results from RCP4.5 in Text S3 in Supporting Information S1 to augment our discussion. For each timeframe there was a 55-year scenario spin-up and a 30-year assessment period (Figure 3) for a total 85-year of simulation with the fire model turned on. Each spin-up period was initialized with the same initial conditions generated in the vegetation initialization described above.

View Image - Figure 3. Future fire simulation scenarios using the ProDrought storyline as an example.b, e, and h show the assessment of the 2040s (2031–2060); c, f, and i show the assessment of 2070s (2061–2090); a, d, and g are for baseline scenarios. There were three simulation periods: the vegetation initialization period (not shown), the scenario spin-up period, and the assessment period. For the scenario spin-up period, there were two different climate inputs: baseline and RCP8.5; and two different CO2 fertilization scenarios (with and without). We used the historical output from individual GCMs as a baseline for calculating future climate change effects. For the scenario spin-up period, the baseline scenario used historical data from 1951 to 2005. For the assessment period (30 years), the baseline scenario used data that concatenated the first 15 years of the historical record (1950–1964) to its last 15-year records (1991–2005). We considered the future climate and CO2 fertilization scenarios in isolation and together to build different landscape fuel conditions for the subsequent assessment periods. For the CO2 fertilization effect, we used transient CO2 concentrations at an annual time step from IPCC as the model input; for the no CO2 fertilization scenarios, we used a constant CO2 concentration (i.e., 353 ppm). For the assessment period (30 years), we used the spun-up fuel conditions as initial conditions and ran the fire model to assess the relative contribution of different factors in driving fire regimes.

Figure 3. Future fire simulation scenarios using the ProDrought storyline as an example.b, e, and h show the assessment of the 2040s (2031–2060); c, f, and i show the assessment of 2070s (2061–2090); a, d, and g are for baseline scenarios. There were three simulation periods: the vegetation initialization period (not shown), the scenario spin-up period, and the assessment period. For the scenario spin-up period, there were two different climate inputs: baseline and RCP8.5; and two different CO2 fertilization scenarios (with and without). We used the historical output from individual GCMs as a baseline for calculating future climate change effects. For the scenario spin-up period, the baseline scenario used historical data from 1951 to 2005. For the assessment period (30 years), the baseline scenario used data that concatenated the first 15 years of the historical record (1950–1964) to its last 15-year records (1991–2005). We considered the future climate and CO2 fertilization scenarios in isolation and together to build different landscape fuel conditions for the subsequent assessment periods. For the CO2 fertilization effect, we used transient CO2 concentrations at an annual time step from IPCC as the model input; for the no CO2 fertilization scenarios, we used a constant CO2 concentration (i.e., 353 ppm). For the assessment period (30 years), we used the spun-up fuel conditions as initial conditions and ran the fire model to assess the relative contribution of different factors in driving fire regimes.

For the baseline scenario, we first ran the model for 55 years using historical forcing inputs from each of the four selected GCMs (1950–2005; Table 3). We then repeated the climate data over the years 1950–1964 and 1991–2005 to generate the 30-year baseline assessment period. Climate change effects were assessed against the patterns observed in the baseline time period.

For the future scenarios, we simulated a 55-year spin-up period using each of the four storylines and starting with the same vegetation initialization described above. We then repeated the last 15 years of the climate data and combined it with the next 15 years to produce the assessment period. For example, for the 2040s simulation we spun the model up for 55 years using climate data from 1991 to 2045, then simulated an additional 30-year assessment period using climate data from 2031 to 2060. For the 2070s simulation we spun the model up for 55 years using climate data from 2021 to 2075, then simulated an additional 30-year assessment period using climate data from 2061 to 2090. All simulations were repeated with and without CO₂ fertilization for each of the four storylines.

“Climate change” refers to the climate change effect with no CO₂ fertilization (using constant CO₂ concentration −353 ppm as the model input). “Climate change and increasing CO₂” refers to their combined effects (using both RCP8.5 climate data and transient CO₂ concentrations as inputs; Meinshausen et al., 2011).

For evaluating the changes in fire regimes, we only analyzed outputs for the 30-year assessment period. We calculated four main fire characteristics for each of the 200 independent Monte Carlo replicates for each scenario: mean number of patches burned per fire, 95th percentile fire size in the 30-year assessment period, annual area burned (AAB, mean area burned each year), and number of fire starts (fires that surpass 30 burned patches in the 30-year assessment period). We then compared distributions of these variables among the model scenarios.

We also considered patch-level summaries of burn probability (P_burn), fuel aridity, and fuel loading. For each patch, the burn probability (P_burn) was calculated as: [Image Omitted. See PDF]

Fuel aridity was calculated by WMFire as relative deficit and was calculated monthly as: [Image Omitted. See PDF]

We calculated patch-level mean burn probability, fuel loading, and fuel aridity to understand the relationships among these variables in the different model scenarios.

Results

In the following descriptions, we use “overall effect” to represent both the climate change and CO₂ fertilization effects on fire regimes (e.g., burned area, fire size, fire starts), while “climate change effect” represents the climate change only effects without CO₂ fertilization.

Effects of Climate Change and CO₂ Fertilization on Fire Starts and Size

During the 2040s assessment period, the climate change effect decreased burned area and fire size relative to baseline (Figures 4a and 4d). In contrast, the overall effect increased or only slightly decreased burned area and fire size (Figures 4a and 4d). These patterns correspond to climate change and CO₂ effects on fuel and vegetation. Climate change alone decreased overall fine fuel loading by increasing litter decomposition (Figure S5 in Supporting Information S1), whereas CO₂ fertilization increased burned area and fire size by enhancing vegetation productivity and thus fuel loading (Figure S6 in Supporting Information S1).

View Image - Figure 4. Boxplots showing projected fire characteristics for the assessment period of 2031–2060 under each climate change and increasing atmospheric CO2 storyline: (a) mean fire size (mean number of patches burned per fire), (b) 95th percentile fire size in the 30-year assessment period, (c) number of fire starts (fires that burned more than 30 patches in the 30-year assessment period), and (d) annual area burned (mean area burned each year). Red lines represent the median value for the baseline scenario. Box plots show medians, 25th and 75th percentiles, and 95% confidence intervals. “Climate change (2040s)” refers to the climate change effect without CO2 fertilization, and “climate change and increasing CO2 (2040s)” refers to the overall effect. All four fire characteristics were calculated first within each independent simulation replicate (there were 200 simulation replicates for each scenario). Their distribution across 200 simulations is shown for each scenario.

Figure 4. Boxplots showing projected fire characteristics for the assessment period of 2031–2060 under each climate change and increasing atmospheric CO2 storyline: (a) mean fire size (mean number of patches burned per fire), (b) 95th percentile fire size in the 30-year assessment period, (c) number of fire starts (fires that burned more than 30 patches in the 30-year assessment period), and (d) annual area burned (mean area burned each year). Red lines represent the median value for the baseline scenario. Box plots show medians, 25th and 75th percentiles, and 95% confidence intervals. “Climate change (2040s)” refers to the climate change effect without CO2 fertilization, and “climate change and increasing CO2 (2040s)” refers to the overall effect. All four fire characteristics were calculated first within each independent simulation replicate (there were 200 simulation replicates for each scenario). Their distribution across 200 simulations is shown for each scenario.

Fire responses differed among these four storylines. When considering the climate change effect, the ProDrought storyline exhibited the greatest decreases in burned area and the ProVeg storyline exhibited the smallest decreases (Figure 4d). This occurred because the ProDrought storyline had the largest decline in fuel loading, which outstripped the effects of increasing fuel aridity (Figures S2 and S3 in Supporting Information S1). When considering the overall effect (climate change and CO₂), the ProVeg storyline projected the largest increase in fire size and AAB with no noticeable increase in the number of fire starts (i.e., fires that burned more than 30 patches). This occurred because the storyline promoted productivity and, therefore, increases in fuel loading (Figures S2 and S4 in Supporting Information S1). The MultiMean storyline also projected an increase in fire size for the overall effect but the increase was smaller in magnitude than ProVeg. ProFire and MultiMean predicted the most significant increases in the number of fire starts for the overall effect, but differed in their fire size responses (Figure 4c).

Fire size and burned area had a non-monotonic response to exogenous drivers over the 21st century (i.e., in the 2070s; Figure 5). When only considering climate change (for all four storylines), there was a dramatic decrease in all fire metrics in the 2070s compared to the historical scenario, consistent with the simulated decrease for the 2040s. The effect of CO₂ fertilization, however, differed between the two timeframes. The overall effect of climate change and CO₂ fertilization increased fire size and burned area in the 2040s, but in the 2070s, it either had no effect or decreased them (Figure 5). Warming increased aridity and reduced vegetation productivity (Figures 8b and 8d), which reduced fuel accumulation. It also increased decomposition rates, which further reduced fuel loading (Figures S6 and S9 in Supporting Information S1). Faster decomposition of litter and reduced vegetation productivity were the dominant factor reducing fuel loading. CO₂ fertilization, however, increased fuel loading by enhancing vegetation productivity (Figure S8 in Supporting Information S1), which counteracted warming-induced decreases in fuel loading in the 2040s. However, by the 2070s, climate change-driven decreases in fuel loading outstripped any compensatory effects of CO₂ fertilization (Figures S2 and S6 in Supporting Information S1). This led to an overall decrease in fire size and burned area from the 2040s to the 2070s (e.g., Figure 5d).

View Image - Figure 5. Boxplots showing projected fire characteristics for the assessment period of 2061–2090 under each climate change and increasing atmospheric CO2 storyline: (a) mean fire size, (b) 95th percentile fire size in the 30-year assessment period, (c) number of fire starts, and (d) annual area burned. Red lines represent the median value for the baseline scenario. Box plots show medians, 25th and 75th percentiles, and 95% confidence intervals.

Figure 5. Boxplots showing projected fire characteristics for the assessment period of 2061–2090 under each climate change and increasing atmospheric CO2 storyline: (a) mean fire size, (b) 95th percentile fire size in the 30-year assessment period, (c) number of fire starts, and (d) annual area burned. Red lines represent the median value for the baseline scenario. Box plots show medians, 25th and 75th percentiles, and 95% confidence intervals.

Effects of Exogenous and Endogenous Drivers on Burn Probability (P_burn)

Fuel conditions varied among vegetation cover types (Figure 6). For example, region A (grass-dominated) was very fuel-limited (fuel load <0.3 kg/m², Figure S10 in Supporting Information S1 and Figure 6b) and had a broader range of fuel aridity conditions; region B (shrub-dominated) was fuel-limited (fuel load <0.5 kg/m², Figure S10 in Supporting Information S1 and Figure 6b) and very arid (fuel aridity >0.6); and region C (evergreen forest) was fuel abundant (fuel load >0.4 kg/m²) and relatively mesic (0.5 < fuel aridity < 0.85, Figure S10 in Supporting Information S1 and Figure 6b).

View Image - Figure 6. Relationships among fuel load (i.e., litter carbon), fuel aridity, and percent changes in burn probability (relative to the baseline scenario) under various climate change and atmospheric CO2 fertilization scenarios in the 2040s. Panels (a)–(d) show the distribution of burn probability against fuel load and fuel aridity with the baseline scenario as the reference. Other panels show bivariate effects of fuel load and fuel aridity on percentage changes in burn probability (climate change scenario minus baseline scenario). Data are binned with 0.05 window length for both fuel load and fuel aridity. The value is the median of percent changes within each bin. Panels (e)–(h) show the climate change effect and panels (i)–(l) show combined climate change and CO2 fertilization effects (i.e., the overall effect). Region A has low fuel load with a broader range of aridity, representing grasses, region B is dry with low fuel loading, where shrubs dominate, and region C is relatively mesic with high fuel loading, where forests dominate. For more information about vegetation distribution of each scenario see Figure S10 in Supporting Information S1.

Figure 6. Relationships among fuel load (i.e., litter carbon), fuel aridity, and percent changes in burn probability (relative to the baseline scenario) under various climate change and atmospheric CO2 fertilization scenarios in the 2040s. Panels (a)–(d) show the distribution of burn probability against fuel load and fuel aridity with the baseline scenario as the reference. Other panels show bivariate effects of fuel load and fuel aridity on percentage changes in burn probability (climate change scenario minus baseline scenario). Data are binned with 0.05 window length for both fuel load and fuel aridity. The value is the median of percent changes within each bin. Panels (e)–(h) show the climate change effect and panels (i)–(l) show combined climate change and CO2 fertilization effects (i.e., the overall effect). Region A has low fuel load with a broader range of aridity, representing grasses, region B is dry with low fuel loading, where shrubs dominate, and region C is relatively mesic with high fuel loading, where forests dominate. For more information about vegetation distribution of each scenario see Figure S10 in Supporting Information S1.

In the 2040s, climate change reduced P_burn (Equation 1) across the watershed by reducing fuel loading. However, increasing CO₂ tempered that effect, and in many locations even changed its direction (except for in region B, which is shrub-dominated). The counteracting effects of climate change and CO₂ fertilization depended on location and vegetation cover type. Region B (shrub-dominated) had the largest decreases in P_burn due to climate change (except under the ProVeg storyline). This occurred because warming reduced shrub productivity (Figures 2 and 8d) and enhanced litter decomposition, which synergistically reduced fuel loading (Figure S2 in Supporting Information S1).

In the 2070s, changes in P_burn were more consistent across storylines and fuel conditions. With climate change, the fuel conditions in regions B (shrub-dominated) and C (evergreen forest) converged with region A (grass-dominated), where the watershed as a whole became intensely fuel-limited (caused by climate change-driven reductions in vegetation growth and increases in litter decomposition). The CO₂ fertilization effect tempered the magnitude of decreasing fuel loads in the 2070s but was not sufficient for changing the direction of P_burn, with a few exceptions (e.g., in regions A and C under the MultiMean climate change storyline, Figure 7i). However, in region B, where shrubs were distributed broadly across the whole basin (Figure 2), the dominant mechanism differed among climate change storylines.

View Image - Figure 7. Relationships among fuel load (i.e., litter carbon), fuel aridity, and percentage changes in burn probability (relative to baseline) under various climate change and atmospheric CO2 fertilization effect scenarios in the 2070s. Panels (a)–(d) show the distribution of burn probability against fuel load and fuel aridity for the baseline scenario (the same baseline as 2040s). Other panels show bivariate effects of fuel load and fuel aridity on percentage changes in burn probability (climate change scenario minus the baseline scenario). For more information about the vegetation distribution of each bin see Figure S11 in Supporting Information S1.

Figure 7. Relationships among fuel load (i.e., litter carbon), fuel aridity, and percentage changes in burn probability (relative to baseline) under various climate change and atmospheric CO2 fertilization effect scenarios in the 2070s. Panels (a)–(d) show the distribution of burn probability against fuel load and fuel aridity for the baseline scenario (the same baseline as 2040s). Other panels show bivariate effects of fuel load and fuel aridity on percentage changes in burn probability (climate change scenario minus the baseline scenario). For more information about the vegetation distribution of each bin see Figure S11 in Supporting Information S1.

In all cases, the effect of CO₂ fertilization had an upper limit (i.e., around 610 ppm, which was the concentration at the beginning of the assessment period for the 2070s) at which it could no longer counteract the effects of climate change. As the climate continued to warm and the fuel load continued to decrease, the system shifted from flammability limited to fuel limited and the dampening effect of CO₂ fertilization appeared to diminish.

To better understand why the ProDrought and ProVeg storylines sometimes exhibited different P_burn responses, we analyzed spatial distributions of P_burn, fuel aridity, fuel load, and NPP for the baseline and the 2040s scenarios (Figures 8 and 9). For the climate change only scenario, both ProDrought and ProVeg led to decreases in P_burn across the watershed, however this occurred for different reasons. In both storylines, fuel loading decreased, but fuel aridity increased under ProDrought, Figure 8b; and decreased under ProVeg, Figure 9b). Therefore, under ProVeg, decreases in both fuel loading and aridity compounded to decrease P_burn (Figure 9a). Under ProDrought on the other hand, decreases in fuel load and increases in fuel aridity counteracted one another; however, the effects of decreasing fuel loads outstripped increasing fuel aridity to decrease P_burn (Figure 8a). The differences we observed between the ProDrought and ProVeg storylines demonstrate how fire regimes can respond to the counteracting and/or compounding effects of exogenous (climate change vs. CO₂ fertilization) and endogenous (fuel load vs. fuel aridity) drivers.

View Image - Figure 8. The individual and combined effect of climate change and CO2 fertilization on burn probability, mean annual fuel aridity, mean annual fuel load (i.e., litter carbon), and mean annual NPP over the 30-year assessment period in the 2040s under the ProDrought climate change storyline (effects were calculated as the difference between the future scenario and the baseline). Red colors represent increases; blue colors represent decreases. Panels (a), (e), and (i) are changes in burn probability; panels (b), (f), and (j) are changes in fuel aridity; panels (c), (j), and (k) are changes in fuel load; and panels (d), (h), and (i) are changes in NPP.

Figure 8. The individual and combined effect of climate change and CO2 fertilization on burn probability, mean annual fuel aridity, mean annual fuel load (i.e., litter carbon), and mean annual NPP over the 30-year assessment period in the 2040s under the ProDrought climate change storyline (effects were calculated as the difference between the future scenario and the baseline). Red colors represent increases; blue colors represent decreases. Panels (a), (e), and (i) are changes in burn probability; panels (b), (f), and (j) are changes in fuel aridity; panels (c), (j), and (k) are changes in fuel load; and panels (d), (h), and (i) are changes in NPP.

View Image - Figure 9. The individual and combined effects of climate change and CO2 fertilization on burn probability, mean annual fuel aridity, mean annual fuel load (i.e., litter carbon), and mean annual NPP over the 30-year assessment period in the 2040s under the ProVeg climate change storyline. The red colors indicate increases; blue colors indicate decreases. Panels (a), (e), and (i) are changes in burn probability; panels (b), (f), and (j) are changes in fuel aridity; panels (c), (j), and (k) are changes in fuel load; and panels (d), (h), and (i) are changes in NPP.

Figure 9. The individual and combined effects of climate change and CO2 fertilization on burn probability, mean annual fuel aridity, mean annual fuel load (i.e., litter carbon), and mean annual NPP over the 30-year assessment period in the 2040s under the ProVeg climate change storyline. The red colors indicate increases; blue colors indicate decreases. Panels (a), (e), and (i) are changes in burn probability; panels (b), (f), and (j) are changes in fuel aridity; panels (c), (j), and (k) are changes in fuel load; and panels (d), (h), and (i) are changes in NPP.

Mechanisms Driving Fire Regime Changes in the Dry Future Storyline (e.g., ProDrought)

Under the climate change only scenario and ProDrought storyline, fuel loading and fuel aridity had counteracting effects on fire regimes and those effects varied among locations. What are the mechanisms that result in fuel load and fuel aridity competing with each other in the ProDrought storyline? Fuel Loading: warming increased NPP and fuel loading in the relatively mesic, forested areas and decreased NPP and fuel loading in water-limited areas (Figure 8d). However, at the same time, warming decreased fuel loading by increasing decomposition rates in both dry and mesic areas (Figure S5 in Supporting Information S1). Overall decreases in fuel loading suggest that decomposition dominates over the increase in NPP in forested areas. While some studies corroborate this finding in semiarid systems (Matthews et al., 2012), future decomposition rates are a key source of uncertainty in C cycling models (Luo et al., 2015; Tang & Riley, 2020). Given our current understanding and representation of decomposition under climate change, the net effect, therefore, is a decrease in fuel loading across the basin. Fuel aridity: warming increased fuel aridity across the basin because it increased PET but not AET; this occurred because the watershed is largely water-limited (Bradstock, 2010; Littell et al., 2016). Net Effect: the decrease in P_burn occurred because the decrease in fuel loading dominated over the increase in fuel aridity.

What are the mechanisms that result in climate change and CO₂ fertilization counteracting one another in the ProDrought storyline for the Overall Effect? Climate Change: this decreased P_burn for the reasons described above. CO₂ fertilization, on the other hand, increased NPP and fuel loading, especially in arid locations (Figure 8h) because the vegetation became more energy- and water-efficient under greater ambient CO₂ concentrations (Becklin et al., 2017). By reviewing evidence from different approaches (field observations and vegetation models), Lewis et al., 2009 also found rising CO₂ concentrations were the most likely driver of increased NPP. Net Effect: fuel loading decreased overall, suggesting that the climate change effect dominated over the CO₂ fertilization effect (Figure 8g). CO₂ fertilization also further increased fuel aridity by increasing LAI in water-limited areas (Figure 8f). This occurred because greater LAI can intensify the difference between PET and AET for water-limited ecosystems (Tague, 2009; Warren et al., 2011). With the overall effect, P_burn increased in the relatively mesic locations and decreased in the relatively arid locations (Figure 8i). This suggests that, in the relatively mesic locations that were historically flammability-limited, increases in fuel aridity can increase burn probability even while reduced fuel loading. However, in arid locations, burn probability was more sensitive to changes in fuel loading than fuel aridity (Figure 8i).

Mechanisms Driving Fire Regime Changes in the Mesic Future Storylines (e.g., ProVeg)

What are the mechanisms that result in climate change and CO₂ fertilization competing with each other in the ProVeg storyline for the Overall Effect? Climate Change: both fuel loading and fuel aridity decreased (Figures 9b and 9c). Fuel loading decreased in response to warming while increases in precipitation decreased fuel aridity (Figure S12 in Supporting Information S1). CO₂ Fertilization: as with all storylines, CO₂ fertilization increased fuel loading. However, it had a mixed effect on fuel aridity, with slight increases occurring in relatively arid locations and slight decreases occurring in the more mesic locations (Figure 9f). Net Effect: We observed a net increase in fuel loading, suggesting that CO₂ fertilization can dominate over climate change-induced drought under wetter future scenarios (Figure 9k). For fuel aridity, CO₂ fertilization only outstripped the climate change effect in very arid locations (at the lower part of the basin where CO₂ fertilization changed the fuel aridity response from a decrease to an increase, Figure 9j). P_burn increased across the entire basin (even with less arid fuel), indicating that elevated fuel loading was the main driver of P_burn for the overall effect, under the ProVeg storyline (Figure 9i).

Discussion

Vegetation productivity and litter decomposition are critical processes that integrate exogenous and endogenous drivers (e.g., climate and fuels) to shape wildfire regimes (fire size, burned area, and burn probability). For example, we found that in the mid-21st century (2040s), CO₂ fertilization increased NPP to a greater extent than it was reduced by climate change-induced drought, resulting in a net increase in fuel loading and resultant fire activity (fire size, burned area and burn probability). However, by the late-21st century (2070s), climatic warming (and associated drought) outstripped the effects of CO₂ fertilization, leading to a decrease in fire activity. The timing and magnitude of these trade-offs were modified by local aridity gradients and vegetation composition. The local-scale variability and the non-monotonic responses we observed suggest that linear or empirical extrapolation of fire regime (burned area, fire size, and burn probability) from the baseline and 2040s scenarios to the 2070s may not always be appropriate and depends on the degree of exogenous influence on the endogenous drivers of fire regimes (Littell et al., 2018; Westerling et al., 2011).

The Role of Exogenous Drivers in Future Fire Regimes The Role of Climate Change and CO₂ Fertilization in Influencing Future Fire Regimes

Exogenous drivers affect fire regimes through their effects on vegetation productivity, litter decomposition, and endogenous fire drivers (i.e., fuel loading and fuel aridity (Kennedy et al., 2021). Climate warming and CO₂ fertilization are two key exogenous drivers that can either counteract or compound one another to influence fire regimes. For example, given sufficient moisture, climate warming can increase vegetation growth and fuel loading by increasing photosynthetic activity (Kurz et al., 2008). However, warming can also accelerate litter decomposition, which can offset such increases (Bradstock, 2010; Keane et al., 1999; Kurz et al., 2008). We found that in Trail Creek, litter decomposition dominated over increases in vegetation productivity (Figures 8c and 9c).

Aridity and CO₂ fertilization can also compete to influence fuel and fire dynamics. In water-limited locations for example, climate change can actually decrease fuel loading and burn probability by promoting drought (Hanan et al., 2021). CO₂ fertilization can enhance vegetation productivity and resultant burn probability by enabling higher rates of photosynthesis when stomata are partially closed to reduce transpiration (Becklin et al., 2017; Lewis et al., 2009). Consequently, drought effects on plant productivity can be tempered by CO₂ fertilization, and the final fuel loading is determined by two competing mechanisms: climate change and CO₂ fertilization. We found that the ProVeg and ProDrought storylines led to different fuel loadings, which suggests precipitation patterns (and aridity) play a strong role in the relative balance between these mechanisms (Figures 8k and 9k).

Climate warming and CO₂ fertilization can also compound one another to influence fuel aridity. For example, in water-limited ecosystems, climate warming can increase the water budget deficit (i.e., PET–ET, Bradstock, 2010; Littell et al., 2016), while CO₂ fertilization can increase leaf area index and PET thus making plants more water stressed (Tague, 2009; Warren et al., 2011). However, this compounding effect is location-dependent and can be modified by precipitation patterns, which will be discussed in the next section (Figures 8j and 9j).

Our modeling results suggest that plant response to CO₂ fertilization is an important factor in predicting future fire regimes in fuel-limited watersheds such as Trail Creek. However, physiological responses to rising CO₂ can vary among species and over time and there is still a great deal of uncertainty in projecting how different species will respond to increasing atmospheric CO₂ (Becklin et al., 2017; Lewis et al., 2009; Norby et al., 2016; Warren et al., 2011). While RHESSys accounts for the direct effect of increasing water use efficiency on carbon assimilation and the resulting potential increases in growth and biomass (including LAI), other responses such as reduced stomatal conductance under elevated CO₂ (Swann et al., 2016) or changes in allocation or leaf physiology (Warren et al., 2011) are not currently included, and may dampen or alter projected effects of CO₂ fertilization.

For projecting future fire regimes, the role of CO₂ fertilization increasing productivity and corresponding fuel loading is more certain than potential offsetting mechanisms, such as decreases in leaf area and stomatal conductance. The leaf-level stomatal conductance response is not a robust predictor of transpiration at the canopy and stand levels in part because larger-scale responses are confounded with many other factors (Norby & Zak, 2011). For example, reduced stomatal conductance could decrease canopy transpiration and increase soil moisture, but CO₂ fertilization can also increase LAI thus counteracting the effect of reduced stomatal conductance (Franks et al., 2013). If our model framework had accounted for reductions in stomatal conductance that could occur under higher CO₂, this would likely have offset the effects of CO₂ fertilization, thus reducing fuel aridity in water-limited locations and possibly increasing it in mesic systems (Figures 8f and 9f). These offsets may be less significant in a fuel-limited system because such systems are more sensitive to changes in fuel loading.

Another source of uncertainty in modeling future fire regimes is how litter/fuel loading will change through time. Because litter stores are a major control on fire spread, modeling fire regimes requires high skill in representing both litterfall and decomposition. Decomposition in particular is complex and is affected by temperature, moisture, nitrogen, pH, and microbial dynamics (Lin & Webster, 2014). Although RHESSys accounts for many of these drivers, it necessarily includes some simplifications that may ignore important mechanisms in semiarid, fire-prone systems. For example, spatial partitioning of moisture and nitrogen in litter stores can accelerate or decelerate decomposition during drying and rewetting cycles (Birch, 1959). In addition, modeled decomposition rates can be highly sensitive to parameter and model structure uncertainties and this sensitivity increases with climate warming (Hanan et al., 2022). This can be problematic when projecting the future fuel loading under climate change, particularly over long timescales such as when extending projections from the 2040s to the 2070s.

The Role of Precipitation in Influencing Future Fire Regimes

Precipitation was also an important factor in predicting future fire regimes in our relatively fuel-limited watershed. At an annual timescale, increases in precipitation can decrease burn probability by increasing fuel moisture and decreasing the length of the fire season. However, over longer timescales, increased precipitation can also increase burn probability by increasing NPP and therefore fuel loading. Other studies have also found that precipitation is important in driving fuel loading and fire (Bradstock, 2010; Littell et al., 2016; Pausas & Paula, 2012). For example, in a drier future, there may be less fire due to decreases in fuel loading; whereas in a wetter future, there may be more fire due to increases in productivity (Figure 10, Halofsky et al., 2020; Littell et al., 2016). Williams et al. (2019) also argue that interannual variability (i.e., the sequencing of dry and wet extremes) is as important as changes in average precipitation.

View Image - Figure 10. Conceptual diagram illustrating how exogenous drivers, vegetation, biogeochemical processes, and endogenous drivers interact to influence burn probability. “+” indicates an increase (e.g., increases in productivity cause increases in fuel loading; “−” indicates a decrease (e.g., increases in litter decomposition lead to decreases in fuel loading). “Mesic” and “arid” indicate that the specified effect is location dependent and occurs in mesic or arid locations. Relative deficit is calculated as (1 – ET)/PET.

Figure 10. Conceptual diagram illustrating how exogenous drivers, vegetation, biogeochemical processes, and endogenous drivers interact to influence burn probability. “+” indicates an increase (e.g., increases in productivity cause increases in fuel loading; “−” indicates a decrease (e.g., increases in litter decomposition lead to decreases in fuel loading). “Mesic” and “arid” indicate that the specified effect is location dependent and occurs in mesic or arid locations. Relative deficit is calculated as (1 – ET)/PET.

Although our results illustrate the importance of precipitation in predicting future fire regimes, they did not cover all plausible futures. We found that ProVeg, which had a 6% increase in precipitation during the fire season, led to much greater increases in fire size and burned area in the 2040s than were observed in other storylines, which instead had decreases in precipitation (Figure S12 in Supporting Information S1). Under ProVeg, higher precipitation increased productivity and fuel loads, thereby increasing fire size and burned area (Figure 10; Kennedy et al., 2021). For Trail Creek, precipitation was as critical in driving future fire regimes as warming (e.g., Figure 8d vs. Figure 9d). Given that all 20 GCMs are considered plausible futures and we did not weight GCMs by skill, the range of future precipitation considered here is illustrative but not predictive (Figure S1 in Supporting Information S1). That is, since projections of water deficit in this region vary from around ∆10 mm to ∆160 mm and water deficit in our selected GCMs range from around ∆10 mm to ∆135 mm (Figure S1b in Supporting Information S1), our future fire regime projections represent a range of possible outcomes.

The Role of Endogenous Drivers in Future Fire Regimes

We found that increases in fuel aridity could increase burn probability (Figure 8) but these increases were tempered by decreases in fuel loading. Decreases in fuel loading can occur in response to drought-driven declines in NPP (as simulated in our model framework), or in response to plant functional type transitions, such as the displacement of forests and shrub lands by grasslands due to fire-climate feedbacks (e.g., Bowman et al., 2020).

Plant functional types (e.g., grasses, shrubs, and trees) are generally distributed along gradients in aridity and productivity leading to differences in their fire regimes (Bradstock, 2010; Littell et al., 2016; Williams et al., 2019). To examine the role of vegetation type in driving fuel conditions and future burn probability (P_burn), we calculated changes in P_burn in response to changes in fuel loading ( ${\Delta}$ fuel loading) and fuel aridity ( ${\Delta}$ fuel aridity) for each vegetation type (Figure 11). For all vegetation types, when fuel loading and aridity both increased, they compounded to increase P_burn. However, as fuel loading decreased, and fuel aridity increased, the dominant driver varied among vegetation types. We investigated this difference by drawing a line in each panel that separated negative and positive changes in P_burn. For conifer and shrub patches, the line had a negative slope ( ${\Delta}$ fuel loading: ${\Delta}$ fuel aridity is 1:1), which represents a threshold separating the negative and positive changes in P_burn. A sloped line indicates that both fuel load and fuel aridity drive P_burn. Therefore, conifer stands and shrubs in Trail Creek (region C in Figure 6) can be classified as co-limited by fuel and flammability. While many previous studies using empirical models suggest that forest fire activity (i.e., burned area and burn probability) will increase under future warming due to increases in fuel aridity (Bradstock, 2010), our process-based model results agree with conceptual frameworks and suggest that in semiarid forests, fire activity may ultimately decrease due to decreasing fuel loads (even without wildfire-driven self-limitation). Similarly, studies in other dry forests have found that burned area and burn probability are likely to decrease in the future due to drought (Halofsky et al., 2020; Littell et al., 2018). In grass-dominated locations, the line that separates P_burn increases or decreases is vertical, which suggests that these areas are fuel-limited and fuel load is the main driver of P_burn (i.e., P_burn is insensitive to changes in fuel aridity; Figures 11c and 11f). The lines separating the negative and positive changes in P_burn had the same pattern in the 2070s as in the 2040s, indicating that these thresholds represent a consistent characteristic of this watershed over time (assuming no vegetation type-conversion). In the 2070s, the cluster of patches moved toward lower fuel loading (to the left), and higher fuel aridity (upwards). Therefore, even though we observed a non-monotonic trend in fire response to climate change from the 2040s–2070s (Figures 4 and 5), here we found that the ${\Delta}$ fuel loading: ${\Delta}$ fuel aridity threshold, above which increases in fuel aridity dominate over decreases in fuel loading (and vice versa), was stationary, and this stationary threshold varied among vegetation types (pines and shrubs showed a similar pattern while grasses stood out as highly fuel-limited). Because the distribution of plant functional types is shifting in many locations due to climate change (Batllori et al., 2020), identifying the stationary threshold for new vegetation types may be useful for predicting future fire regimes.

$View Image - Figure 11. The percent changes in burn probability in response to changes in fuel loading and fuel aridity relative to the baseline. Panels (a)–(c) illustrate how burn probability changes in response to changes in fuel load and fuel aridity for 2040s. Panels (d)–(f) show responses for 2070s. Changes are binned by every 5 percent change, and changes in burn probability are the median of these bins (note that the data for these bins are drawn from all four storylines). We removed bins that had less than 100 observations. The panels are as follows: a and d: conifer; b and e: shrub; and c and f: grass. All these changes are between future climate change only and overall effect (combined effect of climate change and increasing CO2) scenarios and the baseline scenario. The red line is the threshold Δ ${\Delta}$ fuel loading:Δ ${\Delta}$ fuel aridity that separates negative and positive changes in burn probability.$

Figure 11. The percent changes in burn probability in response to changes in fuel loading and fuel aridity relative to the baseline. Panels (a)–(c) illustrate how burn probability changes in response to changes in fuel load and fuel aridity for 2040s. Panels (d)–(f) show responses for 2070s. Changes are binned by every 5 percent change, and changes in burn probability are the median of these bins (note that the data for these bins are drawn from all four storylines). We removed bins that had less than 100 observations. The panels are as follows: a and d: conifer; b and e: shrub; and c and f: grass. All these changes are between future climate change only and overall effect (combined effect of climate change and increasing CO2) scenarios and the baseline scenario. The red line is the threshold Δ ${\Delta}$ fuel loading:Δ ${\Delta}$ fuel aridity that separates negative and positive changes in burn probability.

The Role of Aridity Gradients in Influencing Future Fire Regimes

Interactions between exogenous and endogenous drivers can vary along aridity gradients (Figure 10). Although top-down climate warming can be the dominant driver of wildfire at large scales, local aridity (P/PET, Figure 2) also responds to bottom-up drivers such as topography and vegetation composition (Hanan et al., 2021; Littell et al., 2018). As a result, climate warming can have different effects on fuel loading in arid versus mesic locations (e.g., Figure 8d). In arid locations, warming can increase drought stress and reduce productivity and resultant fuel loading, while in more mesic locations warming can increase productivity by increasing photosynthetic activity. Similarly, the effects of CO₂ fertilization also differ between arid and mesic locations (e.g., Figure 9f). In water-limited locations, CO₂ fertilization can increase fuel aridity by increasing leaf area index and therefore water deficit (as explained in previous Section 4.1). In mesic locations, CO₂ fertilization is more likely to decrease fuel aridity by increasing ET, thereby decreasing water deficit (Becklin et al., 2017; Duursma et al., 2014; Warren et al., 2011). To further complicate matters, aridity gradients are not stationary under climate change. With warming, some historically mesic areas are likely to become increasingly arid (McKenzie & Littell, 2017). Thus, interactions between exogenous and endogenous drivers can vary over space and time. It is worth noting, however, that as the frequency and mean size of very large fires increases, single fires regularly burn across these gradients, further challenging the ability to model fire-climate-vegetation feedbacks that are already difficult to disentangle.

Study Limitations

A current limitation to the RHESSys-WMFire framework is that it does not simulate vegetation transitions. Under climate change, such transitions in Trail Creek are also important for studying the climate-vegetation-fire relationships (Littell et al., 2018; Wirth et al., 2008). For example, fire and warming can promote the growth of fire-tolerance species (e.g., grasses and certain shrubs), which can replace sagebrush and evergreen forest (Halofsky et al., 2020). With more intensive climate change in the 2070s, grasses and shrubs that were previously limited to lower elevations may begin to occupy higher elevations and replace evergreen forests (Bart et al., 2016; Collins & Roller, 2013; Goforth & Minnich, 2008). Because grasses and shrubs have different fire regimes than evergreen forest (Bradstock, 2010; Hessburg et al., 2005; Littell et al., 2018), further research is needed to understand how vegetation conversions will influence fire regimes in the future.

Another limitation of our study is that our projection of future fire regimes mainly focused on the high-end warming scenario of RCP8.5. However, to better-support management-related decision-making, it is important to select a range of robust future scenarios, such as including RCP4.5 in our analysis (Snover et al., 2013). To address this, we compared the differences in projected future fire regimes (burned area, fire size, number of fire starts) between the 2040s and 2070s (Figures S13, S14 and Text S3 in Supporting Information S1). We found out that changes in burned area were consistent between RCP4.5 and RCP8.5 in the 2040s, but diverged in the 2070s. That is, we found that in the 2040s burned area increased for both RCP4.5 and RCP8.5 scenarios (when considering the combined effects of climate change and increasing CO₂, Figure S13d in Supporting Information S1). However, in the 2070s burned area under RCP4.5 continued to increase, while under RCP8.5 it decreased (Figure S14d in Supporting Information S1). Although RCP4.5 scenarios do not change our conclusion that the linear extrapolation of fire regimes from the past to the future may not always be appropriate, RCP4.5 did change the 2070s fire regime response.

Conclusions

We found that climate change and CO₂ fertilization effects can counteract one another to alter fuel loads and hence burn probability in a semi-arid watershed. Climatic warming reduced fuel loading by decreasing vegetation productivity and increasing fuel decomposition rates, while CO₂ fertilization increased fuel loading by enhancing vegetation productivity. On the other hand, both climatic warming and CO₂ fertilization increased fuel aridity. In the 2040s, CO₂ fertilization outstripped climatic warming to increase burned area and burn probability; however, in the 2070s, climatic warming became so intense that the mitigating effects of CO₂ fertilization could no longer keep up and therefore, burned area and burn probability decreased. This non-monotonic response occurred because decreases in fuel loading dominated over the increases in fuel aridity. However, the decreases in fuel loading (caused by higher litter decomposition rates) outstripped changes in fire weather and caused a dramatic reduction in wildfire under the climate change in the far future (2070s)—this process needs to be explored further given the uncertainties in modeling future decomposition (Hanan et al., 2022).

Vegetation type is an important factor modifying how future fire regimes respond to tradeoffs between fuel loading and aridity. For example, we found that ${\Delta}$ fuel loading: ${\Delta}$ fuel aridity thresholds that determine whether fire regimes are influenced more by fuel loading or fuel aridity varied between grasses (in which fuel load is limiting) and conifer and shrub stands (where fuel load and fuel aridity can both be limiting depending on underlying aridity gradients). For all vegetation types, this threshold does not change between the 2040s and the 2070s, suggesting that thresholds are a stationary characteristic of a given vegetation type. This stationary threshold can be used as a tool to predict future fire regimes.

Given that changes in wildfire regimes can have catastrophic consequences for both human infrastructure and wildlands, identifying appropriate management actions to reduce vulnerability is a high priority. However, fire regimes are changing (Littell et al., 2018; Liu et al., 2013; Liu & Wimberly, 2016), and therefore, allocation of limited resources to reduce fire risks and societal vulnerability will need to take these changes into account. Our modeling approach demonstrates that fire regimes will likely change in different ways across watersheds, but there is still substantial uncertainty in the models. As scientists work to reduce key model uncertainties—including future precipitation patterns, future decomposition rates, and species-specific responses to CO₂ fertilization—future projections will continue to be refined, providing better support to wildfire management decisions.

Acknowledgments

This project was supported by the National Science Foundation of the United States under award numbers DMS-1520873 and DEB-1916658. William Burke and Tung Nguyen provided helpful support for setting up the RHESSys model. We thank Jan Boll, Nicholas Engdahl, Rebecca Gustine, and Ames Fowler for providing valuable suggestions on the manuscript. We are grateful to the editor and two anonymous reviewers who helped improve the quality of this paper.

Data Availability Statement

The coupled RHESSys-WMFire model code is available online at: https://doi.org/10.5281/zenodo.5594757 (Ren et al., 2021a). The data used in this study are available at: https://doi.org/10.17605/OSF.IO/AX45N (Ren et al., 2021b).

© 2022. This work is published under http://creativecommons.org/licenses/by-nc-nd/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.

Projecting Future Fire Regimes in a Semiarid Watershed of the Inland Northwestern United States: Interactions Among Climate Change, Vegetation Productivity, and Fuel Dynamics

Content area

Abstract

Full text