skip to main content

Explore Research Products in the NSF-PAR

Title: Machine learning to predict final fire size at the time of ignition
Fires in boreal forests of Alaska are changing, threatening human health and ecosystems. Given expected increases in fire activity with climate warming, insight into the controls on fire size from the time of ignition is necessary. Such insight may be increasingly useful for fire management, especially in cases where many ignitions occur in a short time period. Here we investigated the controls and predictability of final fire size at the time of ignition. Using decision trees, we show that ignitions can be classified as leading to small, medium or large fires with 50.4±5.2% accuracy. This was accomplished using two variables: vapour pressure deficit and the fraction of spruce cover near the ignition point. The model predicted that 40% of ignitions would lead to large fires, and those ultimately accounted for 75% of the total burned area. Other machine learning classification algorithms, including random forests and multi-layer perceptrons, were tested but did not outperform the simpler decision tree model. Applying the model to areas with intensive human management resulted in overprediction of large fires, as expected. This type of simple classification system could offer insight into optimal resource allocation, helping to maintain a historical fire regime and protect Alaskan ecosystems.  more » « less
Award ID(s):
1633631 1839336
NSF-PAR ID:
10182754
Author(s) / Creator(s):
; ; ; ; ;
Date Published:
Journal Name:
International Journal of Wildland Fire
Volume:
28
Issue:
11
ISSN:
1049-8001
Page Range / eLocation ID:
861
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ( , Fire)
    In recent decades, wildfires in many areas of the United States (U.S.) have become larger and more frequent with increasing anthropogenic pressure, including interactions between climate, land-use change, and human ignitions. We aimed to characterize the spatiotemporal patterns of contemporary fire characteristics across the contiguous United States (CONUS). We derived fire variables based on frequency, fire radiative power (FRP), event size, burned area, and season length from satellite-derived fire products and a government records database on a 50 km grid (1984–2020). We used k-means clustering to create a hierarchical classification scheme of areas with relatively homogeneous fire characteristics, or modern ‘pyromes,’ and report on the model with eight major pyromes. Human ignition pressure provides a key explanation for the East-West patterns of fire characteristics. Human-dominated pyromes (85% mean anthropogenic ignitions), with moderate fire size, area burned, and intensity, covered 59% of CONUS, primarily in the East and East Central. Physically dominated pyromes (47% mean anthropogenic ignitions) characterized by relatively large (average 439 mean annual ha per 50 km pixel) and intense (average 75 mean annual megawatts/pixel) fires occurred in 14% of CONUS, primarily in the West and West Central. The percent of anthropogenic ignitions increased over time in all pyromes (0.5–1.7% annually). Higher fire frequency was related to smaller events and lower FRP, and these relationships were moderated by vegetation, climate, and ignition type. Notably, a spatial mismatch between our derived modern pyromes and both ecoregions and historical fire regimes suggests other major drivers for modern U.S. fire patterns than vegetation-based classification systems. This effort to delineate modern U.S. pyromes based on fire observations provides a national-scale framework of contemporary fire regions and may help elucidate patterns of change in an uncertain future. 
    more » « less
  2. ; ; ( , Environmental Research Letters)
    Abstract

    Increasing fire activity and the associated degradation in air quality in the United States has been indirectly linked to human activity via climate change. In addition, direct attribution of fires to human activities may provide opportunities for near term smoke mitigation by focusing policy, management, and funding efforts on particular ignition sources. We analyze how fires associated with human ignitions (agricultural fires and human-initiated wildfires) impact fire particulate matter under 2.5µm (PM2.5) concentrations in the contiguous United States (CONUS) from 2003 to 2018. We find that these agricultural and human-initiated wildfires dominate fire PM2.5in both a high fire and human ignition year (2018) and low fire and human ignition year (2003). Smoke from these human levers also makes meaningful contributions to total PM2.5(∼5%–10% in 2003 and 2018). Across CONUS, these two human ignition processes account for more than 80% of the population-weighted exposure and premature deaths associated with fire PM2.5. These findings indicate that a large portion of the smoke exposure and impacts in CONUS are from fires ignited by human activities with large mitigation potential that could be the focus of future management choices and policymaking.

     
    more » « less
  3. ; ; ; ; ; ; ; ( , Remote Sensing)
    The size and frequency of wildland fires in the western United States have dramatically increased in recent years. On high-fire-risk days, a small fire ignition can rapidly grow and become out of control. Early detection of fire ignitions from initial smoke can assist the response to such fires before they become difficult to manage. Past deep learning approaches for wildfire smoke detection have suffered from small or unreliable datasets that make it difficult to extrapolate performance to real-world scenarios. In this work, we present the Fire Ignition Library (FIgLib), a publicly available dataset of nearly 25,000 labeled wildfire smoke images as seen from fixed-view cameras deployed in Southern California. We also introduce SmokeyNet, a novel deep learning architecture using spatiotemporal information from camera imagery for real-time wildfire smoke detection. When trained on the FIgLib dataset, SmokeyNet outperforms comparable baselines and rivals human performance. We hope that the availability of the FIgLib dataset and the SmokeyNet architecture will inspire further research into deep learning methods for wildfire smoke detection, leading to automated notification systems that reduce the time to wildfire response. 
    more » « less
  4. ; ; ; ; ; ; ; ( , Proceedings of the National Academy of Sciences)
    Fire is an important climate-driven disturbance in terrestrial ecosystems, also modulated by human ignitions or fire suppression. Changes in fire emissions can feed back on the global carbon cycle, but whether the trajectories of changing fire activity will exacerbate or attenuate climate change is poorly understood. Here, we quantify fire dynamics under historical and future climate and human demography using a coupled global climate–fire–carbon cycle model that emulates 34 individual Earth system models (ESMs). Results are compared with counterfactual worlds, one with a constant preindustrial fire regime and another without fire. Although uncertainty in projected fire effects is large and depends on ESM, socioeconomic trajectory, and emissions scenario, we find that changes in human demography tend to suppress global fire activity, keeping more carbon within terrestrial ecosystems and attenuating warming. Globally, changes in fire have acted to warm climate throughout most of the 20th century. However, recent and predicted future reductions in fire activity may reverse this, enhancing land carbon uptake and corresponding to offsetting ∼5 to 10 y of global CO 2 emissions at today’s levels. This potentially reduces warming by up to 0.11 °C by 2100. We show that climate–carbon cycle feedbacks, as caused by changing fire regimes, are most effective at slowing global warming under lower emission scenarios. Our study highlights that ignitions and active and passive fire suppression can be as important in driving future fire regimes as changes in climate, although with some risk of more extreme fires regionally and with implications for other ecosystem functions in fire-dependent ecosystems. 
    more » « less
  5. ; ; ( , Natural Resource Modeling)
    Abstract

    The number of large‐scale, high‐severity forest fires occurring in the United States is increasing, as is the cost to suppress these fires. We use a technique developed by William Reed to incorporate the stochasticity of the time of a forest fire into our optimal control problem. Using this optimal control problem, we explore the trade‐offs between prevention management spending and suppression spending, along with the overall economic viability of prevention management spending. Our goal is to determine the optimal prevention management spending rate and the optimal suppression spending, which maximizes the expected value of a forest. We develop a parameter set reflecting the 2011 Las Conchas Fire and numerically solve our optimal control problem. Furthermore, we adapt this problem to simulate a sequence of fires and corresponding controls. Overall, our results support the conclusion that the prevention management efforts offset rising suppression costs and increase the value of a forest.

     Recommendations for Resource Managers

    Increasing wildfire size and increasing federal suppression costs have prompted investigations into alternative methods to help prevent and manage large wildfires.

    Fire prevention lowers the risk of experiencing large fire events, but investment in fire prevention is risky because its benefits are realized at an unknown time in the future.

    Results illustrate that there are real economic costs associated with using funding directed to fire prevention to fund immediate fire suppression.

    In our work with unknown fire sequences, we observe an 88% reduction in suppression spending on average with fire prevention, and a 55% reduction in spending overall.

     
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email