Wildfires are globally widespread and affect both ecosystem processes and distribution patterns of flora and fauna [1,2,3]. Investigators have characterized burning in different ways. Here we use the term fire activity to refer to the extent of burning, necessitated by the fact that some investigators report area burned [4], others report number of fires >200 ha [5], or >400 ha [6], or >5000 ha [7], the latter three apparently as surrogates for area burned; though in California number of fires >400 ha is only moderately correlated with annual area burned (r² = 0.44, p < 0.001 for the years 1963–2013, [8]). Although these studies demonstrate that the determinants of fire activity are multi-factorial, climate has long been considered to play a key role [9].

Since at least the beginning of the Holocene there has been documentation of a significant link between fires and climate [10], and drought has typically been associated with years having anomalously high area burned [11,12,13,14]. One of the earliest papers implicating potential global warming impacts on wildfires was the demonstration by Westerling et al. [6] that, since 1980, the number of moderate to large forest fires has increased in western US forests and is correlated with years having earlier snow melt, and by implication higher spring temperatures (see [5] for a somewhat expanded version). This association between high temperatures in certain seasons and extent of burning has been borne out by other historical studies in western US forests [4,15,16,17].

Beyond the direct effect of weather (primarily winds) during a fire event, climate primarily affects fires in two ways. Warm, dry weather during the months immediately preceding the fire season reduces moisture content of both live and dead fuels, increasing the likelihood of ignition and spread of fires. In contrast, high precipitation a year or two prior will increase the volume of herbaceous fuels, which later increases the probability of ignitions and fire spread [18]. Climate impacts on fuel moisture vs. fuel volume are of varying importance in different vegetation types.

For example, both closed canopy mixed conifer forests and ponderosa forests in the Sierra Nevada of California have high fire activity in drought years (Figure 1), indicating both fuel types respond to low fuel moisture. However, the prior year’s precipitation has no relationship with fires in the mixed conifer forests, where fire spread is dependent on litter. On the other hand, prior year’s precipitation is strongly correlated with fires in the more open ponderosa forests, where fire spread is dependent on dried herbaceous fuels that increase in volume during years of high rainfall. Not surprisingly, this effect of prior rainfall is generally only observed in vegetation types where fire spread is dependent on herbaceous fuels, primarily grasslands and savannas. Sometimes this effect is evident in shrublands that burn as active crown fires (i.e., fire is not spread by surface fuels), because on such landscapes fires ignite in associated herbaceous vegetation, which acts as a wick that spreads fire into dense shrublands. This relationship with prior-year rainfall is also seen in desert ecosystems where non-native grasses are now contributing to fire spread and increased occurrence of large fires [19], and in Sierra Nevada woodlands and savannas [4].

Temperature effects on fuel moisture act by reducing moisture of live and dead fuels. Westerling et al. [6] hypothesized that high spring temperatures contributed to a more rapid melting of the snowpack, causing a longer season of drier soil conditions. Drier soils limit water uptake and thus reduce live fuel moisture, but it is not clear how drier soils per se would impact dead fuel moisture. However, higher temperatures during spring and summer, although correlated with early snow melt, will reduce moisture of both live fuels and dead surface fuels by increasing evaporative demands during the dry season. Distinguishing between these two processes is difficult with the Westerling study [6] as it reports only the timing of peak river flow in associated watersheds as a surrogate for snowpack melt-off date, and by implication spring temperature. Littell et al. [15] concluded that snowpack melt-off was important as they found a negative relationship between winter precipitation and fire activity in the Sierra Nevada and Rocky Mountains. However, Medler et al. [20] measured areal and temporal extent of snow cover in the western US and found no relationship with area burned. Keeley and Syphard [4] found a negative correlation between spring snowpack depth and area burned in Sierra Nevada forests, but doubted there was a causal relationship since the climate variables most strongly tied to snowpack melt-off were not parameters most strongly affecting area burned. They hypothesized that although high spring temperatures did result in early melt of the snowpack, fires were more strongly controlled by direct effects of spring and summer temperatures on live and dead fuel moisture. Likewise, Abatzoglou and Kolden [16] found that although early snow melt-off did correlate with higher early season fire activity, it was only weakly associated with annual area burned and suggested that temperature and precipitation within the fire season were more important determinants of area burned in western US forests.

In summary, high temperatures and drought have substantial impacts on the extent of burning in the western US, and elsewhere in the world [21,22]. There is general agreement that climate operates through effects on fuel moisture and fuel volume, and the relative importance of each varies with vegetation type. It is now widely recognized that the relationship between climate and fire activity has important implications for future climate change impacts on fire regimes. Fire response to climatic variation is shown to be correlated at some level to global circulation patterns such as the Pacific Decadal Oscillation or El Nino-Southern Oscillation [19,23]. However, here we focus on more direct and local climate drivers of fire activity.

Global warming is clearly projected by countless climate-modeling studies that support a strong relationship between a largely anthropogenically caused rise in atmospheric CO₂ and annual global temperatures. Although the rate of future temperature rise will depend on lifestyle decisions, there is a consensus that annual temperatures will increase by 3–6 °C in the 21st century [24] and will not be easily reversible [25].

What does this mean in terms of future fire regimes? Modeling results generally predict increased fire frequency and fire severity for much of the globe [26,27,28,29]; however, these projections need to be viewed with caution as they are based on rather uncertain and spatially variable relationships between temperature and fire activity, and comprise ecosystems with very different fuel structures. In general, models that predict future fire regimes depend on relationships between annual temperature and fire activity that are based on historical studies correlating annual climate variation with fire activity, and often at very different spatial scales. In addition, in order to model future regimes certain simplifying assumptions are needed, such as fires are not limited by ignitions [30], a position that a number of fire scientists find unacceptable [31,32].

Historical studies of annual variation in climate and fire have revealed several important points that need to be understood in future modeling efforts. One is that the scale of focus has profound impacts on recognizing which climate parameters are most critical to fire activity. Even for a single region such as California, the resulting relations between climate and fire are rather different dependent on the size of the subregion under consideration (Table 1).

Secondly, climate models focus on annual temperature change, yet annual variation in area burned is not strongly tied to average annual temperature, but rather to temperatures in particular seasons [4,15,16]. Understanding how this seasonal relationship plays out relative to future climate models is important to future forecasts of fire regimes.

Thirdly, there is an implicit assumption in these analyses that there is a climate signal correlated with annual fire activity on all fire-prone landscapes, yet some fail to show a significant fire-climate relationship (Table 1), and likely are controlled by more direct anthropogenic impacts such as ignition patterns.

Broad regional scale analyses of fire and climate relationships for the western US are widely reported; however, because climates vary remarkably across this landscape it is difficult with this broad brush approach to sort out spatial from temporal variation. For example, considering all western US forests collectively, as the year-to-year temperature anomalies (i.e., the difference from the long-term average) during spring and summer increases, the number of fires over 200 ha increases [17]. It is tempting to interpret this pattern as reflecting temporal patterns of temperature anomalies and their effect on fire activity. However, analyses over such broad regions often reflect as much spatial variation in climate as temporal changes. Failing to differentiate between temporal and spatial variation can have profound effects on the resulting climate model for fire activity.

Using California as an example, consider US Forest Service (USFS) lands, which cover an extensive latitudinal range. Here temperatures are inversely correlated with latitude; r² = 0.68, 0.59, 0.69, 0.73, p < 0.001, for winter, spring, summer and autumn, respectively, on USFS lands in California [8]. In addition, annual area burned is inversely correlated with latitude; r² = 0.15, p < 0.001. Thus, before investigating temporal patterns of how fire activity varies with climate, we need to appreciate there is significant spatial variation as well as temporal variation. In our studies of fire-climate relationships in California we have investigated these relationships within climatically homogeneous divisions as defined by NOAA [33], focusing on the main fire-prone landscapes in the state. At this scale we find that fire-climate relationships are remarkably different from one climate division to another. For example, a 104 years of records reveal highly significant correlations between temperature and area burned in the Sierra Nevada (Figure 2a), but not in the southern part of the state (Figure 2b). In addition, where there are significant relationships (Figure 2a), winter and autumn temperatures are not associated with annual area burned, whereas spring and summer temperatures are.

Clearly spatial scale will greatly alter our choice of the best model for predicting future fire regimes under a changing climate. This in part accounts for the differing models in Table 1. Abatzoglou and Kolden [16] used large management units that cut across different climate regimes in California; for example, their region SO comprised such a large area that it included both the NOAA climate divisions of the Sierra Nevada and South Coast (Table 1), as well as the Central Coast, regions that have very different fire-climate relationships (Figure 2, Table 1). It is important if we are to make sound predictions about how global warming will impact fire regimes we work within climatically similar regions, e.g., the NOAA divisions shown in Table 1 as well as within similar vegetation types [4]. Indeed, Abatzolou et al. [34] contended that even a finer scale spatial resolution is needed to capture regional climate modes due to orgographic and coastal effects. There might also be a need for even finer spatial scales as mountainous landscape climates vary from one watershed to another; however, at some point this becomes counterproductive since large fires burn across broad landscapes with variable climates. In addition, we may be better served by focusing on more direct biophysical parameters such as evapotranspiration rather than temperature and precipitation as demonstrated by Abatzoglou and Kolden [16].

Spatial scale is also critically important in terms of making future fire regime projections. Indeed, in central California it appears the Sierra Nevada temperature change has been relatively unremarkable over the past century in comparison to the Central Valley [35]. Using broad regional fire-climate patterns [5,26], when it is apparent that there is marked sub-regional variation (e.g., Figure 2), may not lead to productive predictions.

These data support the following points: First, increasing annual temperatures in and of themselves do not predict higher fire activity, rather the effect is highly dependent on season, primarily spring and summer temperatures (Figure 2a). Secondly, it is important to focus climate-fire analyses on climatically homogeneous divisions and preferably separating montane forests from foothill savannas and shrublands, in large part because these generally have very different fuel structures that respond to different climate signals. Thirdly, not all parts of the landscape are equally sensitive to climate change. In the southern part of the state (Figure 2b), and in lower elevations (Cal Fire lands) of the Sierra Nevada [4] higher temperatures in any season are not reflected in greater fire activity. In other words, fire activity is apparently not strongly climate-limited. This is consistent with the conclusion that climate plays a larger role in dictating fire regimes in mesic than arid environments [36,37]. On these landscapes climatic conditions are suitable most years for massive wildfires (e.g., the coolest summer temperatures in the South Coast region (Figure 2b) are higher than the highest temperatures in the Sierra Neavada subregion (Figure 2a), so that other direct anthropogenic impacts may play a larger role; in other words, fire activity appears to be ignition-limited [38], which calls into question premises behind some modeling exercises. These nuances in fire-climate relationships are particularly important to consider because broad brush approaches to future fire regimes sometimes predict extraordinary increases in fire activity, e.g., [28], that may not be realistic.

Drought plays a key role in driving fire regimes and annual precipitation is a primary driver of drought variability [39]. In some parts of the western USA droughts have become more intense in recent decades, particularly so in southern California (Figure 3), though it is unclear if this is a trajectory for future drought. Future climate change projections of precipitation range between a 5% increase or decrease in precipitation by 2060 in California [40,41].

At present it is hard to distinguish current droughts from natural drought cycles, which complicates parsing out the influence of anthropogenic climate change [42]. Although natural drought cycles may dominate the current picture, it appears that anthropogenic-caused global warming is exacerbating the effects of drought; it is estimated to have contributed 5%–18% to the severity of one of the worst recent droughts in 20th-century California history [39]. This recent drought has been associated with severe die-off of Pinus ponderosa in mid-elevations of the Sierra Nevada (Figure 4). While we might expect future droughts on average to be more intense, it is open to debate as to whether or not droughts in the future will be worse than our worst droughts in the past.

It is unclear to what extent droughts, which may not be the result of anthropogenic warming, may have contributed to increased temperatures in recent decades. Droughts are associated with decreased cloud cover, which contributes to elevated temperatures [43]. This is not a recent global warming phenomenon as in California over the past 100 years there has been a negative relationship between spring temperatures and spring precipitation (r² = 0.28, p < 0.001) [8]. Thus, a recent trajectory of increased temperature may not be solely led by anthropogenically driven warming. This of course is not meant to downplay the importance of global warming, only to illustrate one more complexity in determining the appropriate fire-climate model for any given region (e.g., Table 1).

Another consideration is that, even if precipitation does not decline in the future, increased temperatures may exacerbate water deficit, and thus evaporative demand and plant water stress, which has been linked to forest mortality [44,45], and thus altered fuel conditions, as well as possibly the impact of fire severity [46]. Potential evapotranspiration will likely play a significant role in modeling future drought impacts [47].

One of the complications in making forecasts of future fire regimes is that the nature of fire-climate relationships can change over time. Forests in both the Sierra Nevada of California [4] and the northern Rocky Mountains [48] have very strong fire-climate relationships that span the 20th and 21st centuries. Of particular interest is the fact that the climate parameters driving high fire activity in the early 20th century are not always those driving current fire activity. For example, in Sierra Nevada forests from 1910 to 1959 the strongest variable controlling area burned was spring precipitation followed by winter precipitation (r² = 0.41, p < 0.001) [4]. However, from 1960 to 2010, mean summer temperature was the strongest variable followed by spring precipitation (r² = 0.53, p < 0.001). Recently we expanded this analysis from 1960 to 2013 (to include the massive Rim Fire that burned into Yosemite National Park), subsequently predicting area burned by spring temperature followed by summer temperature (r² = 0.52, p < 0.001) [8].

Another way of looking at this is to examine decadal patterns of burning and the associated temperature and precipitation anomalies. On USFS lands in the Sierra Nevada there were peaks in fire activity in 1920 and in 2000 and the first half of the 2010 decade (Figure 5a). While the recent peaks in fire activity are associated with positive anomalies in spring temperature, the 1920s peak was not (Figure 5b). On the other hand, there was a deficit in precipitation for all peaks in fire activity (Figure 5c).

These patterns are consistent with global warming expectations, but it also raises the possibility that future fire-climate relationships may change as well. Models of future fire regimes based on past fire-climate relationships may have limited predictive capacity. Of course it is appreciated that model development has to start somewhere and thus rudimentary modeling efforts have value in the development of a sound modeling framework; however, at the present time, such models may not be ready for directing future management decisions on fire-prone landscapes.

A further complication is that, while most predictions about future fire regimes are conditioned on relationships observed in the past, there will likely be novel combinations of temperature and precipitation, varying across different temporal and spatial scales, that have no analog in the past [49,50]. Predictions of climatic impacts to future fire regimes will need to account for these types of conditions that have no historical precedent. In addition, there is reason to believe that future impacts are dependent on the order, timing and magnitude of many contingencies [51]. This suggests that statistical models may be insufficient to guide future predictions and more mechanistic models will be required.

Anyone who says they have a grasp on how climate change will impact future fire regimes possesses an impressive level of optimism. It is doubtful that anyone in the early 20th century could have predicted future fire regimes and the same applies to the early 21st century. We now have a much improved grasp on the relative importance of climate across different ecosystems. Montane forested ecosystems far removed from direct human impact have historically been affected by climate much more so than lower elevation non-forested ecosystems. However, projecting future fire regimes requires a focus on expected seasonal changes in temperature. Regional analyses of fire climate relationships potentially mix sub-regions with different climate drivers and this may produce misleading forecasts of future fire regimes. In contrast to montane landscapes, in lower elevations and lower latitudes, fires are not strongly climate-limited but are ignition-limited, and on these landscapes increased population growth may ultimately be a more important factor driving future fire regimes than climate.