Evaluating long-term monitoring data for fire and resource management applications often differs from traditional experimental research designs. A major challenge is that management objectives are typically specified as a range of acceptable values, bounded by a lower and upper acceptable limit (e.g., to reduce fuel loading by 50 to 65

Dry forests of the western United States evolved with frequent, spatially extensive, low-severity fires, followed by a century of fire exclusion. Recently, these same forests have experienced large and increasingly severe fires causing widespread forest loss. To what extent do modern high-severity fires depart from pre-1900 fire regimes, and how do these departures affect the future persistence of dry-conifer forests? We address these questions using a multiscalar approach that combines the largest landscape-scale tree-ring fire-scar network in North America with neutral landscape model simulations in the Jemez Mountains of northern New Mexico. Fire scars recorded a continuous 400-y history (ca. 1500-1900 CE) of frequent, low-severity fires across individual sites, landscapes, and centuries. We evaluated whether forests exhibiting this record could plausibly persist if they had historically burned under modern patterns of high-severity fire. Across the 240,000-ha Jemez landscape, high-severity fires from 1995-2024 burned 32% of fire-scar sites and 25% of dry-conifer forests. Within the Las Conchas Fire perimeter-a 61,000 ha area reburned by multiple fires-66% of fire-scar sites and 71% of forests were burned at high severity, erasing multiple centuries of forest and fire history. Simulations indicate that the probability of the observed fire-scar network persisting under modern high-severity fire rates is near-zero (P < 0.001). Projecting these dynamics forward indicates rapid loss of both dry-conifer forests and tree-ring fire histories-50% of modern dry-conifer forest will remain by 2055, and 24% by 2115. Modern high-severity fire regimes appear incompatible with dry-conifer forest persistence.

Co-varying disturbance and environmental gradients can shape vegetation dynamics and increase the diversity of plant communities and their features. Pinyon–juniper woodlands are widespread in semi-arid climates of western North America, encompassing extensive environmental gradients, and a knowledge gap is how the diversity in features of these communities changes across co-varying gradients in fire history and soil. In pinyon–juniper communities spanning soil parent materials (basalt, limestone) and recent fire histories (0–4 prescribed fires or managed wildfires and 5–43 years since fire) in Grand Canyon-Parashant National Monument (Arizona, USA), we examined variation at 25 sites in three categories of plant community features including fuels, tree structure, and understory vegetation. Based on ordinations, canonical correlation analysis, and permutation tests, plant community features varied primarily with the number of fires, soil coarseness and chemistry, and additionally with tree structure for understory vegetation. Fire and soil variables accounted for 33% of the variance in fuels and tree structure, and together with tree structure, 56% of the variance in understories. The cover of the non-native annual Bromus tectorum was higher where fires had occurred more recently. In turn, B. tectorum was positively associated with the percentage of dead trees and negatively associated with native forb species richness. Based on a dendroecological analysis of 127 Pinus monophylla and Juniperus osteosperma trees, only 18% of trees presently around our study sites originated before the 1870s (Euro-American settlement) and <2% originated before the 1820s. Increasing contemporary fire activity facilitated by the National Park Service since the 1980s corresponded with increasing tree mortality and open-structured stands, apparently more closely resembling pre-settlement conditions. Using physical geography, such as soil parent material, as a landscape template shows promise for (i) incorporating diversity in long-term community change serving as a baseline for vegetation management, (ii) customizing applying treatments to unique conditions on different soil types, and (iii) benchmarking monitoring metrics of vegetation management effectiveness to levels scaled to biophysical variation across the landscape.

Burned area and proportion of high severity fire have been increasing in the western USA, and reducing wildfire severity with fuel treatments or other means is key for maintaining fire-prone dry forests and avoiding fire-catalyzed forest loss. Despite the unprecedented scope of firefighting operations in recent years, their contribution to patterns of wildfire severity is rarely quantified. Here we investigate how wildland fire suppression operations and past fire severity interacted to affect severity patterns of the northern third of the 374 000 ha Dixie Fire, the largest single fire in California history. We developed a map of the intensity and type of suppression operations and a statistical model of the Composite Burn Index (CBI) including weather, fuels, and terrain variables during the fire to quantify the importance of operations and prior fires on wildfire severity. Wildfire severity was estimated without operations and previous fires and then compared with modeled severity under observed conditions. Previous low and moderate-severity fire without operations decreased CBI by 38% and 19% respectively. Heavy operations and offensive firing in the footprint of past fires lowered fire severity even more compared to prior fire alone. Medium operations and defensive firing reduced but did not eliminate the moderating effects of past fires. This analysis demonstrates important interactions between suppression operations and previous burns that drive patterns of fire severity and vegetation dynamics in post-fire landscapes. Given the need to reduce wildfire severity to maintain forest resilience, particularly with a warming climate, increased attention to using operations and severity patterns of previous fires known to reduce wildfire severity in megafires are likely to increase forest resilience and improve ecological outcomes.

First posted April 17, 2023 For additional information, contact: Western Ecological Research CenterU.S. Geological Survey3020 State University Drive EastSacramento, California 95819 Mechanical thinning and prescribed fire are common mitigation treatments to reduce fire hazards. However, these treatments are infrequently applied together within national parks. The Northwest Gateway project at Lassen Volcanic National Park is an exception to this pattern. Various thinning prescriptions were applied to the project area in 2014, with a subset of the area prescribed burned in 2018 and 2019. To determine responses to these treatments, we analyzed forest structure and fuels data across a network of long-term monitoring plots measured before treatments and in multiple years following treatments. Additionally, we assessed patterns in individual tree growth from cores taken from 101 individual yellow pines (ponderosa pine, Pinus ponderosa Douglas ex Lawson & C. Lawson, and Jeffery pine, P. jeffreyi Grev. & Balf.) within the project unit.Basal area and stem density were reduced following thinning treatments for pole-sized (≤15-centimeter diameter at base height) and overstory trees (>15-centimeter diameter at base height), with sharper reductions in pole-sized trees. Proportional change in live basal area after thinning was highest for pole-size Abies, with more than 80-percent basal area and stem density removed on average. There were large reductions in pole-sized Pinus and Populus. However, Populus trees were not targeted for removal, suggesting that these trees died via other mechanisms. Thinning treatments also resulted in reductions in stand density index values and in surface fuel loading when followed by prescribed fire, particularly for small fuels size classes (such as litter/duff, 1-hour, and 10-hour fuels). Growth of individual residual yellow pine, measured in terms of annual basal area increment, indicated a strong growth release in the years following thinning treatments.Taken together, these results indicate that forest restoration treatments at the Northwest Gateway project area were effective at reducing stand density and encouraging growth of residual Pinus. Interestingly, our results also indicated that although thinning followed by prescribed fire was most effective at reducing surface fuel loads, harvest techniques such as whole tree yarding may effectively reduce the accumulation of post-treatment residual fuels, especially when combined with hand piling and other targeted treatments.

The reestablishment of natural fire regimes can have numerous benefits for forest ecosystems, including the restoration of stand structure through a reduction in tree densities and increased representation of large diameter trees. However, fire effects may depend on how departed the ecosystem is from its historical fire frequency. Red fir (Abies magnifica) forests occupy a broad geographic area across which historical fire return intervals and stand structures vary. Using historical stand inventory data from the Vegetation Type Mapping (VTM) project, we evaluated red fir forests in the Sierra Nevada in California and the Cascade-Klamath region of northwestern California and southern Oregon to determine how reintroduced fire effects vary regionally and if these differences are related to historical fire return intervals or structural conditions. We sampled a total of 29 overlapping fires and found that reestablishing fire in red fir forests consistently restored historical forest structure across a wide geographic range by reducing the density of small trees and maintaining large trees. However, the effect of fire was most evident in the Sierra Nevada where the percent difference in total tree density between unburned and burned plots was significantly greater (77% difference) than in the Cascade-Klamath (53% difference), and burned plots in the Sierra Nevada had significantly lower densities of both small (<30 cm dbh) and medium sized trees (30-60 cm dbh). These stronger fire effects may be related to greater departure from reference fire return intervals in the Sierra Nevada, as well as the region's warmer and drier conditions increasing the availability of fuel to burn and susceptibility of trees to fire-related mortality. We found that departure from reference fire return intervals followed a similar pattern to departure from historical tree density in both study regions. Unburned plots were 61% departed from reference fire return intervals in the Cascade-Klamath and 69% departed in the Sierra Nevada. In these same plots, departure from VTM total tree density estimates were 37% in the Cascade-Klamath and 44% in the Sierra Nevada. We suggest that incorporating historical references for structural conditions together with regional or local estimates of historical fire return intervals contributes to an improved understanding of how reference conditions varied at local and regional scales, and their importance in the restoration of fire-dependent forests.

Prior research suggests that Indigenous fire management buffers climate influences on wildfires, but it is unclear whether these benefits accrue across geographic scales. We use a network of 4824 fire-scarred trees in Southwest United States dry forests to analyze up to 400 years of fire-climate relationships at local, landscape, and regional scales for traditional territories of three different Indigenous cultures. Comparison of fire-year and prior climate conditions for periods of intensive cultural use and less-intensive use indicates that Indigenous fire management weakened fire-climate relationships at local and landscape scales. This effect did not scale up across the entire region because land use was spatially and temporally heterogeneous at that scale. Restoring or emulating Indigenous fire practices could buffer climate impacts at local scales but would need to be repeatedly implemented at broad scales for broader regional benefits.

A classical analysis for measurements from a paired design is the paired t tool, useful for both estimation via confidence intervals and testing. It is usual to assume the pairs are independent of one another. There is, however, correlation between data points within pairs; indeed, it is this correlation that gives the design its power as a statistical tool. This tool operates on increments of change (i.e., change measured by subtraction) within each pair by taking the differences, and analyzing the resulting single sample thereof, or by taking the difference in the two means. On the other hand, inference on relative change is often of interest with biological data. For example, the National Park Service implements controlled burns with management goals that are often stated on a percentage basis (e.g., aiming to reduce fuel load by 50%, or to limit postfire tree mortality to <20%). Relative change objectives are useful when a trend over time is more important than the specific current or future state (USDI National Park Service 2003). For example, a reduction in tree density of 40% may be more useful to resource managers than a decline to 500 individuals per hectare across multiple burn units (USDI National Park Service 2003). In monitoring a fish population from one time point to another, the relative increase (or decrease) in the population size may be of more interest than the increment of change. For example, many small-bodied fish may produce multiple cohorts of larvae in a summer, with some of the progeny reaching maturity by the end of their first summer. Thus, and depending on environmental conditions and prey availability both owing to variable levels of recruitment and natural mortality, population sizes from one year to the next can be highly variable (Herwig and Zimmer 2007). Consequently, relative increases across a longer time period are useful to draw inference to size and status of populations. In what follows, we will show that the ratio of means is often superior to a more traditional approach (mean of ratios), examine some theory behind the approach, and exemplify the method with several real examples. For purpose of studying relative change, we will here consider the ratio of condition b to condition a, without prescribing which order is best for a given real situation; that is, in fact, a story-telling choice on the part of the researcher. The two possible analogs to the aforementioned approach to incremental change are the mean of the individual ratios (MoR) and the ratio of the two means (RoM). For relative change, it does matter how one approaches the estimation. Unlike incremental change where the mean of differences is equal to the difference of means, the ratio of the means and the mean of the ratios will usually not yield the same answers, and these differences are in fact meaningful. Table 1 illustrates these points using a simple example between two hypothetical conditions a and b (n = 2). If asked, “On average, by how much did these individuals change?” the answer would be fourfold. On the other hand, the relative change over both of the sampling units is only a doubling (twofold). Manley et al. (2002) discuss choosing between RoM and MoR in their resource selection book. They note that the two statistics estimate different population parameters; RoM estimates whereas the average of the ratios estimates the population mean of all such ratios. However, and as illustrated in Table 1, and explained more fully below, the two in fact are not identical. The ratio of means appears in the sampling literature, albeit not by that name. Ratio estimation is one of several regression-based model-assisted methods for estimating the mean (or, if rescaled) total of some variable in a population. These methods require information, ideally for the whole population, of an auxiliary variable that is correlated with the variable of primary interest. See, for example, Section 6.2 of Schaffer et al. 2012. Lohr (1999) develops the RoM more fully, showing its use as a regression tool in its own right. Suppose the response variable (Y) is proportional to a predictor (X). then it is reasonable that (1) Y and X can only take non-negative values; (2) that a straight line relationship through the origin would be a good fit, and (3) that variation in Y also grows with increasing X values. Under those conditions, the RoM is a natural estimator for the slope. The basic ideas are presented in Section 3.1 (pages 60–73) of Lohr (1999), with model formalities addressed in Section 3.4 (pages 81–88). The sample RoM is slightly (downward) biased, although the bias is usually small and shrinks with increasing sample size. An estimator of the bias is presented on page 67. Investigation of the properties of the RoM of two normally distributed variables (with sufficient sample sizes, the two means will be so) is a classical problem in statistics, known as the Fieller-Creasy Problem; various modern approaches to that problem include Bayesian methods (e.g., Mendoza and Guitierrez-Pena 1999 and Kim et al. 2018) and constrained normal approximations (see Diaz-Frances and Sprott (2004) and Diaz-Frances and Rubio (2013). (The latter studies the ratio of two independent variables; in the current setting, they will be correlated.) The novelty in the current paper is our proposal that has a useful and largely unexploited role in estimation for paired data. The leading terms sum to unity, and so they qualify as weights in a weighted average. Thus, RoM can be seen as a weighted form of MoR. MoR will always be greater than or equal to RoM. They will be equal when the denominator terms are the same. It sometimes happens that the values in the denominator have small variation; in that case, MoR and RoM will yield similar answers numerically. By way of analogy, when distributions are approximately symmetric, the mean and the median may be similar numerically, but they are still conceptually meaningfully different. Here, we argue that the choice between MoR and RoM is similarly germane. Using the individual ratios is appropriate when change (on average) at the level of individuals is of interest; a ratio of the means wins when change at the population level is of interest. When the question of interest in many studies is with regard to change over the population, we argue that RoM will be the better choice. We illustrate the process of choosing among the classical paired t tool, MoR and RoM in Figure 1. When using individual ratios (MoR), a value of 0 in the denominator will yield an undefined ratio (or infinite). One can choose to replace the 0 with a judiciously chosen (the choice will affect the resulting estimate) small value (akin to what is sometimes done to accommodate a 0 when using logarithmic transformations) or drop that data point. Some studies do indeed focus on individual responses. For instance, in studies of changes in oxygen consumption rates in seals while diving as compared to resting (e.g., Webb et al. 1998, Green et al. 2007) measuring the change on a per-seal basis makes sense, and if relative change is of interest, then the mean of individual ratios is attractive. In studies of changes in percent body fat of hibernating animals (e.g., Buck and Barnes 1998, Kunz et al. 1998), answering the question of change on average of individuals also makes sense. When using the MoR, the central limit theorem applies, since the statistic being considered is the mean of individual values (ratios). For studies like these, the RoM does not answer a relevant question (e.g., the total sum of body fat across the entire bear population). On the other hand, when interest is in change at the population (rather than individual), the RoM comes into its own. A simple and elegant statistic, it comes with a few twists, as we now investigate. Analysis of the mean of individual ratios can proceed in the usual one-sample way; the t-distribution is viable given that the sample size is sufficiently large so that the distribution of the mean is approximately normal. However, those conditions do not hold for the RoM, for two reasons. First, the central limit theorem does not apply to the ratio of two means, except under certain constraints (e.g., Diaz-Frances and Rubio 2013). Second, the standard error (SE) formula does not always work well. The conventional advice is that the estimator works well when is estimated precisely (necessary since the derivation of the estimator treats it as essentially constant), typically described as having a coefficient of variation less than 0.1. Ironically, it is precisely when this is not true that choosing between RoM and MoR is critical numerically. Aside from the sometimes unstable behavior of the SE formula, the matter is further complicated by the fact that the central limit theorem does not routinely apply to the distribution of the ratio ; we cannot argue that it is approximately normal, hence the need for bootstrapping. As a nonparametric method for creating confidence intervals, bootstrapping was the creation of Efron (1979), but see Efron and Tibshirani (1993) for a thorough introduction. The simplest way to form bootstrap confidence intervals is the empirical method. Suppose one uses B = 1000 bootstrap replicates with intent to make a 95% confidence interval. Sort the 1000 bootstrapped estimates from smallest to largest. Then, 2.5% of the simulated distribution will be less than the 26th smallest value and 2.5% will be larger than the 97.5th smallest. Those two values are the endpoints of the interval. Currently the bca-bootstrap procedure is commonly used; bca stands for “bias-corrected and accelerated.” Many statistics are biased estimators of their relevant parameter (the usual Pearson product-moment correlation coefficient and the standard deviation are two oft-used examples). The bootstrap estimate of bias in a statistic is measured by the proportion of simulated estimates that are less than that in the original sample. An acceleration parameter is calculated that is proportional to the skewness in the bootstrap distribution. These two values adjust the endpoints of the interval. For instance, if the simulated distribution is skewed right, the CI is adjusted to the right. To illustrate the differences among the three statistical methodologies, we present three different examples below. In southern Arizona (USA), seven fish species are endemic to the Río Yaqui watershed and four remain at and around San Bernardino and Leslie Canyon National Wildlife Refuges (NWRs). The Yaqui catfish Ictalurus pricei is functionally extinct (Stewart et al. 2017a), the beautiful shiner Cyprinella formosa and Yaqui chub Gila purpurae are federally listed as threatened, and the Yaqui topminnow Poeciliopsis occidentalis sonoriensis is federally listed as endangered (USFWS 1994). Our focus is on the Yaqui chub. This species occupies more than 35 wetland ponds. From 2007 to 2012, Federal, State, and multiple conservation organizations annually monitored populations of these fish to insure their persistence and evaluate their recover (Stewart et al. 2017b, 2019). Of the 35 wetland ponds, 10 were selected and ranged in size from 28 to 8813 m2, and their depths ranged from 0.4 to ~4 m. Most ponds are clear and dominated by submergent aquatic vegetation (e.g., spiny naiad Najas minor, sago pondweed Potamogeton pectinatus, coontail Ceratopyllum demersum, and horned pondweed Zannichellia palustris). Typically, the number of traps used to survey each wetland pond ranged from 8 to 30 and was set following a proportional stratified random sampling design based on vegetation and depth surveys conducted 1 wk prior to sampling (Stewart et al. 2017b). For example, if 50% of the pond was dominated by a certain submergent aquatic vegetation type, then 50% of the traps were randomly set in vegetation and 50% randomly outside the vegetation (Stewart et al. 2017b). We calculated the total number of fish trapped from the historical survey data set for Yaqui chub by pond and sample year, to determine population trajectories from 2007 to 2012, on a relative basis (by choice), using the MoR and RoM statistics (Table 2). Based on the Yaqui chub data, if we used the mean of individual ratios, data from pond 3 must be excluded due to the observed 0 in 2007. After discarding pond 3, and based on n = 9, the MoR is 56.8, and a t-based 95% CI is (−40.8, 154.4). The Yaqui chub data are so heavily skewed that the assumption of approximate normality is likely invalid; that and the fact that the resulting interval has a negative value for the lower endpoint also makes it clear that the t-based inference here is not valid. A bca-bootstrapped interval for these data is 2.0, 125. If inference is to be made to the population of ponds rather than average behavior of individual ponds, then the ratio of the means is the more appropriate statistic. Here, we use data for all 10 ponds. The RoM is 2.5, and a 95% bca-bootstrapped CI is 1.2, 6.9. This example illustrates how different the two estimates can be. One suggests an almost 60-fold increase in fish abundance. The other more realistically suggests a 2.5-fold increase. The reason for the dramatic difference can be traced to the ponds with only a very small number of fish detected in 2007 (Table 2). Thus, if the objective is to infer overall species population status and trend through time at the species and NWR level, then the RoM is the more appropriate statistic. The National Park Service (NPS) fire management program conducts systematic monitoring of prescribed burns to determine whether management goals are being achieved. Within each park, a series of fire management objectives are established for each major vegetation community, referred to hereafter as monitoring types. Fire ecologists typically use random sampling within each monitoring type to collect pre- and postburn data at regular time intervals. In this area of work, the choice to focus on relative change often rises to the level of routine policy. For example, an important metric used to predict fire intensity and potential fire impacts is the surface fuel loading, defined as the total amount of dead biomass (wood, duff, and litter) on the forest floor available for consumption (measured in dry weight as tons per acre). A common management objective listed in Fire Management Plans is to reduce the total surface fuel load in a given monitoring type by some designated percentage. The data for Example 2 show surface fuel loading values for 11 random plots established in the mixed conifer monitoring type before (pre) and 1-yr after (post) a prescribed burn in 2008 in Lassen Volcanic National Park (northern California, USA) (Table 3). The park's programmatic management objective for this monitoring type is to reduce surface fuel load by 50%–65% one year after burning. Since inference is to be made to the entire population in the burn unit as opposed to estimating how, on average, individual plots behaved, estimation using the means is preferred. The RoM is 0.42 with a 95% bca-bootstrapped CI of (0.27, 0.55), indicating that the postfire fuel load was 42% of the prefire fuel load. To interpret the RoM in the context of the management objective, we subtract the RoM from 1 (and rescale) to show the percent reduction in fuel loading from prefire levels: with a 95% CI of (45%, 73%). Thus, we can conclude that the burn reduced fuel loading by 58%, and the management objective of reducing fuel load by 50%–65% was achieved. For sake of comparison, the analogous result using the MoR shows an estimated fuel load reduction of 66%, CI of (50%, 78%), slightly greater than the upper management objective range (although not statistically significant). The MoR suggests a greater overall reduction than the RoM because plots with relatively low preburn fuel loads experienced a larger percent reduction than plots with higher fuel loads; these plots receive a lower weight in the RoM because they contribute less to the total loading of the burn unit (population). This is not an unusual occurrence in spring burns because areas with open forest canopies (and often lower corresponding surface fuel loads) can dry quicker than areas with more dense canopies (and often higher corresponding fuel loads). Thus, the burn may consume a greater percentage of biomass in those open forests, even though more total tons per acre of fuels are being consumed in the denser forests. The RoM better reflects the population fuel load across the unit. This example illustrates that the two methods not only usually produce different estimates numerically, but that they can actually point in opposite directions! Here, we have native grass percent cover from 42 paired plots, preburn and five years postburn, in the sagebrush-steppe monitoring type in Lava Beds National Monument (northern California, USA) (Figure 2; Table 4). These data are intended to measure park-wide trends across the sagebrush monitoring type, so the RoM is the appropriate metric here. The MoR is 1.17, suggesting a slight (17%) increase in native grass cover five years after burns, whereas the RoM (the correct metric) yielded an 8% decrease. Here, the MoR approach based on individual ratios yields a 95% CI of 0.96–1.53 strongly suggestive of an increase, whereas the RoM yields a 95% CI of 0.80, 1.07, equally strongly suggestive of a decrease! Caveat emptor! The reason for the discrepancy between the two metrics is a small handful plots with very low prefire native grass cover that exert high leverage on the MoR due to equal weighting (Figure 2). For example, grass cover in one plot increased from 1% to 5% after the burn, which is a fivefold increase even though real cover increased by only 4 percentage points! In fact, just four plots with prefire cover of ≤4% (no other plot in the park had <12%) have an MoR of 3.2, compared to 0.95 for the other 38 plots (more in line with the RoM of 0.91). These plots contribute very little to overall grass cover across the monitoring type in the park, which is what the monitoring was intended to measure. In this example, the preburn percent cover of grasses was approximately 30%. If a specific management goal had been declared (it was not here) in relative terms, it might have been, say, “a 50% increase.” Since the measurement units are percentage, this statement is unclear. A 50% relative change would be 15 percentage points (an increase to 45% cover), whereas a 50 percentage point increase would yield 80% cover after the burn. In our examples, data were paired across time. This is a common circumstance, but not exclusively so. For example, pairing can have a spatial basis. Long et al. (2017) employed paired plots in a study examining the effect of removing feral pigs on soil and nutrient availability. One of each pair was inside an exclosure, the other out. Additionally, paired sampling is often done in ecology to measure the effects of natural or intentional experimental treatments (e.g., burned vs unburned, or sprayed vs unsprayed, etc.) for practical reasons, yet the overall inference is still on the population rather than individual pairs. The same theory and application of RoM vs MoR would apply. RoM and MoR can yield wildly different results driven by, in particular, high variation in the denominator terms. In our first example, the conclusions were approximately an order of magnitude different between the two methods. In the third example, the two estimators pointed in opposite directions (MoR suggested an increase, while MoR pointed to a decrease)! Just as it is possible that one might, for different purposes, use both the mean and median in turn (applicable if the data are skewed), so it is also possible that one might, for different ends, use both the RoM and the MoR in a given study. The mean of ratios can be handled with standard statistical tools, since summarized data are simply a single sample of values, which happen to be ratios. The ratio of means, however, is better served by use of bootstrapping. Gerow has created an interactive Excel app for confidence intervals and will gladly send such to interested readers. The intention when using paired data in ecological studies is often to make inference at the population level: how much change has there been at the landscape or watershed level. In that case, the ratio of the two means is estimating the parameter of interest, whereas the mean of ratios is useful if, on average, individual change is of interest. Our examples showed that the two approaches (RoM or MoR) can produce markedly different results, so careful attention to choosing between them is paramount. For the MoR, any standard single sample procedure will work (e.g., use a one-sample t if sample sizes are large enough). Bootstrapping is useful for confidence intervals when using the RoM; we recommend that bca-bootstrap procedure in that case. If interest happens to be at the individual level, in which case the mean of pairwise ratios is suitable, we note that zero values in the denominator require either dropping that data point or replacing the zero with some arbitrary small value (much as we deal with zero values when deploying logarithmic transformation).

Fire regimes in North American forests are diverse and modern fire records are often too short to capture important patterns, trends, feedbacks, and drivers of variability. Tree-ring fire scars provide valuable perspectives on fire regimes, including centuries-long records of fire year, season, frequency, severity, and size. Here, we introduce the newly compiled North American tree-ring fire-scar network (NAFSN), which contains 2562 sites, >37,000 fire-scarred trees, and covers large parts of North America. We investigate the NAFSN in terms of geography, sample depth, vegetation, topography, climate, and human land use. Fire scars are found in most ecoregions, from boreal forests in northern Alaska and Canada to subtropical forests in southern Florida and Mexico. The network includes 91 tree species, but is dominated by gymnosperms in the genus Pinus. Fire scars are found from sea level to >4000-m elevation and across a range of topographic settings that vary by ecoregion. Multiple regions are densely sampled (e.g., >1000 fire-scarred trees), enabling new spatial analyses such as reconstructions of area burned. To demonstrate the potential of the network, we compared the climate space of the NAFSN to those of modern fires and forests; the NAFSN spans a climate space largely representative of the forested areas in North America, with notable gaps in warmer tropical climates. Modern fires are burning in similar climate spaces as historical fires, but disproportionately in warmer regions compared to the historical record, possibly related to under-sampling of warm subtropical forests or supporting observations of changing fire regimes. The historical influence of Indigenous and non-Indigenous human land use on fire regimes varies in space and time. A 20th century fire deficit associated with human activities is evident in many regions, yet fire regimes characterized by frequent surface fires are still active in some areas (e.g., Mexico and the southeastern United States). These analyses provide a foundation and framework for future studies using the hundreds of thousands of annually- to sub-annually-resolved tree-ring records of fire spanning centuries, which will further advance our understanding of the interactions among fire, climate, topography, vegetation, and humans across North America.

Prescribed fire in dry coniferous forests of the western U.S. is used to reduce fire hazards. How large, old trees respond to these treatments is an important management consideration. Growth is a key indicator of residual tree condition, which can be predictive of mortality and response to future disturbance. Using a combination of long-term plot records and dendrochronological samples, we analyzed the effects of prescribed fire treatments from the early 1990 s on forest structure and individual tree growth in mixed-conifer forests of Lassen Volcanic National Park in northern California. Prescribed fire reduced stand live tree basal area and stem density at our sites up to 10 years following fire. Within two prescribed fire burn units and two adjacent unburned stands, we analyzed tree cores from 136 large (mean stem diameter > 70 cm) yellow pine (Pinus jeffreyi and P. ponderosa) and 136 large (mean stem diameter > 50 cm) white fir (Abies concolor). After accounting for annual precipitation, basal area increment for individual trees initially declined up to < 3 years post-fire for white fir and > 10 years post-fire for yellow pine, presumably in response to tree injuries. Growth improved for both species at a site that was burned twice, particularly for white fir. Recent average basal area increment was positively related to crown ratio and negatively associated with an index of local competition. Our findings suggest that forest management, such as prescribed fire and mechanical thinning, may be beneficial in terms of maintaining or improving tree growth among large residual trees. However, managers may want to balance the benefits of these treatments against inadvertent injury and mortality of large trees.

Fuels treatments and fire suppression operations during a fire are the two management influences on wildfire severity, yet their influence is rarely quantified in landscape-scale analyses. We leveraged a combination of datasets including custom canopy fuel layers and post-fire field data to analyse drivers of fire severity in a large wildfire in the southern Cascade Range, California, USA. We used a statistical model of tree basal area loss from the fire, factoring in weather, fuels and terrain to quantify the extent to which prescribed burning mitigated wildfire severity by simulating potential wildfire severity without prescribed fire and comparing that with modelled severity from areas burned with prescribed fire. Similarly, using a map of operations intensity, we calculated predicted fire severity under a scenario with no operations and used these predictions to quantify the influence of operations. We found that prescribed fires and operations reduced tree basal area loss from the wildfire by an average of 32% and 22% respectively, and that severity was reduced by 72% in areas with both prescribed fire and operations. Our approach could be applied to other wildfires and regions to better understand the effects of fuel treatments and fire suppression operations on wildfire severity.

Quantifying historical fire regimes provides important information for managing contemporary forests. Historical fire frequency and severity can be estimated using several methods; each method has strengths and weaknesses and presents challenges for interpretation and verification. Recent efforts to quantify the timing of historical high-severity fire events in forests of western North America have assumed that the "stand age" variable from the US Forest Service Forest Inventory and Analysis (FIA) program reflects the timing of historical high-severity (i.e. stand-replacing) fire in ponderosa pine and mixed-conifer forests. To test this assumption, we re-analyze the dataset used in a previous analysis, and compare information from fire history records with information from co-located FIA plots. We demonstrate that 1) the FIA stand age variable does not reflect the large range of individual tree ages in the FIA plots: older trees comprised more than 10% of pre-stand age basal area in 58% of plots analyzed and more than 30% of pre-stand age basal area in 32% of plots, and 2) recruitment events are not necessarily related to high-severity fire occurrence. Because the FIA stand age variable is estimated from a sample of tree ages within the tree size class containing a plurality of canopy trees in the plot, it does not necessarily include the oldest trees, especially in uneven-aged stands. Thus, the FIA stand age variable does not indicate whether the trees in the predominant size class established in response to severe fire, or established during the absence of fire. FIA stand age was not designed to measure the time since a stand-replacing disturbance. Quantification of historical "mixed-severity" fire regimes must be explicit about the spatial scale of high-severity fire effects, which is not possible using FIA stand age data.

Prescribed fire is commonly used for restoration, but the effects of reintroducing fire following a century of fire exclusion are unknown in many ecosystems. We assessed the effects of three prescribed fires, native ungulate browsing, and conifer competition on quaking aspen (Populus tremuloides Michx.) regeneration in four small groves (0.5 ha to 3.0 ha) in Lassen Volcanic National Park, California, USA, over an 11 yr period. The effects of fire on aspen regeneration density and height were variable within and among sites. Post-fire aspen regeneration density generally decreased with greater conifer basal area (rs = −0.73), but there was a wide range of aspen regeneration densities (4000 to 36 667 stems ha−1) at transects with no live conifers post-fire. The height of aspen regeneration increased as a function of increasing years-since-fire (1 yr to 11 yr), but heavy browsing by mule deer (Odocoileus hemionus Rafinesque) may alter future growth trajectories. Median percent of aspen regeneration browsed was high in burned (91 %) and unburned (81 %) transects. Only 7 % (282 stems ha−1 to 333 stems ha−1) of post-fire aspen regeneration in 11-year old burns exceeded the height necessary to escape mule deer browsing (150 cm). Browsing may also be altering aspen growth form, such that multi-stemmed aspen regeneration was positively associated with proportion of aspen regeneration browsed. These four case studies indicate that the effects of prescribed fires on quaking aspen in the southern Cascade Range of northern California were highly variable and, when coupled with biotic factors (such as deer browsing and competing vegetation) and varying fire severity, fire may either benefit or hasten the decline of small aspen groves.

Reconstructions of dry western US forests in the late 19th century in Arizona, Colorado and Oregon based on General Land Office records were used by Williams & Baker (2012; Global Ecology and Biogeography, 21, 1042-1052; hereafter W&B) to infer past fire regimes with substantial moderate and high-severity burning. The authors concluded that present-day large, high-severity fires are not distinguishable from historical patterns. We present evidence of important errors in their study. First, the use of tree size distributions to reconstruct past fire severity and extent is not supported by empirical age-size relationships nor by studies that directly quantified disturbance history in these forests. Second, the fire severity classification of W&B is qualitatively different from most modern classification schemes, and is based on different types of data, leading to an inappropriate comparison. Third, we note that while W&B asserted surprising' heterogeneity in their reconstructions of stand density and species composition, their data are not substantially different from many previous studies which reached very different conclusions about subsequent forest and fire behaviour changes. Contrary to the conclusions of W&B, the preponderance of scientific evidence indicates that conservation of dry forest ecosystems in the western United States and their ecological, social and economic value is not consistent with a present-day disturbance regime of large, high-severity fires, especially under changing climate.

Fire suppression has been the dominant fire management strategy in the West over the last century. However, managers of the Gila and Aldo Leopold Wilderness Complex in New Mexico and the Saguaro Wilderness Area in Arizona have allowed fire to play a more natural role for decades. This report summarizes the effects of these fire management practices on key resources, and documents common challenges in implementing these practices and lessons for how to address them. By updating historical fire atlases, we show how fire patterns have changed with adoption of new policy and practices.

Fire history researchers employ various forms of search-based sampling to target specimens that contain visible evidence of well preserved fire scars. Targeted sampling is considered to be the most efficient way to increase the completeness and length of the fire-scar record, but the accuracy of this method for estimating landscape-scale fire frequency parameters compared with probabilistic (i.e. systematic and random) sampling is poorly understood. In this study we compared metrics of temporal and spatial fire occurrence reconstructed independently from targeted and probabilistic fire-scar sampling to identify potential differences in parameter estimation in south-western ponderosa pine forests. Data were analysed for three case studies spanning a broad geographic range of ponderosa pine ecosystems across the US Southwest at multiple spatial scales: Centennial Forest in northern Arizona (100ha); Monument Canyon Research Natural Area (RNA) in central New Mexico (256ha); and Mica Mountain in southern Arizona (2780ha). We found that the percentage of available samples that recorded individual fire years (i.e. fire-scar synchrony) was correlated strongly between targeted and probabilistic datasets at all three study areas (r=0.85, 0.96 and 0.91 respectively). These strong positive correlations resulted predictably in similar estimates of commonly used statistical measures of fire frequency and cumulative area burned, including Mean Fire Return Interval (MFI) and Natural Fire Rotation (NFR). Consistent with theoretical expectations, targeted fire-scar sampling resulted in greater overall sampling efficiency and lower rates of sample attrition. Our findings demonstrate that targeted sampling in these systems can produce accurate estimates of landscape-scale fire frequency parameters relative to intensive probabilistic sampling.

Fire managers often need data that is spatially explicit at a fine scale (30 m) but is also laborious and time consuming to collect. This study integrates Landsat 5 imagery and topographic information with plot and tree based data to model and map four key canopy fuels variables: Canopy Bulk Density (CBD), Canopy Cover (CC), Canopy Base Height (CBH), and canopy Height (HT). We sampled 223 plots of 500 m(2) each in Lassen Volcanic National Park. Within each plot we recorded every tree by species, diameter, condition, and canopy position. Additionally, we measured each tree's height, height to live crown base and height to dead crown base. Finally, we took three hemispherical photographs of the forest canopy above each plot. We developed five topographic variables-elevation, slope, aspect, and two measures of topographic position-and used Landsat 5 spectral bands 1-5, and 7 as well as the Normalized Difference Vegetation Index (NDVI) and the Tasseled Cap Greenness, Brightness, and Wetness to model and then predict these canopy fuels variables for both 2009 and 2003 across LVNP. RF models relating predictor variables to canopy fuels characteristics had pseudo-r(2) values ranging from 0.55 to 0.68. To demonstrate the potential utility of our mapping procedure, we used our 2003 canopy fuels map along with a previously unpublished contemporary surface fuels map and the fire behavior modeling program FlamMap (R) to relate predicted fire behavior of our fuels maps with fire severity from the Monitoring Trends in Burn Severity (MTBS) dataset for the Bluff (2004) fire. (C) 2012 Elsevier B.V. All rights reserved.

Anticipating future forest-fire regimes under changing climate requires that scientists and natural resource managers understand the factors that control fire across space and time. Fire scars - proxy records of fires, formed in the growth rings of long-lived trees - provide an annually accurate window into past low-severity fire regimes. In western North America, networks of the fire-scar records spanning centuries to millennia now include hundreds to thousands of trees sampled across hundreds to many thousands of hectares. Development of these local and regional fire-scar networks has created a new data type for ecologists interested in landscape and climate regulation of ecosystem processes - which, for example, may help to explain why forest fires are widespread during certain years but not others. These data also offer crucial reference information on fire as a dynamic landscape process for use in ecosystem management, especially when managing for forest structure and resilience to climate change.