Results
Watersheds experienced a wide range of fire activity during the study period, with a median of 2 fires (mean = 3; range: 0-48). These fires resulted in a median of 3.6% (mean = 16%; range = 0-250%) of the flammable area burned. The median watershed had a pyrodiversity value of 0.04 (mean = 0.09; range = 0-0.35). Hotspots of pyrodiversity include watersheds in the North Cascades of Washington state, the Northern Rocky Mountains within and around the Frank Church-River of No Return Wilderness, Yellowstone National Park, the Mogollon Rim region of Arizona and New Mexico, and the mountainous regions of California especially the Klamath Mountains, and parts the Sierra Nevada (Fig. 2).
When including watersheds with no recent fire history, variation in burn frequency, patch size, and severity are highly correlated. However, when sequentially excluding areas with less active fire histories, theses correlations quickly dissipate. The correlation between frequency and patch size approximates 0.5 when considering watersheds with 14 or more fires since 1985. The frequency-severity correlation drops below 0.5 once watersheds with fewer than 8 fires are excluded. Patch size and severity plateau at approximately 0.65 when considering watersheds with 10 or more fires. Seasonality is largely uncorrelated with the other three fire regime traits, starting between 0.13 and 0.23 when watersheds with at least one fire are included, and dropping below or near zero when restricting correlations to areas with more active fire histories (Fig. 3).
Climate, topography, and human influence metrics show clear effects on proportion of flammable area burned between 1985 and 2018. Proportion wilderness followed by climatic variables showed the strongest relative effects. Proportion burned area increased with proportion wilderness with a scaled effect (\(\beta_{\text{wild}}\)) of 0.70 (90% confidence interval [CI] = 0.53, 0.86). Climatic water deficit (CWD) also had a strong positive effect (\(\beta_{\text{CWD}}\) = 0.53; CI = 0.47, 0.59), as did actual evapotranspiration (AET; \(\beta_{\text{AET}}\) = 0.10; CI = 0.048, 0.16), and the interaction of CWD and AET (\(\beta_{AET*CWD}\)= 0.27; CI = 0.23, 0.31). Both topographic roughness (\(\beta_{\text{rough}}\) = 0.15; CI = 0.11, 0.19) and elevation (\(\beta_{\text{elev}}\) = 0.084; CI = 0.011, 0.16) are positively associated with burn area, but these variables interact negatively (\(\beta_{elev*rough}\) = -0.10; CI = -0.14, -0.068). Human population density was negatively associated with proportion burned area with an effect estimate (\(\beta_{\text{pop.den}}\)) of -0.15 (CI = -0.183, -0.109) (Table S1).
When proportion of flammable area burned is included as a predictor of pyrodiversity it has by far the greatest effect, with much of the ultimate effects of climate, topography and human influence being mediated by this variable. The proportion burned area is strongly positively associated with pyrodiversity (\(\beta_{\text{prop.burn}}\) = 2.5; CI = 2.4, 2.5), with a negative quadratic term (\(\beta_{\text{prop.bur}n^{2}}\) = -0.78; CI = -0.80, -0.77). These parameter estimates indicate a pyrodiversity peak when an average of 63% (CI = 61%, 65%) of a watershed has burned between 1985 and 2018 (Table S1; Fig. 4a). This apparent maximum equates to a 53-year fire rotation (CI = 51, 54 years), a measure of the time required to burn an area equivalent to the size of a landscape (Heinselman 1973). In some cases, the combined direct and indirect effects on pyrodiversity are reinforcing (e.g. topography) while others dampen their ultimate influence (e.g. climate). For a given level of fire activity, pyrodiversity is negatively associated with CWD (\(\beta_{\text{CWD}}\)= -0.048; -0.068, -0.029) and AET (\(\beta_{\text{AET}}\) = -0.019; CI = -0.035, -0.003) with a positive interaction (\(\beta_{AET*CWD}\) = 0.014; CI = 0.001, 0.027) between the two climate variables. The combined marginal indirect and direct effects show CWD and AET interact to produce low pyrodiversity when watersheds lack an annual dry period but high pyrodiversity in productive areas coupled with dry periods (Fig. 4b). Similar to the burn activity model, elevation (\(\beta_{\text{elev}}\) = 0.035; CI = 0.013, 0.057) and topographic roughness (\(\beta_{\text{rough}}\) = 0.016; CI = 0.004, 0.028) are positively associated with pyrodiversity, with a likely slight negative interaction between the two (\(\beta_{elev*rough}\) = -0.013; CI = -0.026, 0.00). Consequently, pyrodiversity is maximized either at higher elevations or relatively low elevations with variable topography (Fig. 4c). When accounting for the level of burn activity, the direct effect of human population density on pyrodiversity is positive (\(\beta_{\text{pop.den}}\) = 0.029; CI = 0.017, 0.04) and proportion wilderness shows no clear direct effect (\(\beta_{\text{wild}}\) = 0.022; CI = -0.022, 0.066). Combined with a clearly positive indirect effect of proportion wilderness, and negative indirect effect of population density, watersheds in designated wilderness have higher pyrodiversity on average, while more populated areas have marginally lower pyrodiversity (Fig. 4d & e).