Compounded Heat and Fire Risk for Future U.S. Populations

: Climate change is increasing the risk of extreme events, resulting in social and economic challenges. I examined recent past (1971–2000), current and near future (2010–2039), and future (2040–2069) ﬁre and heat hazard combined with population growth by di ﬀ erent regions and residential densities (i.e., exurban low and high densities, suburban, and urban low and high densities). Regional values for extreme ﬁre weather days varied greatly. Temperature and number of extreme ﬁre weather days increased over time for all residential density categories, with the greatest increases in the exurban low-density category. The urban high-density category was about 0.8 to 1 ◦ C cooler than the urban low-density category. The areas of the urban and suburban density categories increased relative to the exurban low-density category. Holding climate change constant at 1970–2000 resulted in a temperature increase of 0.4 to 0.8 ◦ C by 2060, indicating future population increases in warmer areas. Overall, U.S. residents will experience greater exposure to ﬁre hazard and heat over time due to climate change, and compound risk emerges because ﬁre weather and heat are coupled and have e ﬀ ects across sectors. Movement to urban centers will help o ﬀ set exposure to ﬁre but not heat, because urban areas are heat islands; however, urban high-density areas had lower base temperatures, likely due to city locations along coastlines. This analysis provides a timely look at potential trends in ﬁre and heat risk by residential density classes due to the expansion and migration of US populations.


Introduction
Climate change has far-reaching implications for economic and social stability, including loss of services and security, affecting quality and quantity of lives. Climate change is expected to increase severe weather events, including heat waves, fire, flooding, hurricanes, and drought. At recent temperatures, Oppenheimer et al. [1] assigned moderate risk from extreme weather events, high level of risk beginning below 1 • C of warming, and severe and widespread impacts during the next few decades. Indeed, climate change may have already been increasing the number and severity of natural disasters during past decades. During the 1980s, 2.8 events occurred per year that each cost at least a billion dollars, for a total of $127.7B [2]. During the 1990s, 5.2 billion dollar events occurred per year, for a total of $456.7B. During the 2000s, 5.9 billion dollar events occurred per year, for a total of $510.3B. During the 2010s, 11.9 billion dollar events occurred per year, for a total of $802.2B. Additionally, the budget for fire suppression has increased during the past decades to billions of dollars [3].
Climate change is expected to cause migration from coastal areas of inundation and heat [4,5]. This would reverse migration during the past 50 or more years, after air conditioning became available, of movement by households and firms to locations that will experience seawater rise and heat waves (e.g., Florida and southern coastal locations; 5). However, not all 13.1 million people at risk of 1.8 m inundation [4] may be likely to migrate beyond a short distance within a county. Migration decisions

Methods
For ecoregions, I used ecological divisions, slightly modified by merging any spatially disjunct features into the surrounding region ( [21]; Figure 1). Although it is possible to adjust the 'eco' component of the ecoregion based on climate change projections [22], the regions are recognizable and cultural, for example, 'the South'. Therefore, I held the region boundaries constant in time, and then converted the shapefile to a raster with characteristics that matched the Integrated Climate and Land Use Scenarios file (e.g., 90 m resolution; ESRI, Redlands, CA, ArcMap v. 10  I selected a mean ensemble general circulation model and two general circulation models of HadGEM2-ES365, to match one of the general circulation models applied in Integrated Climate and Land Use Scenarios (ICLUS; see below) and MIROC-ESM-CHEM, which also was selected by Morefield et al. [13]. In general, no model is better or worse at representing observed spatial patterns of basic climate and hydrological variables, despite variability across most regions and seasons [23]. I additionally focused on the representative concentration pathway of RCP8.5, which best corresponds with current carbon dioxide concentration trends. If mitigation occurs, RCP8.5 may represent climate of a later interval. Nonetheless, I analyzed RCP4.5 for comparison purposes.
Integrated Climate and Land Use Scenarios (ICLUS; [20]) v2.1.1 land use projections provide population growth every 10 years by residential densities (Figure 1). The ICLUS v2.0 documentation describes the incorporation of climate change into the migration model, although models are available without climate change. Updates to ICLUS v2.1 include a switch to more recent demographic components of change (i.e., rates of fertility and mortality). Additionally, the amenity value of local climate (average precipitation and temperature for summer and winter) was permitted to influence migration patterns. I selected the RCP8.5 representative concentration pathways, which is the best match for current concentration trends. This narrative results in a U.S. population that exceeds 730 million by 2100 [6]. Hadley GEM2-ES and GISS-E2-R general circulation model options were available, but only the Hadley GEM2-ES matched as general circulation model available for fuel moisture data.
In ICLUS, urban residential densities are defined as at least 4 (3.95) housing units per hectare based on the U.S. Census Bureau definition of urban population of 1000 people per square mile. The urban high category at a threshold of at least 25 (24.7) units per hectare is based on typical densities for public transportation, whereas the suburban areas have a 1 unit per hectare threshold, which is the density for municipal sewer and water supply services. For what may be considered the wildland-urban interface, the exurban high category threshold is 0.25 units per hectare and the exurban low category threshold is 0.05 units per hectare.
Maximum temperature during summer (June, July, August) and extreme fire weather days during summer (days with 100-hour fuel moisture below the 3rd percentile) were based on downscaling of general circulation model outputs via the Multivariate Adaptive Constructed Analogs (MACA) method to 4-km resolution [18,19]. The ensemble mean consisted of 18 (extreme fire weather) or 20 (temperature) general circulation models for the recent past , present and near future (2010-2039), and the medium-range future (2040-2069).
I projected fuel moisture and temperature .tif files (comprised of summer fuel moisture and maximum temperature variables for two general circulation models and one ensemble of general circulation models for three climate intervals) to match details of the ICLUS data, including the Albers projection. I then applied zonal statistics (ESRI, ArcMap v. 10

Results
By region, temperature differences were generally consistent by latitude among general circulation models. From 1971-2000 to 2010-2039, the ensemble model increased slightly less than 2 • C, the MIROC model increased about 2 • C, and the Hadley model increased almost 2.5 • C (Table 1)   By residential density categories (Table 2; Figure 3), heat increased over time by about 3.5 • C to 6.5 • C by model, with greatest increases in the exurban low-density class. Approximately a 1 • C difference existed between the coolest residential density category of urban high-density and the warmest residential density category of urban low-density, which increased slightly over time from 0.8 to 1 • C. The urban high-density also had the least number of extreme fire weather days, although wildfires are not likely to occur in urban areas, which do not have continuous vegetation cover to spread fire. The exurban low-density category had the greatest number of extreme fire weather days by 2040-2069, but the differences among residential densities were slight except in the Hadley model, which projected 25 days for exurban low-density, 22 Table S1).     For area of residential categories, all areas increased over time (Table 2). Nonetheless, the percentage of total area (i.e., sum of the five categories) that each category represented over time changed, with high-density urban increasing from about 0.5% to 1.2%, low-density urban increasing from about 8% to 10.4%, suburban increasing from about 9.2% to 10.7%, high-density exurban held steady at about 27%, and high-density exurban decreasing from about 55% to 50.5%. The area of exurban residential density represents the wildland-urban interface, which has greater potential of wildfire. Results varied greatly by region (Supplementary Table S1).

Will United States Residents Move from Areas with Heat Risk into Areas with Fire Risk and are Some
Residential Densities More at Risk?
Both climate change and socioeconomic factors, which drive population growth and distributions and land use, are important determinants of risks [1]. Climate change greatly increased temperature (+3.3 to 6 • C) and number of extreme fire weather days (+3.5 to 20 days), but adaptive response by migration was small, in this model (6; Table 2). Indeed, modeled movement and population growth alone without climate change increased population exposure to heat slightly (+0.4 to 0.8 • C). The relative influence of heat and number of extreme fire weather days was much greater than population dynamics. Climate change risks also varied across regions (Supplementary Table S1).
Nonetheless, continued movement to and growth in urban centers will reduce wildfire exposure. The exurban low-density category had the greatest number of extreme fire weather days by 2040-2069, albeit the difference among residential density categories may be minor depending on the general circulation model. Even though number of extreme fire weather days is based on weather alone, residential categories impart another measure of fire risk. Despite exceptions, the exurban low-and high-density residential classes (i.e., WUI) are most at risk of fire due to the presence of continuous vegetation, which spreads fire, and human-caused ignitions (due to equipment use, smoking, campfires, railroads, arson, debris burning, children, fireworks, and power lines; [16]). Area of exurban low-density decreased over time (55% to 50.5%), as urbanization occurred. Regional variation in area of the exurban low-density category was great (Supplementary Table S1).
Urban areas have their own, warmer climate than surrounding areas, owing to absorption and release of heat by buildings, heat from energy use, and lack of surface moisture [12]. Due to the urban heat island effect, future population growth in urban centers will increase heat risk. The urban high-density residential category by location is coolest, probably because megacities tend to be located along coastlines or rivers, which moderate temperature (Table 2; Figure 3). Therefore, heat-related mortality may be greatest in the urban low-density residential category, which already is 0.8 • C warmer than the urban high-density residential category. However, it may be that densely populated cities make up for the difference in temperature due to a greater heat island effect.
Uncertainty in estimates originates from many sources, including climate general circulation models and concentration projections, downscaling, and future changes in population and land use in response to climate change. Great spatial heterogeneity occurs in projections (Figure 2), reducing the certainty of conclusions. Fire risk is likely to increase due to extreme fire weather alone, but fire risk is moderated by many factors. Models will improve with continued development.

Adaptation
Risk entails hazard (e.g., weather events), exposure to hazard, and vulnerability to harm from hazards. Risk perception is critical for the realization of vulnerabilities and development of adaptive actions to reduce the negative effects of climate change and help transition ecosystems and society to changing conditions [1]. For example, risk in the distant future may not prevail over immediate concerns or motivate outlays of resources to prepare for future risk, particularly when residents doubt both risk and effectiveness of mitigation actions [24]. Both failure to recognize change and deferral of action are able to occur because climate change is slow to develop. Early warning signs include extreme weather events, flooding, drought, wildfires, and record hottest years. However, shifting baselines in the perception of normal conditions may cause tolerance for progressive escalation of event intensity. Moreover, slow development of climate change allows deferral of action to pass to other generations. It is hoped that technology will avert climate change, but so far that is not in evidence.
Adaptation strategies can reduce vulnerability. For example, U.S. residents have been able to adapt well to heat due to air conditioning units, which, paradoxically, promote increased climate warming [25]. Planted trees and other vegetation can likewise help cool cities. Furthermore, switching to a more sedentary and indoor lifestyle prevents heat-related mortality. Crop workers, who are active during the summer and typically not U.S. citizens, are particularly vulnerable to heat stroke [26]. Heat contributes to mortality for people with cardiovascular and respiratory illnesses and the elderly, who are less able to respond physiologically to heat [27].
Residents and communities have been slower to employ less gratifying and more active adaptation and mitigation strategies to reduce vulnerability to fire and climate change. Many exurban residents, who live at the wildland-urban interface with exposure at the intersection of human-caused ignitions with vegetation, do not take actions to create defensible space on their land by reducing flammable vegetation and armoring houses against fire [28]. Trees are valuable features in moderation, but high densities around houses will encourage fires. Wildfire damage to most homes and structures is due to ignition after wind-borne embers land on a roof or wooden deck or blow in through vents [29]. Instead, homeowners rely on neighboring public or private landowners for fuel treatment [30], government emergency services to suppress wildfire, and insurance or federal disaster assistance to cover any property losses [31]. Programs such as Firewise, Wildfire Risk to Communities, and Ready, Set, Go! educate and encourage communities to maintain defensible space, develop wildfire plans, and prepare for evacuation.
Vulnerability to climate change can be exacerbated by increased population and land uses, whereas sustainable development may reduce vulnerability to climate change [1]. Planning guidelines that constrain wildland-urban interface sprawl will result in easier wildfire defense and evacuation of people from WUI communities. Disclosure of wildfire risk may reduce building and movement into fire-prone areas, affecting the spread of future housing development in high-risk locations [32].

Compound Risk
Climate change is affecting ecosystems and society in numerous ways, including increased extreme weather events. Hot temperatures may compromise health and outside activities of both work and leisure, such as farming or gardening and summer sports. Fire may cause road closures, breakdown of infrastructure networks such as electricity, water supply, and health and emergency services, disruption of livelihoods, injury, ill-health, or even mortality. Fire may lead to post-fire flooding and debris flow due to loss of vegetation and soil erosion, intensifying vulnerability. On top of heat, fire, and post-fire events, climate change will increase the likelihood of heavy precipitation and flooding, resulting in a cascade in which greater temperatures will produce more fires that will interact with floods. Urbanization in coastal or riverine cities will additionally increase flood exposure. Extreme events often interact because they are spatially and temporally dependent, leading to underestimation of risk [33,34].
Compound risk occurs due to synergistic interactions among hazards, resulting in greater risk than simple summation of independent events, particularly across sectors. For example, heat-related mortality may be partially related to effects of air quality on respiratory diseases [1]. Hotter weather contributes to greater ground level ozone (smog) in polluted areas, wildfire smoke, and production of pollen, exacerbating respiratory problems [1]. Indeed, Mora et al. [35] documented 467 pathways by which climate hazards such as warming, heatwaves, precipitation, drought, floods, fires, storms, sea-level rise, and changes in natural land cover and ocean chemistry affected human health, water, food, economy, infrastructure, and security.

Conclusions
Climate change will greatly increase temperature and number of extreme fire weather days by 2040-2069. The exurban density classes are most vulnerable to wildfire, due to the intersection of human-caused ignitions and vegetation, and the urban density classes are most vulnerable to heat, due to the urban heat island effect. Population movement and growth to urban centers will reduce the percentage area of exurban density classes, reducing exposure to wildfire, but movement and growth will increase exposure to heat, based on these models. Adaptation strategies can reduce risk. Risk is compounded by interactions among climate change effects, which may impact health, livelihoods, assets and homes, well-being and sense of place, and other sectors such as crop yields, food security, and water resources.