Research articlesFire-driven alternative vegetation states across the temperate Andes
Abstract
The theory of alternative stable states, as applied to terrestrial ecosystems, suggests that under common environmental conditions, different vegetation types may remain stable by contrasting feedback processes. In the temperate Andes, forests and shrublands of Nothofagus species have been proposed as fire-driven alternative vegetation states (AVS): while high fire frequency would promote the stability of pyrophilic shrublands, the absence of fires would stabilize pyrophobic forests. However, to confirm this hypothesis, it must be demonstrated that fire-vegetation feedbacks occur under the same environmental conditions. We aimed to (i) identify to what extent Nothofagus forests and shrublands occur in the same environmental conditions across the temperate Andes and (ii) to understand how the fire regime explains the distribution patterns of these states. We used global environmental databases and local fire data to (i) estimate the environmental niche overlap of forests and shrublands and (ii) to obtain an indicator of the fire activity at micro-basin scale (percentage of the cumulative burned area (PBA); the higher the PBA the higher the fire activity). The environmental niches of forests and shrublands overlapped by more than 70%. Shrublands become more frequent as PBA increases, suggesting that stabilizing fire-vegetation feedbacks promote their persistence. Our results provide broad-scale evidence of fire-driven AVS beyond the tropics.
This article is part of the theme issue ‘Novel fire regimes under climate changes and human influences: impacts, ecosystem responses and feedbacks’.
1. Introduction
The theory of alternative stable states proposes that a system can maintain different stable equilibrium states under the same external condition(s) [1]. In those systems, a complex suite of stabilizing (or negative) feedbacks promote stability of a given state. Stochastic events (disturbances; cf. [2]) may destabilize the system, inducing amplifying (positive) feedbacks that reinforce the effects of the disturbance and may produce a shift to an alternative state [3,4]. The alternative stable states theory has been applied to terrestrial ecosystems as a plausible explanation for the occurrence of strikingly different vegetation types (i.e. alternative vegetation states (AVS)) under the same environmental conditions [5,6]. Different processes have been proposed as drivers of change from one vegetation type to another (e.g. freezing, drought and hurricanes), with fire being the most widely studied, as changes in fire regime may switch between amplifying and stabilizing feedbacks [6].
The pioneering work on what we currently call fire-driven AVS was framed in California and Tasmania [7,8]. Since then, most research on AVS has been performed in tropical ecosystems and tropical forest–savanna mosaics have become the paradigm of fire-driven AVS [6]. These vegetation types can develop in the same environment yet are characterized by functionally distinctive community assemblages [9–12]. Contrasting fire regimes confer stability to both states through different feedback processes [6,13]: in savannas, shade-intolerant, flammable grasses promote frequent fires, thereby favouring the dominance and persistence of these fire-adapted species; in forests, the closed canopy of fire-sensitive trees provide shade that inhibits flammable grasses and maintains conditions that preclude fires (high moisture and low wind). Under extreme weather conditions, severe fires percolate into forests, inducing a shift to an open state [14,15]; conversely, long fire-free intervals permit the establishment of trees in the savanna and ultimately the development of a closed forest [16]. In the last decades, the evidence supporting fire-driven AVS beyond the tropics is growing [6], with some examples in temperate ecosystems of Australia, New Zealand [17–19] and South America [20,21]. Temperate ecosystems are typically characterized by a drought-limited fire regime, with high fuel loads that remain wet for much of the year due to high annual rainfall and low evaporative demand [22,23]. However, a rare dry period can potentially drive the forest to a different stable state. Because these processes occur at slower rates than in tropical ecosystems [24], they are more difficult to depict.
The temperate forests of southern South America represent the most extensive regions of temperate forests in the southern hemisphere [25,26], with widespread mosaics of vegetation dominated by tree species of the genus Nothofagus [27,28]. In this region, subalpine Nothofagus pumilio forests and Nothofagus antarctica shrublands are two of the most common vegetation types along 18° of latitude in Chile and Argentina. The established idea is that N. antarctica shrublands develop under the most harsh conditions, like xeric sites, poor soils and depressions subjected to cold air drainage or waterlogging [28]. In contrast, N. pumilio forests have been associated with more mesic sites, with fertile, well-drained and developed soils [28,29]. However, the differences in the environmental requirements of both formations have not been systematically addressed. On the eastern side of the Andean Patagonia, remarkable field studies have led to the proposal of fire regime as another modulator of the distribution patterns of both vegetation types. Nothofagus pumilio forest are fire-sensitive due to the lack fire-persistence traits of the dominant species, such as resprouting ability, thick bark or fire-resistant seed banks [24]. These pyrophobic forests rarely burn, because the moist, cool and shaded conditions generate a low flammable environment [20,30]. In addition, both the scarcity of dead fine fuels and the vertical discontinuity between the understory and the canopy limit fire propagation [20,21,31]. Under extreme climatic events, high-severity fires can burn the forests, promoting the transition to shrublands [32–34]. Shrublands of the resprouter N. antarctica are fire-resilient, due to the high post-fire resprouting capacity of this species. These pyrophilic shrublands burn readily because the open canopy promotes dry micro-environmental conditions, and there is vertical continuity of highly flammable fine fuels [20,21,33,35]. Taken together, all of this background suggests that different fire-vegetation feedbacks would confer stability to N. pumilio forests and N. antarctica shrublands.
The existence of vegetation-fire feedbacks alone cannot be considered unequivocal evidence of AVS [36]. By definition, alternative states must be able to develop and remain stable in a site under given environmental conditions [1]. Given the difficulty of demonstrating state switches in a given site in long-lived woody plants, spatial analyses emerge as a valuable tool to detect AVS [6,10,37]. Local evidence suggests that N. pumilio forests and N. antarctica shrublands coexist in the same environment, at least in the eastern side of the Southern Andes, which is dry and, thus, fire-prone [27,38]. Here, we ask about the generality of this process across a wide range of environmental conditions, from meso-Mediterranean to subantarctic environments, spanning a broad latitudinal range (18°, which is about the distance between Barcelona and Oslo). Specifically, we aimed to evaluate the hypothesis that N. pumilio forests and N. antarctica shrublands are AVS maintained by different fire regimes and that this occurs over a wide geographical and environmental range across the temperate Andes. We predict that, across the study area, different fire regimes explain the distribution patterns of both states under the same environment, with low flammable forests being more common in areas with less fires, while flammable shrublands dominate sites with high fire activity. Finding evidence for this hypothesis would support the idea that fire and AVS are key mechanisms in understanding vegetation patterns across large areas of South America.
2. Methods
(a) Study area
The study was conducted in the western Andes (36°–54° S), covering most of the distribution of N. pumilio forests and N. antarctica shrublands in Chile (figure 1). Only the western (Chilean) distribution of these species was considered because it is where detailed vegetation and wildfire information are available. In this area, it is possible to find mosaics of coexisting N. pumilio forests and N. antarctica shrublands under similar environmental conditions and separated by abrupt boundaries (figure 2). The climate is predominantly temperate, with Mediterranean influence in the northernmost end, which becomes increasingly rainy and colder towards the south, giving way to a subantarctic climate south of 43° S [39,40]. Soils originate from volcanic ashes, are deficient in elemental nutrients like phosphorus and have high levels of toxic aluminium [41].
Figure 1. Map of the study area. The left panel shows the micro-basins, classified according to the percentage of the cumulative burned area (PBA) during the study period (1985−2018). The four panels on the right show the N. pumilio forests (green) and N. antarctica shrublands (orange) within the micro-basins (grey contours). The number of each of these four panels corresponds to those of the left panel. The scale of each panel varies to enhance visualization.
Figure 2. Abrupt boundary between N. antarctica shrubland (left) and N. pumilio forest (right) in Villarrica National Park (39°35′ S–71°31′ W). Photo: Diego P. Ramírez.
(b) Study sites
To identify patches of N. pumilio forest and N. antarctica shrubland, we used the national vegetation survey maps of the Chilean National Forest Service (CONAF; minimum mapping unit of 0.5 ha) [42]. We define N. pumilio forest and N. antarctica shrubland as any homogeneous patch with at least 25% vegetation cover dominated by the corresponding Nothofagus species.
We characterized the fire activity at the micro-basin level using the micro-basins map from the Chilean Water General Directorate [43]. We selected those micro-basins where both N. pumilio forests and N. antarctica shrublands were present (i.e. both covering at least 1% of the micro-basin’s area). These micro-basins range between 2926 and 351 103 ha (48 197 ha on average). To obtain an indicator of the fire activity for each micro-basin, we used the wildfire database provided by the CONAF. This database records the ignition location and burned area, considering different land covers (e.g. forest, shrublands, forest plantations and others) for all fires larger than 0.01 ha during the 1985−2018 period. Note that the database does not specify the dominant species for each land cover polygon, except in the case of forest plantations. Satellite burned area products (e.g. MODIS Burned Area) were not considered because of the short time window of data available limits their use in our system, with relatively long fire return intervals [44,45]. The fire activity of each micro-basin was defined as the percentage of the cumulative burned area (PBA) of the micro-basin relative to its burnable area (hereafter PBA; figure 1). The cumulative burned area of each micro-basin was obtained as the sum of the burned area in forests and shrublands (regardless of the species), considering those fires starting in the micro-basin. We excluded fires burning less than 10 ha, as they are often confounded with agricultural burning [46], representing only 2% of the cumulative burned area in our study area. The burnable area of each micro-basin was measured as the total surface area of forests and shrublands in the micro-basin (also regardless of the dominant species), obtained from the national vegetation survey maps.
We used global gridded databases to characterize the climatic, edaphic and topographic conditions in which N. pumilio forests and N. antarctica shrublands develop. To achieve this, a stratified random sampling was carried out. First, micro-basins were categorized as ‘NoFire’ (PBA = 0%), ‘LowFire’ (0 < PBA ≤ mean) and ‘HighFire’ (PBA > mean), where the mean refers to the average of PBA for the micro-basin with fire occurrence. Then, 10 000 random points (hereafter, observations) were distributed proportionally across the area occupied by forests and shrublands in each fire category, with a minimum distance of 1 km between each point (electronic supplementary material, table S1). For each point, we extracted a set of environmental variables that characterize N. pumilio and N. antarctica auto-ecology (table 1; see §1 for details), avoiding redundant variables to prevent multicollinearity problems [51]. We obtained the minimum monthly climate moisture index (MImin; table 1) and the number of growing degree days above 5°C (NGD5), with a spatial resolution of 1 km from the CHELSA dataset [47]. Edaphic variables were obtained from SoilGrids [48] and Global Soil Dataset [49], at resolution of 0.25 and 1 km, respectively. Specifically, we obtained the soil pH (in H2O), soil cation exchange capacity (at pH 7; CEC) and the soil organic carbon (Corg) from SoilGrids, as well as the soil volumetric water content at −33 kPa (SWC) from Global Soil Dataset. Edaphic layers were obtained at the greatest available depth, in order to characterize soil properties that depend on bedrock rather than vegetation (table 1). Finally, we obtained the elevation using the digital elevation model (DEM) ALOS-PALSAR (Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar; spatial resolution of 12.5 m) generated by the Japan Aerospace Exploration Agency and available on the Centre of Information of Natural Resources of the Chilean Ministry of Agriculture [50]. Using the DEM, we obtained the terrain ruggedness index (TRI) [52]. The correlation between the predictor variables did not exceed an |r| > 0.72 (electronic supplementary material, figure S1), which is very close to the recommended threshold for this type of analysis [51]. The spatial data processing was performed in QGIS 3.28 [53]. Notice that, to avoid the loss of fine-grained resolution and to obtain a more reliable geographical projection of the ecological niches, we discarded the upscaling of the finer data.
variable | code | description |
|---|---|---|
minimum monthly climate moisture index (kg m−2 month−1)a | MImin | difference between precipitation and monthly potential evapotranspiration of the month with the highest precipitation deficit |
number of growing degree daysa | NGD5 | number of days per year at which the main daily air temperature was >5°C |
soil pHb | pH | edaphic pH in H2O (100–200 cm depth) |
soil cation exchange capacity (mmol+ kg−1)b | CEC | cation exchange capacity of the fine soil fraction (100–200 cm depth) |
soil organic carbon (dg kg−1)b | Corg | edaphic organic carbon concentration (100–200 cm depth) |
soil volumetric water content at −33 kPa (%)c | SWC | percentage of the volume of water retained by a volume of soil at field capacity (138–229 cm depth) |
elevation (m a.s.l.)d | elev | elevation above sea level according to the ALOS-PALSAR satellite DEM |
TRI |
(c) Environmental overlap between forests and shrublands
To estimate the overlap between the environmental conditions of N. pumilio forests and N. antarctica shrublands, we first model the ecological niche for each vegetation type. With this aim, we summarized the environmental variables into three orthogonal axes of variation using a principal component analysis (PCA). Variables were centred and scaled prior to the analysis. The ecological niches were computed using an ellipsoid envelope model based on the minimum volume ellipsoid method. The ellipsoids were delimited to include the 95% of the observations of each vegetation type, thus excluding marginal environmental conditions that could overestimate the ecological niche [54]. Then, to quantify the niche overlap, we first defined the three-dimensional space enclosing all the observations used to create the ellipsoids. Within this volume, 1 000 000 random points were generated, and the points within each ellipsoid and in the overlap zone were recorded. Iterating this process 100 times, we assessed whether the niche overlap was significantly greater than zero using a one-tailed Student’s t‐test. Additionally, the principal components (PCs) scores of the PCA (PC1, PC2 and PC3) of the observations used for the niche modelling were compared between non-overlapping N. pumilio forests, non-overlapping N. antarctica shrublands and the overlap zone using a one-way ANOVA based on trimmed means (with a 20% trimming level), followed by a corresponding post hoc test to compare between groups. This approach was chosen because the data were non-normal and heteroscedastic.
(d) Distribution patterns of forests and shrublands along the fire gradient
To infer whether fire regimes promote the stability of N. pumilio forests and N. antarctica shrublands growing under equivalent environmental conditions, we compared their occurrence along the gradient of PBA (see above) considering exclusively those areas where the niche overlapping conditions are present. For this, we first projected the niche overlap zone onto the geographical space of the study area by creating a raster layer for each of the three PCs of the PCA used to delimit the ellipsoids. Using the three PCA layers, we generated a suitability map for each vegetation type. The two suitability maps were intersected to identify where the niche overlapping conditions are present and thus the areas where N. pumilio forests and N. antarctica shrublands are AVS. Inside this geographical area, we recorded the vegetation type and micro-basin PBA of 10 000 random points. Then, the non-parametric Kolmogorov–Smirnov test was conducted to assess whether the frequency distribution of PBA differs between forests and shrublands. To identify where parts of the PBA gradient differ between the two vegetation types, we compared the percentiles of the two frequency distributions using 95% confidence intervals estimated by bootstrap and controlling the type I error with the Hochberg’s method.
All analyses were performed in R, using packages ellipsenm, SIBER and WRS2 [55–58].
3. Results
(a) Spatial extension of the study area
We selected 177 micro-basins covering 1 475 919 ha of N. pumilio forests and 513 416 ha of N. antarctica shrublands, between the 36° and 55° S (figure 1 and electronic supplementary material, table S1). Overall, the fire activity in the study area was low, with PBA values ranging from 0 to 32.4% (figure 1). Most micro-basins (102) were categorized as ‘NoFire’ (0% PBA). The average PBA of the 75 micro-basins with PBA > 0 was 2.3%, which was the threshold used to categorize them as ‘LowFire’ (63) or ‘HighFire’ (12; electronic supplementary material, figure S2).
(b) Environmental overlap between forests and shrublands
The PCA was performed with a total of 9787 observations after discarding those with no or unreliable environmental data. The three PCs of the PCA explained 69% of the total variance. PC1 explained 34% of the variance, was negatively correlated to the moisture index and soil organic carbon and positively to the number of degree days and pH (electronic supplementary material, figure S3). PC2 and PC3 explained 22 and 13% of the variance, respectively; the former was negatively correlated to the soil water content, elevation and terrain ruggedness, whereas the latter was negatively correlated to CEC and number of degree days and positively to pH (electronic supplementary material, figure S3).
The ellipsoids of N. pumilio forest and N. antarctica shrubland were modelled with 7280 and 2507 observations, respectively (figure 3). In terms of the overlap of the niches, the ellipsoid of N. pumilio forest has a 77% shared volume with N. antarctica (t-value = 5188.9; p‐value < 0.001), while the ellipsoid of N. antarctica shrubland shares 72% of its volume with that of N. pumilio (t-value = 5514.5; p‐value < 0.001). PC scores of overlap and non-overlapping zones were significantly different, with differences observed in all comparisons (figure 3 and electronic supplementary material, figure S4).
Figure 3. The top left panel represents the environmental niches of N. pumilio forest (green) and N. antarctica shrubland (orange) through ellipsoid envelopes. The environmental space corresponds to the first three PCs used for niche modelling. The bottom left and right panels show box plots of the PC scores exclusive for N. pumilio forests, exclusive for N. antarctica shrublands and for the overlapping zone of the ecological niches. The environmental variables more strongly correlated to each PC is shown in blue (positively correlated) and red (negatively correlated). Different letters between box plots indicate significant differences between the PC scores, obtained from the one-way ANOVA analysis based on trimmed means (see §2). The results of the analysis are shown in electronic supplementary material, figure S4. MImin: minimum monthly climate moisture index; NDG5: number of growing degree days >5°C; pH: soil pH; Corg: soil organic carbon; CEC: soil cation exchange capacity; SWC: soil volumetric water content; elev: elevation; TRI: terrain ruggedness index.
(c) Distribution pattern of forests and shrublands along the fire gradient
Niche overlapping conditions are present in 173 of the 177 micro-basins. The area with such conditions extends over 1 669 320 ha, covering 84% of the study area (electronic supplementary material, figure S5). From this, 73% is currently covered by N. pumilio forests (1 226 860 ha) and the remaining (442 460 ha) by N. antarctica shrublands.
Considering only the overlapping niche areas, the PBA frequency distributions differed between forests and shrublands (D = 0.07; p‐value < 0.001). Due the predominance of ‘NoFire’ micro-basins, the difference emerged above the 0.4 percentile and suggest that shrublands become more frequent than forests as the PBA increases, with more pronounced differences in the highest percentile (figure 4).
Figure 4. Left panel: frequency distribution of the PBA for the N. pumilio forests and the N. antarctica shrublands. Different symbols in the x-axis show the percentiles of each frequency distribution. For visualization purposes, a square root transformation was applied on the y-axes, and extreme values on the x-axis were excluded. Right panel: PBA percentile differences (± 95% bootstrapping confidence interval) between shrublands and forests. Black symbols represent significant differences (i.e. bootstrapping confidence interval does not include zero); white symbols were used otherwise. The difference between the percentiles is indicated above each point.
4. Discussion
Our results support the hypothesis that N. pumilio forests and N. antarctica shrublands represent fire-driven AVS across a wide geographical and environmental range. Firstly, the extensive niche overlapping of N. pumilio forests and N. antarctica shrublands provides strong evidence that shrublands occur in conditions suitable for forests and vice versa (figure 3), thus supporting the AVS hypothesis [1]. Secondly, in micro-basins with higher PBA, the niche overlapping areas are preferably occupied by N. antarctica shrublands, supporting the hypothesis that those shrublands are maintained by fire (figure 4) [20].
Our results challenge the traditional notion that the N. antarctica shrublands can only thrive in conditions too harsh for the establishment of N. pumilio forests [28]. In fact, the ecological niche of forests and shrublands overlaps across diverse climatic, edaphic and topographic variables (figure 3 and electronic supplementary material, figure S6) that occur over a wide latitudinal range, from meso-Mediterranean (36° S) to subantarctic (54° S) ecosystems (electronic supplementary material, figure S5). Although the three PCA axes used for the reconstruction of ecological niches captured an important part of the environmental variability of the study area, there is a part of this variability that was not considered (31%). Therefore, we cannot rule out a possible overestimation of the niche overlap. However, it is worth mentioning that the environmental conditions of the non-overlapping N. antarctica shrublands (i.e. the part of the niche that does not overlap with that of N. pumilio forests) are consistent with the restrictive conditions traditionally associated with this vegetation type, that is, on flat terrains with soils that are poor in organic carbon and relatively high pH in warmer and drier sites (higher PC1 values), or on undulating terrain at altitude with waterlogged soils (lower PC2 values; figure 3 and electronic supplementary material, figures S4 and S6) [28]. Thus, our modelled ecological niches reflect both the shared conditions between Nothofagus forests and shrublands, but also the conditions exclusive to N. pumilio and N. antarctica.
The higher frequency of N. antarctica shrublands compared to N. pumilio forests in micro-basins with high PBA is consistent with the hypothesis that fire-vegetation feedbacks promote their stability in areas where forests and shrublands represent AVS [38]. These dynamics may have been occurring for a long period as they are dominated by two contrasting woody life forms: a non-resprouting tree and a resprouting shrubby tree. It is also interesting that they are congeneric, and thus it is tempting to suggest that the AVS may have been the appropriate framework for those evolutionary divergences [59].
However, the two states were equally probable in micro-basins without fires during the studied period (percentiles 0.1–0.4 in figure 4). We argue that this is due to the long history of human-ignited fires in the region and the slow dynamics of this vegetation type. The N. antarctica shrublands expanded in the past due to the use of fire by native settlements and during the Euro-Chilean colonization [32,60], being a common practice until the middle of the twentieth century for agricultural purposes [61]. Therefore, fires prior to the period assessed (i.e. 1985−2018) could have promoted the establishment of shrublands in the current ‘NoFire’ micro-basins, most of them located in the southernmost areas of the study area (figure 1 and electronic supplementary material, figure S2). Both the scarcity of climatic conditions conducive to fires in these areas and the fire-suppression policies established in the country since 1980 would have prevented them from burning during the period evaluated. However, the time elapsed was not enough for the establishment of N. pumilio, which requires between 40 and 80 years to form forests after fires [62], explaining why this state is not dominant in micro-basins without recent fires. Field studies are needed to evaluate the presence of N. pumilio saplings in shrublands of ‘NoFire’ micro-basins. In any case, the fact that N. pumilio forests predominate throughout the study area, being a region with low fire activity (figure 1) evidences the stability of these forests in the absence of fire.
5. Conclusion
Previous studies used field approaches to demonstrate the existence of fire-vegetation feedbacks that stabilize forests and shrublands in the dry and fire-prone region of northeastern Patagonia [20,21,31,34]. Here, we have provided evidence for the existence of fire-driven AVS across the temperate Andes. Using an ecological niche modelling approach, we demonstrated that N. pumilio forests and N. antarctica shrublands can develop under the same set of environmental conditions. Additionally, our results give support for the role of fire in explaining the vegetation distribution patterns over a wide geographical and environmental range, from meso-Mediterranean to subantarctic environments spanning 18° of latitude. Therefore, we provide evidence of the importance of fire in the vegetation dynamics even under cold-wet conditions typically considered non-fire prone. Our study contributes to the understanding of the importance of fire in the vegetation dynamics of the temperate Andes and generates new evidence for the existence of fire-driven AVS beyond the tropics in South America [10,37]. Altogether, this evidence supports the key role of fire-driven AVS as a mechanism for shaping vegetation patterns at the continental scale.
Ethics
This work did not require ethical approval from a human subject or animal welfare committee.
Data accessibility
All data were obtained from published sources as indicated in §2. Our aggregated data are openly available from the Dryad Digital Repository [63].
Electronic supplementary material is available online [64].
Declaration of AI use
We have not used AI-assisted technologies in creating this article.
Authors’ contributions
D.P.R.: conceptualization, data curation, formal analysis, investigation, methodology, writing—original draft, writing—review and editing; S.A.E.: conceptualization, formal analysis, investigation, methodology, writing—original draft, writing—review and editing; A.M.: data curation, investigation, methodology, writing—review and editing; J.G.P.: conceptualization, investigation, writing—review and editing; S.P.: conceptualization, formal analysis, funding acquisition, investigation, methodology, writing—original draft, writing—review and editing.
All authors gave final approval for publication and agreed to be held accountable for the work performed therein.
Conflict of interests declaration
We declare we have no competing interests.
Funding
This research was funded by the ANID FONDECYT grant no. 1190999 (Chile). S.P. and D.P.R. were funded by the ANID PIA/BASAL FB210006 (Chile). D.P.R. was additionally funded by Anglo American Foundation. J.G.P. contribution was framed within the DISTEPIC project (grant no. PID2022-141530NB-C21, Ministerio de Ciencia e Innovación, Spain). A.M. thanks ANID FONDAP/15110009, ANID/CONICYT FONDECYT iniciación (grant no. 11240356) and Dirección de Investigación of Universidad de La Frontera. S.A.E. was funded by ANID PIA/BASAL AFB240003.
