Human and climate influences on frequent fire in a high-elevation tropical forest


Correspondence author. E-mail:


  1. Surface fire has increasingly been regarded as a critical threat to tropical forests, but much of the research documenting degradation of tropical forests by fire comes from the low-elevation humid tropics. Fire in high-elevation tropical forests has received less research attention, but these forests are of high conservation value because they support unique ecosystems, which are often isolated due to their restriction to widely separated peaks.
  2. We investigated the frequency and ecological impact of fire on a high-elevation tropical forest of Pinus hartwegii in Pico de Orizaba National Park in central Mexico. This forest was previously thought to have been degraded by excessive human-caused fires. We assessed human-caused changes to the fire regime as well as the impact of climate on fire occurrence, both previously undocumented in this region.
  3. We found no increase in fire frequency or evidence of degradation of the forest. We found that the forest was uneven-aged and contained many large and old trees (maximum age 483 years). In the twentieth century, the forest experienced a frequent surface fire regime, with fires scarring trees in 90 of 100 years. However, most fires were small and asynchronous among sites. Inter-annual climatic variability was not an influential driver of fire, and El Niño–Southern Oscillation was not significantly related to the occurrence of widespread fire.
  4. Synthesis and applications. Our results show that this high-elevation tropical forest has not been degraded but has existed with frequent fires for at least a century. A trend in the 21st century towards less-frequent fire could be cause for concern, as a decrease in fire frequency could lead to an increase in tree density and a loss of resilience in the face of climate change and other future disturbance. We recommend that managers take into account historical fire regimes in their local areas: frequent surface fires in the case of Pico de Orizaba. It is important to recognize that although fire can be detrimental in many low-elevation tropical forests, it is an integral part of this high-elevation tropical forest ecosystem, and other high-elevation forests may show similar patterns.


Surface fire has increasingly been regarded as one of the most critical threats to tropical forests (Laurance 2003). However, much of the research documenting degradation and destruction of tropical forests by fire comes from the low-elevation humid tropics (Cochrane 2003). Although fire has been acknowledged as an important part of the ecology of high-elevation tropical forests, it has received little research attention (e.g. Smith & Young 1987). Similar to low-elevation tropical forests, high-elevation tropical forests face threats including climate change, livestock grazing and illegal logging (Toledo-Aceves et al. 2011). High-elevation tropical forests are important conservation targets because they form isolated archipelagoes of rare ecosystems, often separated by low-elevation ecosystems and human development.

The fire regime in the high-elevation tropical forests of central Mexico is relatively unknown; no long-term dendrochronological reconstructions of fire have been carried out in the area. Rodríguez-Trejo & Fulé (2003) suggested that three categories of forest exist in Mexico: (i) forests that have had less fire in recent decades as compared to historical fire occurrence due to human-induced fire exclusion, (ii) forests that have continued to burn at frequencies and severities similar to historical patterns and (iii) forests that have experienced excessive fire with deleterious ecological consequences due to human practices of setting fire for agricultural or other uses (as documented in Román-Cuesta, Retana & Gracia 2004). Fire history studies that have been conducted in Mexico to date have found either regular, frequent fires continuing up to the present (e.g. Fulé et al. 2011), or an abrupt cessation of fires correlated with increased human land use including livestock grazing, road building and timber harvesting, and often associated with the formation of ejidos (e.g. Heyerdahl & Alvarado 2003; Yocom et al. 2010).

In central Mexico, a series of tall volcanic peaks form a chain of geographically isolated islands of high-elevation tropical forest. At the tree line on these volcanoes, the forests consist of monospecific stands of Pinus hartwegii Lindl., a fire-resistant species found in Mexico's highest-elevation forests (Rodríguez-Trejo & Fulé 2003). As P. hartwegii is restricted to the uppermost elevations, its populations are highly isolated and this has led to significant genetic divergence (Schaal & Leverich 1996). Fire history studies are rare in high-elevation tropical forests (but see Martin & Fahey 2006) and in tropical forests in general, but two fire history studies have been completed in P. hartwegii forests, both several hundred kilometres north of Pico de Orizaba in the Sierra Madre Oriental of north-eastern Mexico. Those studies showed that the historical fire regime was characterized by frequent surface fires with mean fire intervals (MFIs) ranging from 8·6 to 16·4 years (Yocom et al. 2010; Yocom 2011). However, the range of P. hartwegii extends from northern Mexico south to Guatemala and Honduras, and it is possible that other populations of this species, including populations south of the Tropic of Cancer, may experience different fire regimes. This study is the first of its kind in high-elevation forests south of the Tropic of Cancer in Mexico (Biondi, Hartsough & Galindo Estrada 2005), designed to analyse the ecological effects of fire on these rare and valuable ecosystems.

Pico de Orizaba is a high volcanic peak in a national park in Mexico (Fig. 1) where forest degradation has been attributed to human-caused fires. Researchers who studied the timberline at Pico de Orizaba in the 1970s observed that people using the lower slopes of the mountain for livestock grazing set fires for agricultural reasons (Lauer & Klaus 1975). Although Lauer & Klaus (1975) did not quantitatively study the fire regime, nor did they specify exactly when human-caused degradation began, they stated that recent human-ignited fires on these volcanoes were different in several ways from natural lightning-caused fires: (i) they took place almost every year, whereas natural fires were estimated to occur every 6–7 years, (ii) they were usually set in February and March while natural fires occurred at the beginning of the rainy season in May and (iii) they were started below the timberline and swept up into the crowns of trees by the upslope wind, while natural fires typically started at timberline and moved downslope as surface fires. They speculated that the upper timberline at Pico de Orizaba had become lower in elevation due to human-caused fire (Lauer & Klaus 1975; Lauer 1978). The current study tests several of their ideas.

Figure 1.

(a) Map of Mexico with a black square showing the location of Pico de Orizaba National Park (square is not to scale). Circles indicate 775 weather stations with at least 90% complete data for at least 30 years. Filled circles indicate a negative correlation (<−0·1) between precipitation and Southern Oscillation Index values, and unfilled circles indicate a positive correlation (>0·1). Circle size is proportional to strength of the correlation. (b) Map showing timberline location of six sites between Pico de Orizaba and Sierra Negra in Pico de Orizaba National Park. Image from Google Earth. (c) Photograph of Site 1, with Sierra Negra in the background. (d) Map showing distribution of Pinus hartwegii in isolated sites in Mexico and Central America (Data from Critchfield & Little 1966).

The geographical location of Pico de Orizaba National Park is also ideal for testing fire relationships with El Niño–Southern Oscillation (ENSO). The dipole between northern Mexico, where La Niña events tend to be correlated with dry conditions, and southern Mexico, where El Niño events tend to be correlated with dry conditions, is located close to the Tropic of Cancer (Fig. 1). In north-western Mexico, several studies have linked fire occurrence to La Niña events (Fulé & Covington 1999; Heyerdahl & Alvarado 2003; Fulé, Villanueva-Díaz & Ramos-Gómez 2005; Skinner et al. 2008). In north-eastern Mexico, the situation is more complicated, with a finding at Peña Nevada in the Sierra Madre Oriental that fires were associated with La Niña historically, but the relationship was unstable over time (Yocom et al. 2010). In southern Mexico, fire is more likely to occur during El Niño events such as those of 1983 and 1998 (Román-Cuesta, Gracia & Retana 2003). The current study allowed us to investigate long-term climate–fire relationships in this region south of the dipole, which were previously undocumented.

Based on the observations of Lauer & Klaus (1975), we tested the following hypotheses related to forest ecology: (1a) fire regime changes in the mid-twentieth century included an increase in fire frequency, a change in fire type from surface fire to crown fire and a change in seasonality of fire to more early-season fires; (1b) forest degradation is evidenced by high tree mortality and little regeneration; and (1c) forest structure or demographics make the forest less resilient to future disturbance such as drought, fire or insect attacks. We also tested two hypotheses related to fire–climate relationships: (2a) El Niño events and low Palmer Drought Severity Index (PDSI) values are associated with fire occurrence; and (2b) the relationship between ENSO and fire has been consistent over time. Investigating the influences of human and climate drivers of fire along with the ecological effects of fire in Pico de Orizaba National Park provides us with a picture of the ecological role of fire in a high-elevation tropical forest, the current conservation status of a rare ecosystem and identifies information gaps for further research.

Materials and methods

Study area

Pico de Orizaba (also called Citlaltépetl, meaning ‘star mountain’ in the Náhuatl language) is the highest peak in Mexico and the third highest peak in North America, reaching an elevation of 5,675 m a.s.l. Its companion peak, Sierra Negra, c. 7 km to the southwest, reaches 4600 m a.s.l. The volcano is located on the border of the states of Veracruz and Puebla at the eastern end of the Transvolcanic Belt, a chain of high volcanoes that stretches across central Mexico. The forest at the timberline, between 3900 and 4200 m, is dominated by P. hartwegii. The understorey consists mostly of thick bunchgrasses including Calamagrostis tolucensis (Kunth) Trin. ex Steud., Festuca tolucensis (Kunth) and Muhlenbergia quadridentata (Kunth) trin. (Lauer & Klaus 1975).

Using temporary weather stations, Lauer & Klaus (1975) estimated that mean annual temperature at 4000 m a.s.l on the north side of Pico de Orizaba was 5 °C with an average 6 °C mean daily range. Mean annual precipitation at 4000 m was 900 mm. The majority of precipitation (>80%) falls in the summer rainy season between 1 May and 31 October.

ENSO–precipitation correlation map of Mexico

To characterize the relationship between ENSO and precipitation in Mexico, we created a map of correlations between weather station precipitation data and the Southern Oscillation Index (SOI) (see Caso, González-Abraham & Ezcurra 2007 for a similar map, but restricted to the Pacific coast of Mexico). SOI is the difference in surface air pressure between Darwin, Australia, and Tahiti. From the ERICIII climatic data base (IMTA 2005), we extracted data for weather stations in Mexico that had ≥90% complete data for ≥30 years. We calculated correlations between the resulting high-quality precipitation data and SOI and mapped the results, coding each weather station on the map by direction and strength of the correlation (Fig. 1).

Field methods

We established six 12-ha sites at the timberline in Pico de Orizaba National Park in 2010. The sites were arrayed in the saddle between Pico de Orizaba and Sierra Negra (Fig. 1). We chose the sites based on the presence of old trees or old remnant wood, with the goal of compiling the longest possible tree-ring record for the analysis of fire history and fire–climate relationships. Because of our emphasis on finding old wood, the study area is not necessarily representative of the whole park. The average elevation of the six sites ranges from 3912 m to 4132 m a.s.l, and the average slope ranges from 38% to 61% (Table 1).

Table 1. Characteristics of six sites (each 12 ha in area) at Pico de Orizaba National Park, Mexico
SiteAspectAverage elevation (m)Average slope (%)

Fire history

We systematically searched each site and collected fire-scarred samples that would provide good spatial distribution throughout each site and the longest possible fire record. We removed small partial cross-sections of fire-scarred trees, both alive and dead.

In the laboratory, using tree cores and fire-scarred samples, we developed a standard tree-ring chronology for the area. We then visually crossdated each sample, measured the ring widths and checked the crossdating with the COFECHA software program (Holmes 1983). We identified fire scars to the year of formation by noting the crossdated ring in which the fire injury occurred (Baisan & Swetnam 1990). Where possible, we also noted the position of each fire scar within the annual growth ring. We assigned scars on the ring boundary to the following year.

Fire interval statistics were calculated with FHX2 version 3.2 software (Grissino-Mayer 2001). For each site, we calculated the twentieth-century composite MFI between all fires, including fires that only scarred one tree, between fires that scarred at least two trees and between more widespread fires that scarred at least 25% of recording trees (with a minimum of two scars). Recording trees are those that have been scarred at least once; after initial wounding, injured trees are more likely to scar in subsequent fires. We used the twentieth century as the time period to compare statistics between sites because all six sites had excellent sample depth during that period. We also calculated fire interval statistics for each site starting from when the site had at least three recording samples: the years for sites 1–6 are 1888, 1853, 1818, 1902, 1805 and 1849. We used the Kolmogorov–Smirnov goodness-of-fit test to determine whether the Weibull distribution fits the data adequately. The Weibull median probability interval (WMPI) is the fire interval associated with the 50th percentile of the distribution (Grissino-Mayer 2001). It is often used in describing fire regimes because it is flexible, able to fit skewed data sets and provides a standard way to compare fire regimes across ecological gradients. Finally, to assess the possibility that humans altered the fire regime, we looked for changes over time in fire frequency, seasonality of fires and percentage of recording trees scarred.

Forest structure and fuels

To measure forest structure and age, we established five permanent plots on a 100 × 100 m grid in each of the six sites, giving a total of 30 plots. We counted seedlings (trees below 1·3 m in height) in 5·6-m-radius circular plots. We measured trees taller than 1·3 m in 11·3-m-radius circular plots, as well as the four closest live trees with a height above 1·3 m outside the plot. We did this to increase our sample depth of measured and cored trees, because in some areas, tree density was low enough that plots had only one or two trees in the 11·3-m-radius area. This method resulted in plots with radii ranging from 12·0 to 26·6 m and a plot area ranging from 0·05 to 0·22 ha. For each tree over 1·3 m in height, we recorded height, diameter at breast height and diameter at stump height. We also took increment cores from all trees over 1·3 m in height to determine tree ages. We cored trees as close as possible to their bases (0–15 cm above the base, depending on what the slope allowed) to determine tree ages most accurately. We re-cored trees when we estimated that the first core missed the centre of the tree by 10 or more rings. We measured canopy cover with a vertical densiometer at 15 points along a 15-m transect running in a random direction from the centre of each plot.

In the laboratory, cores were affixed to core mounts and sanded. We crossdated cores when possible and counted rings when necessary to obtain an estimate of each tree's age. Counting rings was occasionally necessary with cores from young trees, which had relatively few rings that also tended to be complacent, that is, showed little variation in width from year to year. We found few false rings and occasional missing rings. When a core did not pass through the centre of a tree, we used a template of concentric, appropriately sized circles to estimate the number of rings to the centre. We grouped tree ages into 10-year classes for analysis.

We calculated trees per hectare and basal area per hectare for each plot. Distances to the four closest trees were erroneously not recorded in plot 4-2, so for that plot we used the average plot size in Site 4, which was 0·070 ha.

Fire and climate

To evaluate climate conditions related to fire occurrence at our study sites, we used superposed epoch analysis (SEA) in FHX2 version 3.2 (Grissino-Mayer 2001) to compare indices of ENSO and PDSI during fire years, and also for 5 years prior to, and 2 years after each fire year. We used an instrumentally measured index of winter NINO3 (Kaplan et al. 1998; Reynolds et al. 2002) and reconstructed PDSI (Cook et al. 2004). To assess statistical significance in the SEA analyses, confidence intervals (95%) were calculated using bootstrapped distributions of climate data in 1000 trials. To compare climate patterns with more widespread fire years, we identified fire years in which at least three sites had the formation of fire scars on at least two trees. This gave us a list of 25 widespread fire years between 1902 and 1999.


Fire history

We collected samples from 142 fire-scarred trees in the six sites: 102 (72%) were live trees, 21 (15%) were snags, 10 (7%) were logs and 9 (6%) were stumps. All samples were P. hartwegii. We were able to crossdate 118 of the samples (83·1%).

The oldest scar occurred in 1764, and the last scars were formed in 2002 (Fig. 2). We were able to determine seasonality for 57% of the fire scars. Of those, the majority (495 scars, 95·2%) of fire scars for which we could determine seasonality were formed in the dormant period: 14 (2·7%) formed in early earlywood, 8 (1·5%) formed in middle earlywood and 3 (0·6%) formed in late earlywood. No fire scars formed in latewood. All 25 scars that were not formed in the dormant period formed in 1939 or later. Depending on the site, the MFI in the twentieth century for all fires ranged from 2·1 to 3·5 years, the MFI for fires that scarred at least two trees ranged from 3·3 to 5·8 years and the MFI for fires that scarred ≥25% of recording trees ranged from 6·5 to 9·5 years (Table 2). The Weibull distribution fits our data in all but two cases (indicated in Table 2). WMPI values were similar but in each case slightly smaller than corresponding MFI values, indicating that the distribution of fire intervals is skewed towards small intervals.

Figure 2.

Top: all fires in each site and in all sites combined. Middle: years when a minimum of two samples were scarred in each site. Bottom: years when a minimum of 25% of recording samples and a minimum of two samples were scarred. The grey boxes on the left cover data that were not used in analysing fire statistics, because the sample size decreased substantially before 1900.

Table 2. Fire interval characteristics for the years 1900–2009 at six sites in Pico de Orizaba National Park, Mexico
 Site 1Site 2Site 3Site 4Site 5Site 6
  1. a

    Weibull model does not fit the data.

  2. b

    Average sample mean fire interval calculations were calculated for the time period covered by each individual sample.

Number of intervals
All fires373926414534
All fires, ≥2 scarred211716232929
25% scarred111010151114
Mean fire interval (years)
All fires2·72·63·52·42·12·9
All fires, ≥2 scarred4·75·65·84·23·33·4
25% scarred8·99·59·26·58·66·6
Weibull median interval (years)
All fires2·42·3a3·32·22·0a2·5
All fires, ≥2 scarred4·44·95·43·93·13·0
25% scarred8·69·08·95·77·26·1
Minimum interval (years)
All fires111111
All fires, ≥2 scarred113111
25% scarred343132
Maximum interval (years)
All fires10877611
All fires, ≥2 scarred111512101111
25% scarred162316183017
Average sample mean fire intervalb (years)14·614·214·515·416·912·3

In the twentieth century, a fire scarred at least one tree in all six sites in 90 of 100 years. There were 2 years in the twentieth century when at least two trees in all six sites were scarred: 1902 and 1907. Five sites had the formation of at least two scars in 1943 and 1960. Prior to 1900, the sample depth was lower in most sites. However, when we calculated MFI for each site starting in the year for each site when the site had at least three recording trees, the results were within 1·2 years of the MFI for each site during 1900–2009. The fire frequency decreased after 2000: only one fire year was recorded, in one site, during the period 2000–2009. This is an anomaly compared to the twentieth century, when fire was recorded in at least one site in 90% of years. We found no differences over time in the percentage of recording trees scarred in each fire year.

Forest structure

We measured a total of 305 live trees and 24 dead trees in the 30 forest plots. All but one were P. hartwegii; the one exception was a Juniperus species. We were able to estimate ages using cores for 274 of those trees; the other cores were either lost (10 cores) or did not come close enough to the pith to estimate age (21 cores).

Mean tree density ranged from 67·6 to 242·5 trees ha−1 in the six sites, while basal area ranged from 16·3 to 34·5 m2 ha−1 (Table 3). Snags were present in 10 of the 30 plots; where snags were present, snag density ranged from 3·6 to 15·6 snags ha−1 (average 9·6 snags ha−1). We found seven cut stumps in our 30 plots, in sites 1, 3 and 4. Seedling density ranged from 0 to 260 seedlings ha−1 (average 113 seedlings ha−1), and there was no correlation between seedling density and overstorey tree density or basal area.

Table 3. Forest structure characteristics for six timberline sites in Pico de Orizaba National Park, Mexico
SiteTree density (trees ha−1)aBasal area (m2 ha−1)Seedlings ha−1Canopy cover(%)
  1. a

    Trees are classified as stems over 1·3 m in height, and seedlings are classified as <1·3 m in height.


All sites were uneven aged (Fig. 3). Site 3 was the only site where none of the sampled trees had a centre date prior to 1880. However, we had one core from site 3 with an inner date of 1637; this core was broken off near the centre, and we were not able to estimate the centre date of the tree. In the other sites, cores which we could not use for age estimation had inner dates ranging from 1719 to 1963. If we had been able to estimate centre dates from these cores, they would have increased tree density values in the older age classes. The tree size distribution mirrored the age distribution closely. Age and size of individual trees were highly correlated, with a correlation coefficient of 0·81 for all trees that were aged.

Figure 3.

Top: age distributions in sites 1–6. Boxes represent 1st and 3rd quartiles, with the median shown as a vertical line within each box. Whiskers represent oldest and youngest trees in each site, with the exception of a tree that germinated in the early 1500s in Site 1. Bottom: tree density (trees ha−1) in 10-year age classes in all sites combined.

Fire–climate relationships

In the SEA analyses, PDSI values and winter NINO3 values tended to be below average during fire years, but the trends were not significant. In 1902 and 1907, the years of most widespread fire across our sites, winter NINO3 values were close to average. We found that widespread fires have occurred in both El Niño and La Niña years in the past.


Changes to the fire regime

Our first hypothesis was related to whether there was evidence that humans had altered the fire regime in the mid-twentieth century, an observation made by researchers in the 1970s. We hypothesized that fire frequency increased, that the fire type changed from surface fire to crown fire and that a change in seasonality of fire occurred in the mid-twentieth century. Our results do not support this hypothesis. The fire scar record indicates that fires were usually small in area, frequent and asynchronous among sites throughout the twentieth century. The small size of many fires suggests that they may not have been fires that escaped from agricultural burns on the slopes below. Human-started fires in agricultural fields far below the timberline would be more likely to affect a larger area as they spread upslope and scar more trees, although we have no conclusive evidence for this theory. The small size of many fires is probably not due to limited fuel, given the abundant grasses (L. L. Yocom & P. Z. Fulé, personal observation). We speculate that burning conditions, including fuel moisture and temperature, are not normally ideal for high fire spread. In addition, fires develop more slowly at higher elevation due to the lower partial pressure of oxygen, which could also help account for the small fire size observed at our timberline sites (Wieser, Jauch & Willi 1997).

We did not find evidence of high-severity fire in our sites. This is in contrast to P. hartwegii forests in the Sierra Madre Oriental, where we found clusters of pith dates in fire-scarred samples that could be evidence for small patches of high-severity disturbance that opened up growing space for pulses of regeneration (Yocom et al. 2010; Yocom 2011). In the present study, tree centre dates were not episodic but were fairly steady through time.

We did find an interesting trend in fire seasonality, which was that the very small number of scars that formed within a ring as opposed to on the ring boundary (<5%) were formed in 1939 or later. This could indicate a change in seasonality of some fires due to human influence, although the trend is not the direction we had hypothesized; we had predicted an increase in early-season fires rather than an increase in late-season fires.

While we did not see an increase in fire frequency during the period we analysed, there was a striking decrease in fire frequency in the past decade. The drop in fire frequency could be due to increased fire suppression. In the mid-1990s, a telescope was built on Sierra Negra, and with that came people who monitor the telescope full time. Also, although Pico de Orizaba National Park was officially designated a national park in 1937, it did not have personnel until 2004. Since 2004, an effort has been made to extinguish fires immediately when they start (H. Rojas, personal communication, April 2010). We recognize that important resources such as telescopes must be protected, but we recommend that whenever possible, fires be allowed to burn to maintain the disturbance regime and associated forest structure.

Forest degradation

Our results do not support the second hypothesis based on Lauer & Klaus's (1975) observations that frequent fires caused mortality and constrained regeneration. We found old, large trees in every site, few snags and no evidence of high-severity fire. Measured regeneration varied substantially among sites and among plots, which reflected the patchiness of regeneration that we observed. Estimating mortality rates compared to regeneration rates can provide a benchmark of whether regeneration is sufficient to replace mortality. In the simplest calculation (see Mast et al. 1999), maintenance of 119 trees ha−1 (±20%) in a distribution with a maximum age of 270 years (±20%) predicts a mortality rate range of 3·0–6·5 trees ha−1 decade−1. Tree establishment rates in the 1990s and 2000s were well above that mortality rate (9·6 and 11·3 trees ha−1 decade−1) but were lower in the 1970s and 1980s (1·7 trees ha−1 decade−1 in both decades). Overall, our results suggest that forest degradation has not occurred in our sites: we found low mortality, the presence of old, large trees and sustained recruitment of trees over time.

Forest sustainability

Our third hypothesis was related to forest sustainability: current forest structure or demographics make the forest less resilient to severe disturbance such as fire, insect outbreaks or drought. Instead, we found that this high-elevation tropical forest, with a frequent low-intensity surface fire regime and low-density structure, is likely to be resilient to severe disturbance in the future. In terms of forest structure, low-density forests with low canopy bulk density are at less risk of crown fire spread (Agee & Skinner 2005). The severity of insect outbreaks can also be influenced by tree density; bark beetle mortality is often more severe in dense stands (Fettig et al. 2007). Drought effects are also related to forest density; high-density forests can be at increased risk of drought impacts if trees are less vigorous due to high levels of competition. Climate change may influence the occurrence or severity of all of these disturbances, or combinations of these disturbances, and resilience to disturbance is an increasingly important management objective (Allen et al. 2002).

In terms of demographics, the size and age distributions of the forest were similar, with a peak in the middle-sized and middle-aged categories and another peak in the small, young tree range. The relatively high density of small, young trees is found in most forests; however, the high density of young, small trees may also be related to the unusual lack of fire in the 2000s and to some degree in the 1990s. Many of the small, young trees established prior to the recent drop in fire frequency, but the recent lull in fire frequency may be allowing them to survive (Mast & Wolf 2004; Brown & Wu 2005). Overall, the forest has an uneven-aged distribution with populations of old, middle-aged and young trees. This suggests that the population is demographically stable and is unlikely to be at risk of a population crash or irruption due to demographic imbalances. Uneven-aged forests may also be more resilient to disturbance that preferentially affects certain age or size classes, such as some insect attacks (Fettig et al. 2007).

We found no evidence that humans altered the fire regime in the mid-twentieth century or caused forest degradation as we had hypothesized. However, our sites comprised a small portion of the timberline. Elsewhere, we saw evidence of more timber cutting, patchy high-severity fire and younger trees. The hypotheses that human-caused fires altered the fire regime in the mid-twentieth century and degraded the forest might be supported in other parts of the Park.

Climate–fire relationships

Inter-annual climatic variability was not a strong driver of fire occurrence, based on the fact that small fires were recorded in our sites in 90 of 100 years in the twentieth century. Such frequent fire, in sites within close proximity of each other, is unusual compared to most fire history studies in North America, and we conclude that climate conditions suitable for small fires occurred almost every year. This is further supported by the result that PDSI was not significantly associated with widespread fire years. Our hypothesis that El Niño events were associated with fire events was not supported: neither El Niño nor La Niña events were significantly associated with fire. During the strong El Niño event of 1998, central and southern Mexico experienced a record-breaking fire year (Rodríguez-Trejo & Pyne 1999). In our study area in Pico de Orizaba National Park, most samples in site 5 were scarred in 1998. However, 1997 was a big fire year in site 2 and 1999 was a big fire year in sites 1, 4 and 6. Clearly, the El Niño event of 1998, which synchronized fires across much of Mexico, did not synchronize fire events among the sites in Pico de Orizaba National Park.

We were not able to test our second climate hypothesis that ENSO events have been consistently related to fire occurrence over time, because the fire scar record did not consistently extend back past the 1830s, the point at which a change was documented in ENSO–fire relationships in the Sierra Madre Oriental (Yocom et al. 2010). Although we searched for old trees, we did not find much old wood with conserved fire scars. We were surprised by the relatively young age of most of our samples; even most logs and snags that appeared old in the field contained relatively recent fire scars. This may be related to the high frequency of fires in the study area; old remnant wood is probably difficult to find because it has been consumed by frequent fire over the centuries, and older scars on samples we found may have been consumed by subsequent fires as well.

High-elevation tropical forests around the world are under threat by human impacts and climate change. In addition, in densely populated central Mexico, relatively undisturbed forests are rare. Our results show that this high-elevation tropical forest has not been degraded but instead has been maintained in a healthy state by frequent fires for at least a century. However, during the last decade, fire occurrence has abruptly diminished, which is the cause for concern. A decrease in frequent fire could lead to an increase in tree density and a loss of resilience in the face of climate change and other future disturbance such as high-severity fire, drought or insect outbreaks. It is our hope that information from this study will support planning and decision-making by managers, scientists and government officials. We recommend that managers take into account historical fire regimes in their local areas: frequent surface fires in the case of Pico de Orizaba. Although in many low-elevation tropical forests fire can have detrimental effects, it is an integral part of the ecosystem in this high-elevation tropical forest, and other high-elevation tropical forests may show similar patterns.


We thank CONANP, H. Rojas and the Pico de Orizaba National Park for permission to work in the park. J. Cerrano, J. Villanueva-Díaz and D. Rodríguez provided valuable support to the project. R. de Jesús Amador, O. Hernández de la Torre, K. Kent, I. Zamudio and C. Vásquez García were terrific assistants in the field. We appreciate S. Curran's assistance with the ERICIII data base. R. Sheridan's help in the laboratory was greatly appreciated. Thanks also to D. Normandin, other staff and students at the Ecological Restoration Institute. We thank two anonymous reviewers, whose comments helped greatly improve this manuscript. This research was supported by a National Science Foundation Doctoral Dissertation Enhancement Project grant (OISE-1003845) and grant DEB-0640351, and the Ecological Restoration Institute at Northern Arizona University.