Burn severity and fire spread models
This approach to characterizing burn severity and fire spread using the gradient boosting machine learning-based analysis demonstrates that there are quantifiable relationships between weather/climate variables and fire activity at GCNP. Four models were developed for burn severity and fire spread in ponderosa pine and mixed conifer ecosystems based on past fires at GCNP to assess the predictable components of weather forecasts on wildfire within park boundaries.
Over the past two decades, fire managers have been able to increase the number of opportunities to manage unplanned ignitions for multiple objectives (GCNP 2012), which has diversified flora and wildland habitat (Vankat 2011a; Vankat 2011b). Fire occurrences at GCNP have diversified fuel loadings and vegetation both spatially and temporally, resulting in a patchwork mosaic of forest structure and composition and a corresponding mosaic of fire severities (Fulé et al. 2003; Hoff et al. 2014). The unique topography and climate of the GCNP, along with a fire management program with a long history of managing wildfires for resource benefit, have resulted in a fire regime where large, uncharacteristic high-severity fires are comparatively rare within the park’s boundaries. Because of the unique landscape and fire program of GCNP, these models are specifically relevant to the park or other restored landscapes and may have limited application in altered dry-forest ecosystems across the Southwest. Yet, these results indicate that by focusing on predictable components of daily weather conditions, predictive models can be used to better anticipate burn severity and fire spread during advanced single or multi-day planning periods, which can help inform strategic and operational planning on a wildfire or prescribed fire.
Burn Severity and Fire Spread Prediction Tables
Using fire to meet resource objectives contains an inherent level of risk due to several unknowns and uncertainties in what the immediate future will bring. The prediction tables were created to characterize burn severity and fire spread associated with 24-hour weather conditions at GCNP as a reference on what to expect when managing a wildland fire incident or monitoring for changing conditions during a planned prescribed fire. Elevation, wind, and VPD were consistent, important predictors across each model while also being accessible for managers to obtain several days into the future during an incident for planning purposes. The combination of these three components allowed us to capture the expected effects of weather on burn severity and fire spread in dominant forest types at GCNP.
Furthermore, adaptive management that includes tracking of seasonal severity thresholds is a crucial part of the fire management program at GCNP for its continued growth and success as fire managers deal with increased complexity due to multiple fire entries and changes in climate (GCNP 2012). A discussion of current weather conditions and recent trends in comparison to historical conditions can give fire managers insight into the relative severity of fire weather during incident support. Contextualizing the effects of weather on fire behavior within recent and historic trends allows fire managers to develop strategic goals that maximize fire use during favorable weather conditions and seek to limit its application during times of unfavourability (Young et al. 2019). This will help enable fire managers to minimize unwanted high severity fire and large fire ‘runs’, in part by informing when to make proactive fire management decisions and/or include new suppression objectives and operations. However, it should be noted that these models did not directly account for the physical processes that influence fire on the ground, such as fuels arrangement, fine-scale topography, or prior management actions, etc. For this reason, fire managers should be aware that localized conditions not captured within this assessment can affect fire behavior.
Burn severity and fire spread in GCNP
Understanding the drivers of burn severity and fire spread resulting from changing weather conditions can assist fire managers in making informed choices when the effects to ecological and social values are important factors in planning and decision making. Expectedly, it can be a challenge to obtain or estimate weather conditions at the time and location of burning (Collins et al. 2007; Thompson and Spies 2009). However, the increasing availability of daily spatial weather data provides a consistent, albeit coarse, representation of fire weather conditions for burning days on a fire. Therefore, to best address weather and climate effects at the landscape scale, this approach captured the larger spatial extent of the weather and climate conditions consistently over all the fires burned in this analysis. With this approach, a 24-hour average value was captured for each of the climate variables, thereby determining potential landscape-scale thresholds using the Burn Severity and Fire Spread Prediction Tables and assisting managers work within the unique climate and topography of GCNP.
VPD and wind were consistently significant predictors of burn severity and fire spread in all of the models. For the burn severity models, VPD was found to be a strong driver and had a slightly greater influence than wind in determining predicted threshold between low, moderate, and high severity. VPD has been found to be highly correlated with burn severity and burned area, more so than temperature alone, and is gaining traction as an important metric for predicting potential fire conditions and management goals (Mueller et al. 2020; Williams et al. 2015; Young et al. 2020). In fact, although maximum and minimum temperature are often identified as important drivers of fire activity in other studies (e.g., Littell et al. 2009; Westerling 2006), only minimum temperature was found to be an important predictor in the ponderosa pine spread model in this analysis. Fire responds to the combined effects of temperature and moisture. Consequently, a full accounting of water balance is needed to capture the effects of fire on vegetation (Crockett and Westerling 2018; Mueller et al. 2020; Williams et al. 2015), which is singularly captured in VPD.
Importantly, VPD and wind showed limited cointegration with each other or other covariates in this analysis, thereby allowing each metric to provide unique information to the model predictions across changing weather conditions. A significant positive relationship between area burned (Williams et al. 2015) and fire severity (Mueller et al. 2020) with VPD has been found during the wildfire season in the Southwest. Others have found that high wind speeds coincident with periods of favorable fire weather conditions are positively correlated with higher burn severity (Stevens-Rumann et al. 2016). When coupled with high VPD (resulting from high temperatures and/or low humidity levels) windy conditions have the potential to increase overall fire activity quickly (Bessie and Johnson 1995; Crimmins 2006). Similarly, Srock et al. (2018) developed a fire weather index called the Hot-Dry-Windy Index (HDW) which is calculated by multiplying the maximum VPD by the maximum surface wind speed, with the intention of capturing fire management difficulty. The HDW intends to forecast a worst-case scenario of potential fire behavior in contrast to these models, which captures a complex historical relationship across a continuum of fire behavior. Importantly, the Burn Severity and Fire Spread Prediction Tables may have limited use during extreme wildfire events, particularly those driven by high fuel loadings.
Two other index variables, BI and ERC, were important controls in three of the models (BI: mixed conifer severity; ERC: ponderosa pine severity and spread). The fire-danger metric BI, which includes the spread component (forward speed of the head fire and is affected by wind) was not included in either of the spread models, however, ERC, the estimated potential available energy released per unit area in the flaming front of a fire, which does not include wind, was included in the ponderosa pine spread model. Wind by itself did have the highest relative influence in the mixed conifer spread model. Indexed variables have been found to provide a direct link between fuel flammability and weather conditions that increase fire behavior (Abatzoglou and Kolden 2013). Despite BI an ERC being commonly used measures for potential severity (ERC) and spread (BI), these results highlight other factors may hold more direct influence.
In both the ponderosa pine and mixed conifer models, as elevation increased, the threshold for increased severity and spread decreased. Topography influences the biophysical gradients that affect the spatial distribution of fuels at the fine spatial scale at which fire interacts with the landscape while also altering those fuels’ availability for consumption (Holden et al., 2009). Accordingly, the prediction tables capture changes in fire behavior at different elevations but also likely captures changes in vegetation density and continuity. As follows, if fire is ignited, more dense forests typically burn with higher severities and with larger patches even when VPD or winds are marginally lower. The combined effect of windy conditions and an abundance of dry available fuels can increase fire spread (Rothermel 1972; Westerling et al. 2003; Cruz et al. 2022), while extended periods of high VPD depletes moisture from live vegetation, dead fuels, and soils via evapotranspiration, thereby increasing flammability (Karpius et al. 2003; Mueller et al. 2020).
Similar to other studies (Birch et al., 2015; Scott and Reinhardt, 2001; Stevens-Rumann et al., 2016), other topographic variables in addition to elevation were found to heavily influence fire behavior, such as an increase in slope gradient increasing burn severity. Although mean slope was not used as a variable in the prediction tables, it was found to be an important predictor in the spread models and could serve as an additional dimension for further examination due to its limited cointegration with wind and VPD. Furthermore, the topographic roughness index was the most influential predictor in the mixed conifer severity model, and a similar metric, the topographic radiation aspect index was highly influential in the ponderosa pine spread model. This is in alignment with Dillon et al. (2011) who found that topography, including elevation, slope/aspect, position, and complexity variables, had the strongest influence on predicting areas of high burn severity across the Southwest US, regardless of the differences between vegetation and land use. Topography further affects local wind and weather patterns which are poorly represented in our temporally and spatially coarse weather and climate variables.
Finally, as seen in the Burn Severity and Fire Spread Prediction Tables, the window for moderate severity fire was relatively small in ponderosa pine forests, when compared to mixed conifer forest. The moderate-severity category includes a wide range of fire effects and can be particularly difficult to define (Collins et al. 2018; Lydersen et al. 2016a); however, there is a growing interest in managing fire within this window to meet forest restoration and management objectives (Huffman et al. 2017). Research suggests that relatively frequent low- to more moderate-severity fire reduce the potential for uncharacteristically large, high-severity fire during subsequent burns (Collins et al. 2018; Harris and Taylor 2017; Larson et al. 2013; Lydersen et al. 2016b). By assessing trends in fire weather conditions and forecasting the potential for extreme fire behavior, these tables will facilitate informed discussions and decisions when deciding where and when to use fire around severity transition zones.
Implications
Land managers and other stakeholders have a desire to restore degraded dry forests to more historical and resilient conditions as well as protect these ecosystems from uncharacteristic wildfire. Managing wildfires for resource benefit is a flexible management tool for treating large landscapes. Fire-reduction surrogates such as mechanical thinning and low-severity prescribed fire, though attractive to land managers that work within fire risk-averse landscapes, can be time-consuming and expensive to implement (Stephens et al. 2012). Furthermore, NPS policy directs land managers to maintain “natural environments evolving through natural processes minimally influenced by human actions” (National Park Service 1988). The maintenance of natural conditions generally precludes mechanical forestry practices such as thinning or tree removal and is overall seen as socially unacceptable within park boundaries. Therefore, managing wildfires for resource benefit can achieve fuels reduction and typically a level of restoration that points a landscape towards a desired future condition that is more resilient with relatively little investment of time, money, or direct human influence (Young et al. 2022).
Furthermore, given concerns related to climate change and the projected increase in associated wildfire activity, continuing to preserve dry forest ecosystems in the Southwest is tenuous under more severe projections of future climate. However, management practices that reduce fuels and reintroduce frequent, lower severity fire regimes have been shown to prolong the persistence of ponderosa pine and mixed-conifer forest types (Loehman et al. 2018; O’Donnell et al. 2018). Specifically, managing wildfires for resource benefit in these forest types may better satisfy restoration objectives and increase ecological resilience if managed to allow more moderate severity fire to occur (Huffman et al. 2017). This analysis and the resulting Prediction Tables presents a concept that could be extended to similar models with consistent hourly or burning period weather data in the future. Similar models could be developed that account for the anticipated range of future climate/weather conditions, including climate change projections.
However, in order to increase the accuracy and utility of these and future climate-fire models, improved quantification of the weather, climate, and the biophysical controls of burn severity and fire spread on burning days is needed, especially for better predictions of the likelihood and intensity of more extreme fire events or in identifying potential tipping points. Remote sensing approaches will be required to inform future landscape-scale fire planning most efficiently. Therefore, we encourage the continued tracking of fire and weather variables during active fire management in forested ecosystems to help advance scientific activities that improve the wildfire decision-making process in the face of rapid climate change.