Of all nature's destructive forces, wildfire may be the one most intensified by human activity. Encroaching development in wildland areas and fuel buildup due to naive suppression are commonly understood to be exacerbating factors. It is also well known that human carelessness, accidents and damaged infrastructure cause 90% of all wildfires.
Other human disturbances to the environment further compound the wildfire hazard:
- roads and structures disrupt natural runoff and interfere with moisture absorption and retention
- roads, in particular, channel wind and help distribute embers over longer distances
- intrusion into plant and wildlife communities weakens overall health and resilience
- canopy openings in mature vegetation desiccate the natural understorey and promote more flammable ruderal growth
- misguided or accidental introduction of fire-adapted grasses can produce flammable surfaces that quickly spread fire
- vehicles, structures and propane are highly combustible wildfire fuels
Humans are undeniably responsible for a warming planet as well, which is spawning severe and prolonged drought in many areas of the world. Drought alone, however, is not the only adverse effect of climate change:
- hotter summers dry out vegetation faster and raise the ambient temperature of all fuels
- warmer winters produce less moisture from snowmelt during the summer months
- dry conditions persist later into the fall when wind events tend to be more extreme
- destructive insects and pathogens thrive due to the lack of prolonged sub-zero temperatures
- the atmosphere at the surface is drier due to reduced transpiration from stressed vegetation
- the atmosphere above can hold more moisture when fires do burn, leading to potential pyrocumulus collapse
- a slow or precluded recovery of established, healthy vegetation can produce a more vulnerable landscape
Unfortunately, most efforts to model wildfire potential fail to account for many of these factors and additionally suffer from one or more of the following problems:
- a naive view of environmental complexity
- a lack of statistical correlation with actual events
- an over-reliance on conventional fire behavior calculations
- a predisposition to produce a static result
Another source of misunderstanding is the common use of risk to refer to an inherent hazard.
These are very different concepts.
Merriam-Webster's first definitions of these terms are clear and succinct:
hazard ...is "a source of danger"
risk ...is "exposure to the consequences of uncertainty"
Wildfire is clearly a hazard. Being exposed to it is the risk and the quantification of that risk is a function of an asset's value and vulnerability with respect to the danger it faces. When viewed through that clarifying lens, it becomes obvious that risk mitigation is asset-centric and that the hazard of wildfire itself is virtually impossible to reduce. It is even more important to realize that a hazard can fluctuate over time whereas risk accrues over that same period. This makes it doubly important to model for specific forecast periods in order to correctly evaluate long-term risk.
The occurrence and growth of Wildfires is driven by 4 multi-faceted elements:
- Terrain topographic and geologic attributes of the landscape
- Land Cover composition and condition of natural, agrarian and constructed surfaces
- Seasonality phenology and annual patterns of atmospheric and hydrologic behavior
- Weather day-to-day variation in temperature, humidity, precipitation and wind
Only the terrain factors are static. Land cover changes at a relatively low frequency but that variation is still relevant to long-term predictions. Climatic changes occur more frequently and are exhibiting a progressive trend. Weather is the most frequent and most inconsistent pattern of change and it has, by far, the most significant impact on wildfire behavior.
It is interesting to note that there is a causal progression here and well as a temporal one. Terrain drives land cover by segregating plant alliances, convoluting roads and consolidating or isolating human habitation. Dramatic features also invite human investment and insurmountable ones discourage it. Land cover, in turn, contributes to seasonal variation through water re-distribution, transpiration and temperature modification. And the seasons clearly affect the variation and volatility of weather.
That's where simple relationships end. We are just beginning to understand that wildfire is truly chaotic. It conforms to the classic definition of chaos which what occurs when even a small changes in initial conditions suddenly produce drastic and irreversible disorder. Such events are characterized by unpredictability, turbulence and feedback within an infinitely complex system.
Assessing the magnitude of the wildfire hazard across disparate space and over long periods of future time is profoundly difficult. It is tempting to address the inevitable uncertainty of this problem by incorporating more input parameters, collecting them with greater precision and relying on "state of the art" algorithms to simulate fire behavior. This approach, however, is not probabilistic. It will always provide a single output for any given combination of inputs and any deficiency in logic will simply be replicated everywhere.
Consider the conventional algorithms used to predict fire behavior. They were developed as a way to quickly assess the potential impact of an active incident within a bounded environment and under a specific set of weather conditions. The result might be reasonably accurate for that particular case but it can never fairly represent a fluctuating hazard during future months or years. This shortcoming is often "accommodated" by using worse-case weather based on a 97th percentile value for all parameters. Lost are the all-important patterns associated with those values are they extreme outliers or the normal condition? Vegetative fuels are classified by their ligneous and foliar properties with no regard to plant alliances, their adaptation to fire or innate flammability. Non-natural fuels are typically ignored. It should also be noted that the logic behind these algorithms was derived in the 1970s when computational power was extremely limited and there was very little awareness of climate change.
Wildfire is simply not the deterministic result of any set of inputs. Even machine learning is not well-suited to this type of problem because it would need to be trained to recognize the probability of a false positive and it is that very probability that is so elusive.
The principle of parsimony (Occam's Razor) holds that the simplest explanation of a phenomenon is the one most likely to be correct. In the context of wildfire, this suggests that adopting simple proxies for a minimal set of inputs, and correlating those observations with actual wildfire events, might produce the best predictions possible. Correlation, however, cannot simply be input to output it needs to involve statistical comparisons of change over time. This means discarding conventional assumptions about cause and effect and letting wildfire speak for itself. Chaos theory might be a good way to listen!
TerraFire is a wildfire hazard model based on this alternative approach. It is focused on data simplicity, temporal predictions, statistical verification and continual updates. The goal is public awareness. The tragedy and destruction of wildfire will never be effectively reduced without a well-informed public and a shared concern beyond just first responders and property owners in the wildland interface. The potential for wildfire needs to be at the forefront of suburban planning, real estate development, insurance underwriting and legislative initiatives. We all need to understand this hazard before we can effectively manage the risk of confronting it.
SpherAware is currently devloping an experimental version of TerraFire for the western US. This effort is expected to produce a useful tool for raising public awareness regarding the realistic probability of future wildfires and where they are likely to occur. Integral to the design are the following features:
- Free low-resolution coverage everywhere with high-resolution focus areas for subscribers
- Regular updates to underlying meteorlogic, hydrologic and demographic parameters
- Pobabilistic hazard forecasts for 3-month, 1-year and 5-year intervals
- Annual analyses of forecast effacacy
More information will be posted here and elsewhere as progress allows.