Skip to main content
Advertisement
  • Loading metrics

High-velocity upward shifts in vegetation are ubiquitous in mountains of western North America

  • James R. Kellner ,

    Roles Conceptualization, Formal analysis, Writing – original draft, Writing – review & editing

    james_r_kellner@brown.edu

    Affiliations Department of Ecology, Evolution and Organismal Biology, Brown University, Providence, Rhode Island, United States of America, Institute at Brown for Environment and Society, Brown University, Providence, Rhode Island, United States of America

  • Joseph Kendrick,

    Roles Formal analysis

    Affiliations Department of Ecology, Evolution and Organismal Biology, Brown University, Providence, Rhode Island, United States of America, Institute at Brown for Environment and Society, Brown University, Providence, Rhode Island, United States of America

  • Dov F. Sax

    Roles Conceptualization, Writing – original draft, Writing – review & editing

    Affiliations Department of Ecology, Evolution and Organismal Biology, Brown University, Providence, Rhode Island, United States of America, Institute at Brown for Environment and Society, Brown University, Providence, Rhode Island, United States of America

Abstract

The velocity of climate change and its subsequent impact on vegetation has been well characterized at high elevations and latitudes, including the Arctic. But whether species and ecosystems are keeping pace with the velocity of temperature change is not as well documented. Some evidence indicates that species are less able to keep pace with the velocity of climate change along elevational gradients than latitudinal ones. If substantiated this finding could warrant reconsideration of a current cornerstone of conservation planning. Here we use 27 years of high-resolution satellite data to quantify changes in vegetation cover across elevation within nine mountain ranges in western North America, spanning tropical Mexico to subarctic Canada and from coastal California to interior deserts. Across these ranges we show a uniform pattern at the highest elevations in each range, where increases in vegetation have occurred ubiquitously over the past three decades. At these highest elevations, the realized velocity of vegetation varies among mountain ranges from 19.8–112.8 m · decade-1 (mean = 67.3 m · decade-1). This is equivalent, with respect to gradients in temperature, to a 14.4–104.3 km · decade-1 poleward shift (mean = 56.1 km · decade-1). This realized velocity is 4.4 times larger than previously reported for plants, and is among the fastest rates predicted for the velocity of climate change. However, in three of the five mountain ranges with long-term climate data, realized velocities fail to keep pace with changes in temperature, a finding with important implications for conservation of biological diversity.

Introduction

A cornerstone of the global strategy to conserve biodiversity involves protecting mountainous regions [1, 2]. Doing this is important because much of the Earth’s biodiversity is found in montane regions, such as the Andes [2, 3], but also because mountains offer the potential of buffering species from some aspects of climate change [4]. In particular, because montane regions have steep topographic gradients, they have lower overall velocities of climate change than topographically simpler, flatter areas [5]. This means that within mountains, many species would need to shift their distributions only small distances along elevation gradients in order to retain temperatures they experienced historically [6]. Indeed, in response to existing climate change, a great variety of species have already shifted their distributions to higher elevations [79]. However, some shifts in biotic responses with montane regions seem to be exceedingly slow, in some cases an order of magnitude slower than climate change [7]. These shifts seem to be slower overall for plants than for many animal groups [8, 10]. Available evidence suggests that some species are less able to keep pace with the velocity of change along gradients of elevation than latitudinal ones [11]. If additional evidence bears out these sorts of observations then it would call into question our understanding of biotic responses, particularly with respect to velocities of climate change.

In both montane and non-montane regions, understanding the velocity of climate change and corresponding shifts in the distribution of biota have become important aspects of conservation planning [6, 12]. Velocities are derived from changes in climate anticipated or observed across spatial gradients, and describe the instantaneous local rate of change needed to maintain constant conditions (km · yr-1). Current and anticipated velocities of climate change are complex, with substantial variation among and within regions [5, 6]. Mountainous regions in particular differ from topographically homogeneous areas [6, 13]. Velocity itself can be measured against a host of possible factors, and recent work has focused not just on changes in temperature, but on interactions between temperature and water availability [5, 13]. To describe the response of biota, here we use the term ‘realized velocity’, which we define as the rate of change in geographic distribution that is observed or anticipated to occur. Realized velocities differ from climate velocities, because they integrate how organisms respond to changes in the environment, rather than changes in the environment alone. These responses could be less than climate velocity due to intrinsic limitations in dispersal capacity, extrinsic factors such as habitat fragmentation, or tolerances that aren’t matched to changes in climate. Losses of ecosystems or species become more likely when climate velocity exceeds the realized velocity. This occurs when species or ecosystems are not keeping pace with climate change. Available evidence shows that some species and ecosystems are keeping pace, shifting or expanding their geographic distributions poleward or upward in elevation [6, 9, 11, 14], while others are not doing so [7, 15].

One way to evaluate whether species and ecosystems are keeping pace with climate change, in either montane on non-montane regions, is to examine the geographic distribution of vegetation over time and across spatial gradients using high-resolution remote sensing [16, 17]. Quantifying how vegetation is changing facilitates examination of key questions important for conservation of biological resources in response to climate change, such as whether changes in climate observed to date are leading to an increase or decrease in the total habitable portion of mountainous areas, and whether shifts can be detected in the position of the ‘green line,’ which we define as the elevation limit or latitudinal limit beyond which little or no vegetation occurs. Although the velocity of climate change and its subsequent impact on vegetation has been well characterized at high elevations and latitudes, including the Arctic [16, 1822], to date there has been relatively limited examination of how the ‘green line’ is shifting within high-elevation montane regions [23]. Exploring this issue, and understanding how biotic responses might lag climatic changes, offers the potential to better understand this important aspect of global change.

Here we use high-resolution satellite data from the Landsat record during the interval 1984–2011 to examine changes in the quantity of vegetation and the realized velocity of change within and among nine mountain ranges in western North America (S1 Fig, S1 Table). We used signal processing methods that decompose surface reflectance into fractional green vegetation, woody vegetation, and barren substrate [24, 25] in each Landsat pixel within these mountain ranges (c. 105–107 pixels per range per year–see Methods; S2 Table). We examine peak-growing-season measurements within 100 m elevation bands and contrast patterns within six elevation divisions (sextiles) on each mountain range. This relative classification (sextiles) within mountain ranges is necessary because ranges differ in absolute elevation. For example, the Colorado Rockies achieve maximum elevations in excess of 4,000 m, whereas the New Mexico Rockies in our sample do not exceed 3,400 m. Within the top sextile, we determined the average elevation across the first three years that matched vegetation cover during the last three years of our record. We used the difference in elevation between these two points, standardized by time, to compute realized velocity (m · yr-1).

Methods

Our analysis is based on a 27-year time series from the Landsat-5 Thematic Mapper data record. Landsat-5 was a multispectral, whiskbroom scanner that collected measurements in the visible, near-infrared, and shortwave-infrared regions of the electromagnetic spectrum at a ground sample distance of 30 m with a 16-day nominal revisit time. We use peak-growing season observations (August) of the Landsat-5 Surface Reflectance (LEDAPS) data product [26]. We selected peak-growing season observations to minimize snow cover that is present at high elevation in some mountain ranges, and to standardize vegetation phenology. Each Landsat-5 scene was approximately 170 km (north-south) × 185 km (east-west) in size. We excluded scenes if > 30% of the pixels were flagged as cloudy by the CFMask algorithm.

Study areas

We identified nine mountain ranges in western North America (S1 Fig, S1 Table). The ranges included sections of the Rocky Mountains in northern British Columbia, southern Alberta, Wyoming, and Colorado, sections of the Great Basin Ranges in Utah and New Mexico, the Sierra Nevada, a section of the Sierra Madre Oriental, and a section of the Sierra Madre Occidental. These ranges were selected because they sample the major mountain ranges in western North America and have frequent cloud-free Landsat-5 data coverage. We selected grid cells from the Landsat Worldwide Reference System 2 that encompassed the extent of each mountain range. We defined the boundaries of each range using the Nature Conservancy World Conservation Atlas (S1 Fig) [27]. Our analysis of remote sensing data is restricted to the extent of each mountain range within these boundaries.

Analysis

We quantified an index related to fractional cover of photosynthetic vegetation, non-photosynthetic vegetation, and barren substrate using spectral mixture analysis (SMA) [24, 25]. SMA is a signal-processing method that decomposes a surface reflectance spectrum into a linear combination of endmembers. We generated a spectral endmember library from a randomly selected image from the Sierra Nevada (Landsat-5 scene ID LT50420352010228). To identify endmember spectra, we used principal components analysis to identify multivariate outliers. These outliers represent pure endmembers [28]. We classified these endmembers into thematic classes, most of which represented photosynthetic vegetation, non-photosynthetic vegetation, or barren substrate (we omitted endmembers from classes that were infrequent and not of interest to the study, like roads and water). We computed the arithmetic mean of all endmembers within classes, and these aggregated spectra were used for SMA. The linear mixture model is: (1) Where X is the unknown index related to fractional cover for each endmember to be estimated, A is an M × N endmember library matrix, where M is the number of bands and N is the number of endmembers. B is an M × 1 observed data vector (the reflectance observations for a given Landsat-5 pixel). We solved for A−1 using singular value decomposition [28].

For every pixel in the analysis, we computed the root mean-squared error (RMSE) and the shade fraction, which we defined as the difference between 1 and the sum of the estimated endmember fractions [25]. We then excluded any pixel with RMSE > 0.025 or a shade fraction > 0.8. These thresholds were identified empirically and resulted in removal of many pixels containing snow, ice, water, roads, and other materials outside the scope of interest of this study. Finally, we normalized the values of the three endmembers so that they summed to one. Our analyses are based on the fractions for total vegetation (i.e. the sum of photosynthetic and non-photosynthetic vegetation).

We extracted the ground elevation for every Landsat pixel from the Shuttle Radar Topography Mission (SRTM) 30 m resolution digital elevation model [29]. We then computed the weighted mean vegetation cover within each combination of mountain range, 100 m elevation band, and year. Weights were proportional to the number of pixels within each combined mountain range, elevation band and year. For each mountain range and year combination, any elevation band that contained < 1,000 valid pixels was omitted. To determine how changes in the index of vegetation cover over time depended on elevation within and among mountain ranges, we binned 100 m elevation bands into six equal portions (sextiles) within each range and computed the unweighted mean index of total vegetation within each sextile. This relative classification within mountain ranges is necessary because mountain ranges differ in absolute elevation. We computed changes in the index of total vegetation over time within each mountain range and sextile using ordinary linear regression.

Preliminary analysis indicated that vegetation cover increased in the top sextile in all nine mountain ranges examined, and that the top sextile was the only elevation sextile that demonstrated consistent trends in vegetation cover among the nine mountain ranges. To determine whether the association of nine positive trends in one sextile was statistically significant, we computed the binomial probability of observing 9 positive trends in vegetation cover exclusively in the top sextile, given a probability of success of 0.740. The probability of success is the fraction of positive trends in vegetation cover in the entire data set (40/54 = 0.740), where 54 is the total number of sextiles examined (i.e. 6 sextiles in 9 mountain ranges).

Statistical analysis of climate data

Because climate data were exclusively or predominantly available for the coterminous United States, we restricted analyses of climate data to the five USA mountain ranges (Colorado, Nevada, New Mexico, Utah, and Wyoming). Within each of these ranges we quantified changes over time in mean annual temperature and precipitation in the top sextile of each mountain range, which we obtained by overlaying PRISM climate data product AN81m on each USA mountain range in a Geographic Information System [30]. The historical temperature record analyzed here is from 1895–2015, and mean annual precipitation is from 1981–2015.

Interpreting trends over multiple decades using PRISM AN81m must be done with caution, because differences in methodologies and station identity could confound temperature trends. This is especially true in complex mountainous terrain of western North America, where weather station data are sparse and largely unavailable at high elevations and on steep slopes [31]. One way to avoid these limitations is to analyze data from multiple independent sources, since it is unlikely that measurement artifacts related to weather station turnover or instrument changes would simultaneously influence different weather stations in completely different geographic locations, like the five USA mountain ranges considered here. Given these caveats, our analysis of the long-term record indicates a period of relative stability or cooling over the last century and a transition to steadily increasing temperatures in each of the five USA mountain ranges. We estimated the timing of this transition and the change in mean annual temperature · year-1 using breakpoint regression [32]. We quantified the change in total annual precipitation · year-1 using ordinary linear regression.

Quantifying realized velocity

We quantified realized velocity in the top sextile of each USA mountain range in a series of steps. The first step was regressing elevation on mean vegetation cover over the final three years of the data record (S2S6 Figs; Table 1). For all mountain ranges except British Columbia, the final three years were 2009–2011. For British Columbia, the final three years were 2008–2010. We used these relationships to determine the elevations at the beginning of the data record in each mountain range associated with vegetation cover observed during the final three years (i.e. given a relationship between vegetation cover and elevation during the final three years of observation, at what elevation was the same vegetation cover observed during the first three years of observation?). The difference between the predicted and observed values is the change in elevation associated with a given vegetation cover over the duration of the study. Dividing this quantity by the number of observation years in each data record standardizes elevation gains per unit time and is realized velocity (m · decade-1). The first year of observation for each mountain range was 1984 in Alberta, British Columbia, and the Sierra Nevada, 1985 in Colorado, New Mexico, Utah and Wyoming, 1991 in the Sierra Madre Oriental, and 1993 in the Sierra Madre Occidental. We computed the realized velocity for each 100 m elevation band in the top sextile, and then averaged across the 100 m elevation bands (Table 2).

thumbnail
Table 1. Regression coefficients relating elevation in m (response variable) to an index of vegetation cover (predictor variable) in five USA mountain ranges.

https://doi.org/10.1371/journal.pclm.0000071.t001

thumbnail
Table 2. Realized velocities in the geographic distribution of vegetation at high elevation in five North American mountain ranges.

ΔT is the observed temperature change, and realized velocity is the observed change in elevation of vegetation cover per decade. The temperature-elevation regression trend is the rate of change in temperature per meter of elevation. Climate velocity is the velocity necessary for vegetation to keep pace with ΔT given the temperature-elevation regression trend. The right column is the latitude-equivalent velocity assuming 120.9 km · °C-1 at sea level.

https://doi.org/10.1371/journal.pclm.0000071.t002

Predicting realized velocity given changes in temperature

To quantify whether a given change in the geographic distribution of vegetation is keeping pace with changes in temperature, we determined the velocity of elevational migration using the equation, (2) Comparing these predicted velocities to realized velocities is a test of the hypothesis that vegetation is keeping pace with changes in temperature at high elevation in western North America. The terms on the right are the rate of change in temperature in the top elevation sextile (°C · decade−1), and the rate of change in elevation (meters) per unit temperature (°C). The product of these quotients is predicted velocity in units of meters per decade. We computed the rate of change in temperature per decade using the PRISM AN81m data product for the top elevation sextile in each of the five USA mountain ranges (Fig 1; Table 3). To determine the rate of change in elevation per unit temperature we used temperature-elevation regressions (S7S11 Figs; Table 4). Temperature-elevation regressions are the central component of the PRISM algorithm and widely used to quantify spatial gradients in temperature [30, 31]. The approach assumes that temperature changes consistently with elevation. We accessed weather station data for August, 2011 from the National Oceanic and Atmospheric Administration National Centers for Environmental Information for the US States of Colorado, New Mexico, Utah and Wyoming, and for the Sierra Nevada Mountain Range in California and Nevada. Within each region we fit break-point and standard linear regressions to elevation as a function of temperature (S7S11 Figs). We then calculated predicted velocity for each mountain range using Eq 2. To evaluate whether our conclusions were sensitive to the form of the statistical model, we computed predicted velocities in two ways. In the first analysis, we used the slope from the standard linear model (Table 5). In the second analysis, we used the slope from the high-elevation portion of the break-point temperature-elevation trend (Table 4). In Table 2 we report results based on the elevation-temperature trend that predicted the smallest climate velocity, which is conservative with respect to the hypothesis that vegetation is failing to keep pace with changes in temperature, because a smaller climate velocity is easier for vegetation to track than a larger climate velocity.

thumbnail
Fig 1. Trends in mean annual temperature at high elevation in five USA mountain ranges over 120 years.

The data are means in the top elevation sextile for New Mexico, Sierra Nevada, Utah and Wyoming, and the fifth sextile for Colorado (for which data are lacking in the top sextile). Grey lines are best-fit segmented linear regressions, which indicate that significant warming commenced in 1991, 1976, 1978, 1984, and 1993 in Colorado, New Mexico, Sierra Nevada, Utah and Wyoming ranges, respectively. Points before the estimated breakpoint are red, and points after are black. The grey box is the time interval of our Landsat analysis.

https://doi.org/10.1371/journal.pclm.0000071.g001

thumbnail
Table 3. Break-point regression coefficients relating mean annual temperature in °C (response) to time in years (predictor) for each of the five USA mountain ranges.

The data are from PRISM data product AN81m. Slope coefficients are the trend in the relationship before and after the break point (cf. Fig 1). The coefficient of determination and P value refer exclusively to the linear trend after the breakpoint.

https://doi.org/10.1371/journal.pclm.0000071.t003

thumbnail
Table 4. Break-point regression coefficients relating mean August temperature in °C in 2011 (response) to elevation in m (predictor) for each of the five USA mountain ranges.

The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information. Slope coefficients are the trend in the relationship before and after the break point (cf. Fig 1). n is the number of weather stations included in each analysis.

https://doi.org/10.1371/journal.pclm.0000071.t004

thumbnail
Table 5. Regression coefficients relating mean August temperature in °C in 2011 (response) to elevation in m (predictor) for each of the five USA mountain ranges.

The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information. n is the number of weather stations included in each analysis.

https://doi.org/10.1371/journal.pclm.0000071.t005

We calculated the latitude-equivalent realized velocity using the equation, (3) The first term on the right is the realized velocity (Table 2). The second term is the reciprocal of the elevation-temperature regression trend (Tables 4 and 5). The final term is the reciprocal of the rate of change in temperature per kilometer of latitude. We assume 120.9 km · °C-1 at sea level.

Results

There is substantial variation in vegetation cover within and among mountain ranges. As an exemplar of within-range variation, we provide a subset encompassing 2,600 km2 from the Sierra Nevada (Fig 2). In this scene, vegetation cover is greater on western slopes, which are mesic, than on eastern slopes, which are xeric due to a rain-shadow [33]. Vegetation cover also varies strongly with elevation in the Sierra Nevada, such that cover increases moderately with increasing elevation, likely due to increases in water availability, before beginning a precipitous decrease at high elevation (Fig 3). Mountain ranges vary in their relationships between elevation and vegetation cover, with subtropical mountains showing relatively large amounts of cover even at high elevations, whereas subarctic ones have a nearly complete loss of vegetation at high elevations (Fig 3). Other differences in vegetation patterns among mountains also exist, such as the variation in the strength of the inflection point of decreasing vegetation with increasing elevation, as is apparent between the Sierra Nevada and Utah ranges (Fig 3).

thumbnail
Fig 2. Change in vegetation cover over 27 years in the Sierra Nevada Mountains.

Spectral mixture analysis of Landsat-5 from 1984 and 2011. The data are false-color composites with barren substrate red, photosynthetic vegetation green, and non-photosynthetic vegetation blue. Each image is overlaid on 30 m topography from the Shuttle Radar Topography Mission. The scale bar is 10 km. The Landsat and SRTM data used to generate this figure are here and here [26, 29].

https://doi.org/10.1371/journal.pclm.0000071.g002

thumbnail
Fig 3. Elevation profiles of vegetation cover in nine mountain ranges in western North America.

Colored points and lines correspond to individual mountain ranges: AB (Alberta), BC (British Columbia), CO (Colorado), NM (New Mexico), SMOC (Sierra Madre Occidental), SMOR (Sierra Madre Oriental), SN (Sierra Nevada), UT (Utah), WY (Wyoming). The data are from 2011 in all mountain ranges except BC, which are from 2010.

https://doi.org/10.1371/journal.pclm.0000071.g003

Vegetation cover has changed on these mountains over the previous three decades. Within individual mountains, most pixels show little to no directional change over time, but differences are apparent at high elevations, where total vegetation cover has increased (e.g., within the Sierra Nevada, Fig 2). Within elevation bands, there are pronounced changes over time within and among mountain ranges (Fig 4). Some mountains, such as the Sierra Nevada, exhibit a small increase in vegetation cover at all elevations, whereas others, such as the New Mexico and Wyoming ranges, show pronounced variation across elevations. Within elevation sextiles, there is also pronounced variation among mountain ranges at most elevations, with the exception of the highest sextile, which is the only elevation zone to exhibit consistent changes in vegetation cover among the nine mountain ranges examined. The binomial probability of observing 9 positive trends in vegetation cover exclusively in the top sextile is statistically significant (P = 0.052). Although these increases in vegetation in the top sextile are uniform across mountains, there is significant variation in the rate of that change. Indeed, the realized velocity of vegetation among the five USA mountain ranges is as small as 19.8 m · decade-1 in Wyoming, and as large as 112.8 m · decade-1 in New Mexico (Table 2). To test the hypothesis that realized velocities are keeping pace with changes in temperature, we computed temperature-elevation regressions and used these relationships to determine the elevation change required to track observed warming over the duration of our study (Tables 4 and 5; S7S11 Figs). This analysis indicates that the realized velocity in only two mountain ranges–New Mexico and the Sierra Nevada–has kept pace with observed temperature changes, while vegetation in Colorado, Utah and Wyoming did not keep pace with temperature change (Table 2).

thumbnail
Fig 4. Trends in vegetation cover within elevation sextiles and nine mountain ranges in western North America.

The thick pink line shows the mean trend. Colored points and lines correspond to individual mountain ranges: AB (Alberta), BC (British Columbia), CO (Colorado), NM (New Mexico), SMOC (Sierra Madre Occidental), SMOR (Sierra Madre Oriental), SN (Sierra Nevada), UT (Utah), WY (Wyoming). The top sextile shows uniformly increasing vegetation cover within all nine mountain ranges.

https://doi.org/10.1371/journal.pclm.0000071.g004

There have been changes in the climate of these montane regions over the study period. Although warming has occurred in North America over much of the past 200 years [34], a break-point regression analysis of annual temperature from the late 1800s onwards shows that significant warming at high elevations began between 1975 and 1991 for each of the five USA mountain ranges (Fig 1, Table 3). High elevations on the five USA mountain ranges have also experienced significant drying (S12 Fig, S3 Table).

Discussion

The uniformity and overall speed of upward shifts in vegetation at the highest elevations across so many mountains that vary considerably in geographic context is remarkable. Indeed, on two of the mountains examined, realized velocity of vegetation kept pace with velocity of temperature change. These fastest realized velocities (112.8 m · decade-1) for vegetation are as fast as some of the most rapid animal migrations observed in montane regions. For instance, a recent analysis [9] examined 68 bird species in the Andes and found that shifts in the upper-limit of only a few species approached 112.8 m · decade-1. A recent review of species elevational shifts in montane regions, which considered many plant and animal species, found an average upward shift of 25 m · decade-1 over recent decades [35]. In some extreme cases in the Pyrenees, plant species have shifted their distributions upward by 500 m in elevation over more than a century, though average shifts of plant species in these mountains were closer to 200 m [36]. In general, plant species in montane regions show relatively slow shifts in upward distribution; for example, plant species in the Alps had average upward shifts of 20 m over more than a half-century [10]. Even the slowest shift we observed in upward migration of vegetation (19.8 m per decade) still greatly exceeds rates recorded for some individual species [10]. The upward shifts we observed can be further understood by considering the equivalent latitudinal migration within a topographically flat region at sea-level, equivalent to a 14.4–104.3 km · decade-1 poleward shift (mean = 56.1 km · decade-1). The changes we observed are on par with the fastest velocities predicted to occur due to abiotic change alone [6].

Our findings help to characterize patterns of vegetation across elevations in mountains that range from the sub-tropics to the sub-arctic. These vegetation by elevation profiles provide a modern point of comparison with pioneering, centuries-old work of von Humboldt [37], who characterized vegetation across elevations on a high tropical mountain, and with foundational work on gradient analysis [38], but allow now for the characterization of subtle differences among mountains, such as the variation in the strength of the inflection point of decreasing vegetation with increasing elevation. Our work is consistent with a variety of recent studies that take advantage of remotely sensed data to investigate how montane vegetation is related to climate [17, 39]. Although much of this work recapitulates previously known patterns, such as the presence of relatively large amounts of vegetation at high elevations in sub-tropical mountains and the complete absence of vegetation at high elevation in high-latitude mountains, it provides a level of detail in these relationships that was not attainable prior to the era of high-resolution remote sensing. Our analysis extends previous efforts by explicitly testing the hypothesis that vegetation is keeping pace with changes in temperature at high elevation.

The upward shifts in vegetation at the highest elevations could be due to a variety of factors. Although our remote sensing record for vegetation, beginning in 1984, captures most of the period of pronounced warming at these mountains over the past century, identifying a single or suite of causal abiotic factors or other conditions responsible for the observed changes in vegetation at high elevations is difficult. This is likely due in part to severe limitations in availability of climate and other historical data at these elevations, where few weather stations exist [31], as this limits our ability to relate fine-scale changes in vegetation with equally fine-scale climate data. It is also likely due, however, to the complexity of climatic factors and historical land use that may be responsible for these shifts. Indeed, the combination of warming and drying we observed may have pushed vegetation in opposing directions and could be one reason why some realized velocities are failing to keep pace with temperature change [13]. Realized velocity of vegetation shifts can also be influenced by lag processes in the creation of suitable substrate, e.g., through loss of snow cover or creation of soil [40]. Although we did not investigate the contribution of slope or aspect, we believe that topographic differences are unlikely to explain the primary conclusion of our analysis. This is because our analysis is based on very large samples of 105–107 pixels per range per year (see Methods; S2 Table) that vary in slope and aspect, both within and between mountain ranges and individual mountains; further, recent work at high elevations on the Tibetan Plateau showed that topographic slope had some predictive power, but aspect had little impact on changes in vegetation greenness measured using remote sensing, despite overall changes in vegetation cover that were consistent with climate change [41]. Ultimately, finer-scale, more accurate and ubiquitous climate data at upper elevations would help to better disentangle alternative mechanisms for biotic changes in montane systems.

Although the shifts in vegetation we observed at the highest elevations are likely to have been driven by changes in climate, it is conceivable that some aspect of human disturbance or land use change has played a role in driving these patterns. This seems most likely to be a relevant consideration within the two regions observed in Mexico, where even highest elevations are located in close proximity to human settlements. Within the mountains in the United States and Canada, changes in human land use seem unlikely to be driving patterns of vegetation change within the top sextile of elevations. For example, in the Sierra Nevada, the top sextile starts at elevations above 3,600 m (11,811 feet). Although there is a history of high-altitude cattle and sheep grazing in the Sierra Nevada, most of this grazing was probably at elevations less than 3,600 m. Historical grazing in Sierra Nevada meadows of 9,000 or 10,000 feet occurred primarily between 1870–1920, with almost no grazing at high elevations after 1946 [42]. This indicates that grazing history is unlikely to drive recently observed changes in vegetation. Consequently, although some potential influence of human land use cannot be ruled out, it seems unlikely to be a general driver of the uniform patterns of change in vegetation observed in the highest elevations among mountains in western North America.

Our findings show that at high elevations, vegetation cover as a whole was able to expand upward at a rapid rate, even if that rate failed to keep pace with changes in temperature in some cases. Our findings are broadly consistent with recent work showing large increases globally in the fraction of land at the highest latitudes that are becoming vegetated [43]. Our findings also confirm the generality of a host of recent studies that have documented shifts in elevation among plant species on individual mountains [13, 44, 45], but provide mixed news for conservation in the context of climate change [46]. If vegetation at the highest elevations can’t keep pace with the velocity of climate change on some mountains, then high-elevation species and ecosystems in those places may eventually be squeezed at their low-elevation limits by species moving uphill and at their high-elevation limits, not just by declines in total surface-area found in many mountain ranges [28], but also by a failure to occupy available habitat at these highest elevations. On the other hand, in places where vegetation shifts are keeping pace with climate velocity, the expanding available habitat may provide the conditions needed for the long-term survival of high-elevation species. More work is needed to better contextualize how the vegetation shifts we observed, which measure any green vegetation, relate to individual species movements and broader ecosystem transitions–as this sort of integration offers the potential to create a more holistic understanding of change in montane systems.

Conclusion

At the highest elevations and across mountain ranges in western North America, vegetation has shifted upward in elevation over the past few decades. Available evidence suggests that these shifts are primarily driven by warming that has occurred in western North America beginning in the 1980s. The realized velocity of these changes in vegetation is keeping pace with climate velocity on some mountains. Even on those mountains where the realized velocity of vegetation change is slower, these high-elevation changes are still as fast or faster than the rate of most species-level elevational shifts that have been documented worldwide. Still, in cases where vegetation shifts are not keeping pace with climate change, it is likely that the total area of suitable habitat for some species will decline, putting them at increased risk of extinction.

Supporting information

S1 Fig. Geographic distribution and sample areas for nine mountain ranges in western North America.

AB (Alberta), BC (British Columbia), CO (Colorado), NM (New Mexico), SMOC (Sierra Madre Occidental), SMOR (Sierra Madre Oriental), SN (Sierra Nevada), UT (Utah), WY (Wyoming). The base layer for this map is from the Database of Global Administrative Areas.

https://doi.org/10.1371/journal.pclm.0000071.s001

(TIF)

S2 Fig. Relationship between elevation and vegetation cover in the top elevation sextile in the Colorado Rocky Mountains.

Black points are the mean vegetation cover in each elevation band during the first three years of observation. Red points are the mean vegetation cover in each elevation band during the last three years of observation. The black line is the predicted value from linear regression.

https://doi.org/10.1371/journal.pclm.0000071.s002

(TIF)

S3 Fig. Relationship between elevation and vegetation cover in the top elevation sextile in the New Mexico Rocky Mountains.

Black points are the mean vegetation cover in each elevation band during the first three years of observation. Red points are the mean vegetation cover in each elevation band during the last three years of observation. The black line is the predicted value from linear regression.

https://doi.org/10.1371/journal.pclm.0000071.s003

(TIF)

S4 Fig. Relationship between elevation and vegetation cover in the top elevation sextile in the Sierra Nevada Mountains.

Black points are the mean vegetation cover in each elevation band during the first three years of observation. Red points are the mean vegetation cover in each elevation band during the last three years of observation. The black line is the predicted value from linear regression.

https://doi.org/10.1371/journal.pclm.0000071.s004

(TIF)

S5 Fig. Relationship between elevation and vegetation cover in the top elevation sextile in the Utah Rocky Mountains.

Black points are the mean vegetation cover in each elevation band during the first three years of observation. Red points are the mean vegetation cover in each elevation band during the last three years of observation. The black line is the predicted value from linear regression.

https://doi.org/10.1371/journal.pclm.0000071.s005

(TIF)

S6 Fig. Relationship between elevation and vegetation cover in the top elevation sextile in the Wyoming Rocky Mountains.

Black points are the mean vegetation cover in each elevation band during the first three years of observation. Red points are the mean vegetation cover in each elevation band during the last three years of observation. The black line is the predicted value from linear regression.

https://doi.org/10.1371/journal.pclm.0000071.s006

(TIF)

S7 Fig. Relationship between August mean temperature and elevation in the Colorado Rocky Mountains.

The red line is the break-point regression, and the blue line is the linear regression. The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information (formerly Climate Data Online). Coefficients and sample sizes are in Tables 4 and 5.

https://doi.org/10.1371/journal.pclm.0000071.s007

(TIF)

S8 Fig. Relationship between August mean temperature and elevation in the New Mexico Rocky Mountains.

The red line is the break-point regression, and the blue line is the linear regression. The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information (formerly Climate Data Online). Coefficients and sample sizes are in Tables 4 and 5.

https://doi.org/10.1371/journal.pclm.0000071.s008

(TIF)

S9 Fig. Relationship between August mean temperature and elevation in the Sierra Nevada Mountains.

The red line is the break-point regression, and the blue line is the linear regression. The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information (formerly Climate Data Online). Coefficients and sample sizes are in Tables 4 and 5.

https://doi.org/10.1371/journal.pclm.0000071.s009

(TIF)

S10 Fig. Relationship between August mean temperature and elevation in the Utah Rocky Mountains.

The red line is the break-point regression, and the blue line is the linear regression. The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information (formerly Climate Data Online). Coefficients and sample sizes are in Tables 4 and 5.

https://doi.org/10.1371/journal.pclm.0000071.s010

(TIF)

S11 Fig. Relationship between August mean temperature and elevation in the Wyoming Rocky Mountains.

The red line is the break-point regression, and the blue line is the linear regression. The data are from the National Oceanic and Atmospheric Administration National Centers for Environmental Information (formerly Climate Data Online). Coefficients and sample sizes are in Tables 4 and 5.

https://doi.org/10.1371/journal.pclm.0000071.s011

(TIF)

S12 Fig. Trends in mean annual precipitation at high elevation in five USA mountain ranges over 1981–2015.

The figure is scaled to match the 120-year temperature record. The data are means in the top elevation sextile for New Mexico, Sierra Nevada, Utah and Wyoming, and the fifth sextile for Colorado (for which data are lacking in the top sextile). Grey lines are best-fit linear regressions, which indicate significant drying in all five mountain ranges. Points before the estimated temperature breakpoint are black. Points after are red. The grey box is the time interval of our Landsat analysis.

https://doi.org/10.1371/journal.pclm.0000071.s012

(TIF)

S1 Table. The area measured in each of nine mountain ranges in western North America.

https://doi.org/10.1371/journal.pclm.0000071.s013

(CSV)

S2 Table. Sample sizes (number of Landsat pixels) in each elevation sextile in nine mountain ranges in western North America.

Each pixel is 30 × 30 m.

https://doi.org/10.1371/journal.pclm.0000071.s014

(CSV)

S3 Table. Regression coefficients relating total annual precipitation in mm (response) to time in years (predictor) in five mountain ranges in western North America.

The data are from PRISM data product AN81m.

https://doi.org/10.1371/journal.pclm.0000071.s015

(CSV)

Acknowledgments

We thank the Institute at Brown for Environment and Society (IBES), KC Cushman, D. L. Perret, and C. E. Silva for comments that improved this analysis.

References

  1. 1. Kollmair M, Gurung GS, Hurni K, Maselli D. Mountains: Special places to be protected? An analysis of worldwide nature conservation efforts in mountains. Int J Biodivers Sci Manag. 2005;1: 181–189.
  2. 2. Bax V, Francesconi W. Conservation gaps and priorities in the Tropical Andes biodiversity hotspot: Implications for the expansion of protected areas. J Environ Manage. 2019;232: 387–396. pmid:30500702
  3. 3. Myers N, Mittermeier RA, Mittermeier CG, da Fonseca GAB, Kent J. Biodiversity hotspots for conservation priorities. Nature. 2000;403: 853–858. pmid:10706275
  4. 4. Trew BT, Maclean IMD. Vulnerability of global biodiversity hotspots to climate change. Glob Ecol Biogeogr. 2021;30: 768–783.
  5. 5. Dobrowski SZ, Abatzoglou J, Swanson AK, Greenberg JA, Mynsberge AR, Holden ZA, et al. The climate velocity of the contiguous United States during the 20th century. Glob Change Biol. 2013;19: 241–251. pmid:23504735
  6. 6. Loarie SR, Duffy PB, Hamilton H, Asner GP, Field CB, Ackerly DD. The velocity of climate change. Nature. 2009;462: 1052–1055. pmid:20033047
  7. 7. Fadrique B, Báez S, Duque Á, Malizia A, Blundo C, Carilla J, et al. Widespread but heterogeneous responses of Andean forests to climate change. Nature. 2018;564: 207–212. pmid:30429613
  8. 8. Mamantov MA, Gibson-Reinemer DK, Linck EB, Sheldon KS. Climate-driven range shifts of montane species vary with elevation. Glob Ecol Biogeogr. 2021;30: 784–794.
  9. 9. Freeman BG, Scholer MN, Ruiz-Gutierrez V, Fitzpatrick JW. Climate change causes upslope shifts and mountaintop extirpations in a tropical bird community. Proc Natl Acad Sci. 2018;115: 11982–11987. pmid:30373825
  10. 10. Rumpf SB, Hülber K, Klonner G, Moser D, Schütz M, Wessely J, et al. Range dynamics of mountain plants decrease with elevation. Proc Natl Acad Sci U S A. 2018;115: 1848–1853. pmid:29378939
  11. 11. Chen I-C, Hill JK, Ohlemüller R, Roy DB, Thomas CD. Rapid range shifts of species associated with high levels of climate warming. Science. 2011;333: 1024–1026. pmid:21852500
  12. 12. Urban MC. Accelerating extinction risk from climate change. Science. 2015;348: 571–573.
  13. 13. Juvik JO, Rodomsky BT, Price JP, Hansen EW, Kueffer C. “The upper limits of vegetation on Mauna Loa, Hawaii”: a 50th‐anniversary reassessment. Ecology. 2011;92: 518–525. pmid:21618930
  14. 14. Lesica P, Crone EE. Arctic and boreal plant species decline at their southern range limits in the Rocky Mountains. Ecol Lett. 2017;20: 166–174. pmid:28000369
  15. 15. Zhu K, Woodall CW, Clark JS. Failure to migrate: lack of tree range expansion in response to climate change. Glob Change Biol. 2012;18: 1042–1052.
  16. 16. Cavanaugh KC, Kellner JR, Forde AJ, Gruner DS, Parker JD, Rodriguez W, et al. Poleward expansion of mangroves is a threshold response to decreased frequency of extreme cold events. Proc Natl Acad Sci. 2014;111: 723–727. pmid:24379379
  17. 17. Ju J, Masek JG. The vegetation greenness trend in Canada and US Alaska from 1984–2012 Landsat data. Remote Sens Environ. 2016;176: 1–16.
  18. 18. Frost GV, Epstein HE. Tall shrub and tree expansion in Siberian tundra ecotones since the 1960s. Glob Change Biol. 2014;20: 1264–1277. pmid:24115456
  19. 19. Elmendorf SC, Henry GHR, Hollister RD, Bjork RG, Boulanger-Lapointe N, Cooper EJ, et al. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nat Clim Change. 2012;2: 453–457.
  20. 20. Hallinger M, Manthey M, Wilmking M. Establishing a missing link: warm summers and winter snow cover promote shrub expansion into alpine tundra in Scandinavia. New Phytol. 2010;186: 890–899. pmid:20345642
  21. 21. Gottfried M, Pauli H, Futschik A, Akhalkatsi M, Barančok P, Alonso JLB, et al. Continent-wide response of mountain vegetation to climate change. Nat Clim Change. 2012;2: 111–115.
  22. 22. Lenoir J, Gégout JC, Marquet PA, de Ruffray P, Brisse H. A Significant Upward Shift in Plant Species Optimum Elevation During the 20th Century. Science. 2008;320: 1768–1771. pmid:18583610
  23. 23. Tao J, Xu T, Dong J, Yu X, Jiang Y, Zhang Y, et al. Elevation-dependent effects of climate change on vegetation greenness in the high mountains of southwest China during 1982–2013. GSTDTAP. 2018;38.
  24. 24. Asner GP, Heidebrecht KB. Spectral unmixing of vegetation, soil and dry carbon cover in arid regions: Comparing multispectral and hyperspectral observations. Int J Remote Sens. 2002;23: 3939–3958.
  25. 25. Roberts D, Smith M, Adams J. Green Vegetation, Nonphotosynthetic Vegetation, and Soils in AVIRIS Data. Remote Sens Environ. 1993;44: 255–269.
  26. 26. Masek JG, Vermote EF, Saleous NE, Wolfe R, Hall FG, Huemmrich KF, et al. A Landsat surface reflectance dataset for North America, 1990–2000. Ieee Geosci Remote Sens Lett. 2006;3: 68–72.
  27. 27. Hoekstra JM, Molnar JL, Jennings M, Revenga C, Spalding MD, Boucher TM, et al. The Atlas of Global Conservation: Changes, Challenges, and Opportunities to Make a Difference. Brekeley, California: University of California Press; 2010.
  28. 28. Boardman JW. Inversion Of Imaging Spectrometry Data Using Singular Value Decomposition. 12th Canadian Symposium on Remote Sensing Geoscience and Remote Sensing Symposium,. 1989. pp. 2069–2072.
  29. 29. Farr TG, Kobrick M. Shuttle radar topography mission produces a wealth of data. Eos Trans Am Geophys Union. 2000;81: 583–585.
  30. 30. Christopher Daly, Michael Halbleib, Smith Joseph I., Gibson Wayne P., Doggett Matthew K., Taylor George H., et al. Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int J Climatol. 2008;28: 2031–2064.
  31. 31. Scotty Strachan, Christopher Daly. Testing the daily PRISM air temperature model on semiarid mountain slopes. J Geophys Res Atmospheres. 2017;122: 5697–5715.
  32. 32. Muggeo VMR. Estimating regression models with unknown break-points. Stat Med. 2003;22: 3055–3071. pmid:12973787
  33. 33. Galewsky J. Rain shadow development during the growth of mountain ranges: An atmospheric dynamics perspective. J Geophys Res-Earth Surf. 2009;114: F01018.
  34. 34. Trouet V, Diaz HF, Wahl ER, Viau AE, Graham R, Graham N, et al. A 1500-year reconstruction of annual mean temperature for temperate North America on decadal-to-multidecadal time scales. Environ Res Lett. 2013;8: 024008.
  35. 35. Freeman BG, Song Y, Feeley KJ, Zhu K. Montane species track rising temperatures better in the tropics than in the temperate zone. Ecol Lett. 2021;24: 1697–1708. pmid:34000078
  36. 36. Marshall L, Perdijk F, Dendoncker N, Kunin W, Roberts S, Biesmeijer JC. Bumblebees moving up: shifts in elevation ranges in the Pyrenees over 115 years. Proc R Soc B Biol Sci. 2020;287: 20202201. pmid:33171083
  37. 37. von Humboldt A, Bonpland A. Essai sur la géographie des plantes: accompagné d’un tableau physique des régions équinoxiales, fondé sur des mesures exécutées, depuis le dixième degré de latitude boréale jusqu’au dixième degré de latitude australe, pendant les années 1799, 1800, 1801, 1802 et 1803. 1805.
  38. 38. Whittaker RH, Niering WA. Vegetation of the Santa Catalina Mountains, Arizona: A Gradient Analysis of the South Slope. Ecology. 1965;46: 429–452.
  39. 39. Verbyla D, Kurkowski TA. NDVI–Climate relationships in high-latitude mountains of Alaska and Yukon Territory. Arct Antarct Alp Res. 2019;51: 397–411.
  40. 40. Seastedt TR, Oldfather MF. Climate Change, Ecosystem Processes and Biological Diversity Responses in High Elevation Communities. Climate. 2021;9: 87.
  41. 41. An S, Zhu X, Shen M, Wang Y, Cao R, Chen X, et al. Mismatch in elevational shifts between satellite observed vegetation greenness and temperature isolines during 2000–2016 on the Tibetan Plateau. Glob Change Biol. 2018. pmid:30156039
  42. 42. Menke JW, Davis C, Beesley P. Rangeland assessment. Sierra Nevada Ecosystem Project, Final Report to Congress, vol III, Assessments and Scientific Basis for Management Options. Davis: University of California, Centers for Water and Wildland Resources; 1996.
  43. 43. Keenan TF, Riley WJ. Greening of the land surface in the world’s cold regions consistent with recent warming. Nat Clim Change. 2018;8: 825–828. pmid:30319714
  44. 44. Walther G-R, Beißner S, Burga CA, White PS. Trends in the upward shift of alpine plants. J Veg Sci. 2005;16: 541–548.
  45. 45. Koide D, Yoshida K, Daehler CC, Mueller-Dombois D, Wildi O. An upward elevation shift of native and non-native vascular plants over 40 years on the island of Hawai’i. J Veg Sci. 2017;28: 939–950.
  46. 46. Elsen PR, Monahan WB, Merenlender AM. Global patterns of protection of elevational gradients in mountain ranges. Proc Natl Acad Sci. 2018; 201720141. pmid:29784825