Impact Statements
This research utilizes over a century of historical and repeat photography on the Colorado Plateau to document long-term vegetation change on an iconic North American dryland landscape. Understanding long-term vegetation change can be used for predicting ecosystem-level responses to change drivers, but maintaining long-term ecological studies can be challenging. By analyzing 338 photo pairs from 1910 to 2016, our study reveals how three dominant plant communities respond to environmental factors over timescales rarely captured in ecological research. Our findings indicate that vegetation communities have distinct recovery patterns following changes in land use, such as domestic grazing cessation, with grasslands and shrubland communities showing increased plant abundance and cover, while pinyon-juniper woodlands exhibit more complex responses. This analysis provides land managers with an understanding of expectations for restoration timeframes and highlights how managing landscapes can be assessed according to their specific community composition and ecological timing. Results suggest that soil and landform characteristics interact with climate to influence plant community resilience, with deeper, more water-retentive alluvial soils supporting increased vegetation growth across all communities. As drought conditions intensify across southwestern drylands, the varied responses of plant communities highlight the complex interactions between plant functional types, climate, land use history and soil and geomorphic setting, providing important context for anticipating ecosystem responses to future climates. The unique aspect of this research is combining repeat photography with contextual land use and landscape data, demonstrating the synergistic value of historic photos and new spatial data for understanding ecological change across large spatial and temporal scales – a methodology that can be applied in drylands worldwide. This approach creates a bridge between historical land management practices and varied land use, and contemporary monitoring and conservation strategies, offering insights for dryland ecosystems that cover 40% of Earth’s terrestrial surface and support over 2 billion people globally.
Introduction
Dryland vegetation change can be challenging to document due to long-lived and slow-growing perennial species (Ruppert et al., Reference Ruppert, Harmoney, Henkin, Snyman, Sternberg, Willms and Linstädter2015). Understanding long-term change of these varied ecosystems to past climate conditions and land use can be used for understanding future trajectories under climate change and evolving land uses (Maestre et al., Reference Maestre, Eldridge, Soliveres, Kéfi, Delgado-Baquerizo, Bowker, García-Palacios, Gaitán, Gallardo, Lázaro and Berdugo2016; Grünzweig et al., Reference Grünzweig, De Boeck, Rey, Santos, Adam, Bahn, Belnap, Deckmyn, Dekker, Flores, Gliksman, Helman, Hultine, Liu, Meron, Michael, Sheffer, Throop, Tzuk and Yakir2022; Brown et al., Reference Brown, Rodriguez Buritica, Goldberg, Reichenbacher, Venable, Webb and Wilder2024). Inextricably linked vegetation, geomorphic and soils attributes (Reynolds et al., Reference Reynolds, Goldstein and Miller2010; Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022) as well as non-linear or threshold responses of drylands to perturbations further complicate our understanding of vegetation change in dryland ecosystems (Schlesinger et al., Reference Schlesinger, Reynolds, Cunningham, Huenneke, Jarrell, Virginia and Whitford1990; Collins et al., Reference Collins, Belnap, Grimm, Rudgers, Dahm, D’Odorico, Litvak, Natvig, Peters, Pockman, Sinsabaugh and Wolf2014). Because of slow response times, particularly in recovering from perturbations (Lovich and Bainbridge, Reference Lovich and Bainbridge1999; Avirmed et al., Reference Avirmed, Lauenroth, Burke and Mobley2015), long-term data or information on dryland vegetation dynamics can inform understanding of the resilience of these globally important ecosystems that cover about 40 percent of the Earth’s terrestrial surface (Safriel et al., Reference Safriel, Adeel, Neimeijer, Puigdefabregas, White, Lal, Winslow, Ziedler, Prince, Archer, King, Shapiro, Wessels, Nielsen, Portnov, Reshef, Thornell, Lachman, McNab, Hassan, Scholes and Ash2005; Ruppert et al., Reference Ruppert, Harmoney, Henkin, Snyman, Sternberg, Willms and Linstädter2015; Maestre et al., Reference Maestre, Eldridge, Soliveres, Kéfi, Delgado-Baquerizo, Bowker, García-Palacios, Gaitán, Gallardo, Lázaro and Berdugo2016).
The response and recovery times to perturbations of drylands can occur on multi-decadal, even centuries-long scales (Reynolds et al., Reference Reynolds, Smith, Lambin, Turner, Mortimore, Batterbury, Downing, Dowlatabadi, Fernández, Herrick, Huber-Sannwald, Jiang, Leemans, Lynam, Maestre, Ayarza and Walker2007). Therefore, a variety of methods have been used to understand dryland vegetation dynamics at time scales that are contextually logical, such as longitudinal field studies, historical records (Sparks et al., Reference Sparks, West and Allen1990; Gibbens et al., Reference Gibbens, McNeely, Havstad, Beck and Nolen2005; Thomas-Van Gundy and Nowacki, Reference Thomas-Van Gundy and Nowacki2013), environmental tracer work (D’Andrea et al., Reference D’Andrea, Anderson, Cole and Zweifel2023) and space-for-time substitutions (Blois et al., Reference Blois, Williams, Fitzpatrick, Jackson and Ferrier2013; reviewed in Maestre et al., Reference Maestre, Eldridge, Soliveres, Kéfi, Delgado-Baquerizo, Bowker, García-Palacios, Gaitán, Gallardo, Lázaro and Berdugo2016). Long-term field studies can be difficult to maintain but provide important on-the-ground measurements for vegetation shifts that lead to contemporary conditions. For example, in the context of climate change, vegetation can be seen shifting across elevational gradients in long-term studies in response to changing soil moisture conditions and temperature (Kelly and Goulden, Reference Kelly and Goulden2008; Abella et al., Reference Abella, Guida, Roberts, Norman and Holland2019). Historical records such as the use of Public Land Survey records and ‘witness trees’ can provide spatially explicit ecological information to understand how vegetation communities and environmental change have impacted landscapes (Sparks et al., Reference Sparks, West and Allen1990; Dyer, Reference Dyer2001; Thomas-Van Gundy and Nowacki, Reference Thomas-Van Gundy and Nowacki2013 and others). Improvements in technology and availability of both aerial and satellite imagery provide the opportunity to monitor more recent vegetation changes on large landscape scales at higher resolutions which can be coupled with long-term sites, historical records, and historical aerial photos (Kull, Reference Kull2005; Foster et al., Reference Foster, Olsen, Dale and Cohen2010). A challenge is the lack of more complete spatial pattern or patch-level information in field studies and historical records as well as detailed vegetation species information from early aerial photography to provide enough details for comparison with more contemporary imagery (e.g. Browning et al., Reference Browning, Archer and Byrne2009).
Another form of record to help understand long-term landscape change is repeat photography (Rogers, Reference Rogers1982; Kull, Reference Kull2005; Turner et al., Reference Turner, Webb, Bowers and Hastings2021), which is particularly useful to understand change in drylands that includes more detailed patch-level information (Bowers et al., Reference Bowers, Webb and Rondeau1995; Zier and Baker, Reference Zier and Baker2006; Hammond et al., Reference Hammond, Stone and Stone2020). Repeat photography utilizes photos taken from a variety of users, locations and time periods; and as people, whether tourists, locals or scientists, revisit these locations, measurements can be taken from the comparison between photos to quantify and/or qualify landscape changes (Rogers, Reference Rogers1982; Webb et al., Reference Webb, Boyer, Turner and Bullock2007a). Repeat photos can be used to track change in individual plants over time (Bowers et al., Reference Bowers, Webb and Rondeau1995) that may not have been captured by on the ground surveys. For instance, repeat photography found that conifers and deciduous trees increased in extent indicating recovery from Euro-American settlement impacts in the San Juan Mountains of Colorado (Zier and Baker, Reference Zier and Baker2006). Other examples include monitoring debris flows and assessing channel change in rivers (Webb et al., Reference Webb, Leake and Turner2007b), understanding sand aeolian deposition (Lancaster, Reference Lancaster2013), rangeland vegetation change (Rogers, Reference Rogers1982; Clark and Hardegree, Reference Clark and Hardegree2005) and assessing how soils and geomorphic setting determine vegetation patch dynamics (Bestelmeyer et al., Reference Bestelmeyer, Ward and Havstad2006).
Here, we use repeat photography to understand long-term ecological dynamics on the Colorado Plateau (Plateau hereafter), a globally important physiographic dryland region in the southwestern United States that has undergone tremendous land use and land management change since westward European expansion in the mid-1800s (McPherson, Reference McPherson1995; Schwinning et al., Reference Schwinning, Belnap, Bowling and Ehleringer2008; Cannon and Embry, Reference Cannon and Embry2009; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017) and more recently has experienced significant warming and droughts (Duniway et al., Reference Duniway, Finger‐Higgens, Geiger, Hoover, Pfennigwerth, Knight, Van Scoyoc, Miller and Belnap2023), much of which is attributed to anthropogenic climate change (Williams et al., Reference Williams, Cook and Smerdon2022). Available long-term studies on the Plateau suggest that the vast amounts of perennial grasslands and other sensitive plant species have decreased due to a mix of land use and climate change (Schwinning et al., Reference Schwinning, Belnap, Bowling and Ehleringer2008). This may be inextricably linked to some more recent expansion of shrub- and woodlands documented in some areas of the Plateau (Miller et al., Reference Miller, Chambers, Evers, Williams, Snyder, Roundy and Pierson2019). Species symptomatic of overgrazing, such as snakeweed (Gutierrezi sarothrae), yellow rabbitbrush (Chrysothamnus viscidiflorus) and invasive species were not found in paleoecological records until after the introduction of domestic grazing animals on the Plateau (Cole et al., Reference Cole, Henderson and Shafer1997). However, there are not many detailed, analytical long-term perspectives (that is with observations predating 1970) for the Plateau (but refer to Webb et al., Reference Webb, Belnap and Weisheit2004).
To address this information gap, we analyze more than a century of change in three dryland plant communities on the Plateau using interpretation of change from repeat photography and available spatial contextual data on climate, land-use, and geomorphology. The study focuses on areas in and adjacent to National Park Service (NPS) managed lands in southeastern Utah and southwestern Colorado, USA and a repeat photo collection that spans 1910 to 2016. We use the assembled change data and covariates to ask the following questions: How are key plant communities on the Colorado Plateau changing across time? And, what environmental and anthropogenic drivers are associated with observed changes in different plant communities across the Colorado Plateau?
Methods
Study area description
Photos were taken across a 3100 km2 area on the Plateau largely in areas administered by the NPS and Bureau of Land Management (BLM), including (but not limited to) areas now within Arches National Park, Canyonlands National Park, Glen Canyon National Recreation Area, Hovenweep National Monument, Canyons of the Ancients National Monument, Natural Bridges National Monument, Grand Staircase-Escalante National Monument and Bears Ears National Monument (Figure 1). The Plateau is known for its wide range of mountains, high plateaus, deep canyons and open landscapes and is characterized as a cold, semiarid to arid dryland with temperature ranging from warm summers (maximum temperatures June–August: 33–55 °C) and cold winters (minimum temperatures December–February: −7 to −3 °C) and mean annual precipitation of 227 mm/year, ranging between 173 and 323 mm/year (Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022). The Plateau has experienced rapid temperature increases and aridification analogous with much of the southwestern United Stations (Schwinning et al., Reference Schwinning, Belnap, Bowling and Ehleringer2008; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017), with temperatures in Canyonlands National Park recorded as being above average 17 of the 23 years from 1996 to 2019 (Figure 1 in Finger-Higgens et al. Reference Finger-Higgens, Duniway, Fick, Geiger, Hoover, Pfennigwerth, Van Scoyoc and Belnap2022).
Map of study area. Pink circles are locations of stake locations where historic and repeat photographs were taken and used in this analysis. U.S. National Park Service (NPS) boundaries are for those NPS administered lands that were part of the study.
Figure 1. Long description
A topographical map displays study sites across the Southwestern United States.
* Inset Map: A map of the continental United States in the top left corner shows a cluster of pink dots in the Four Corners region, where Utah, Colorado, Arizona, and New Mexico meet.
* Main Map Legend: Pink circles represent Photo Locations. Black outlines represent N P S Boundaries.
* Northern Region: Arches National Park contains a dense cluster of pink dots in its southern section. To the west and south of Arches is the Bureau of Land Management territory.
* Central Region: Canyonlands National Park shows a high concentration of pink dots, particularly in its northern and eastern sections. To the southwest of Canyonlands is Natural Bridges National Monument, which contains a small cluster of dots.
* Western Region: Grand Staircase Escalante National Monument (B L M) and Glen Canyon National Recreation Area form a large contiguous block on the left side of the map. Photo locations are concentrated in the southern portion of Glen Canyon near the Arizona border.
* Southern and Eastern Regions: The Navajo Nation is labeled at the bottom. To the east, Hovenweep National Monument is highlighted with an oval containing four pink dots.
* State Borders: Dashed lines mark the intersections of U T, C O, A Z, and N M.
* Scale: A scale bar at the top left indicates a range from 0 to 50 miles.
Dominant vegetation types of the area include species that utilize C3, C4 and crassulacean acid metabolism photosynthetic pathways and with a broad range of herbaceous to woody vegetation. Biological soil crusts are common on the Plateau and provide important ecosystem functions, including soil stability, biogeochemical cycling and hydrologic processes (Weber et al., Reference Weber, Büdel and Belnap2016). The Plateau has varied land-use history, with the most dominant activity since the onset of this study being domestic livestock grazing, mineral and energy extraction (e.g. oil and gas) and, more recently, tourism (Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). Grazing has had a staggered history of exclusion from within National Park units. Livestock were removed from Arches and Canyonlands National Parks between mid-1970s to early 1980s, respectively, (Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022) but grazing is still an active land use in the BLM units of this study.
Photo collection
We use available photo collections curated by the U.S. Geological Survey and the Southeastern Utah Group of National Parks (NPS, Moab, UT) as part of the Desert Laboratory Collection of Repeat Photography (Webb et al., Reference Webb, Boyer, Turner and Bullock2007a). The earliest photo capture was in 1910, with the first photos for each location taken in the years 1910 through 2012 and designated by a stake location identifier (‘stake’ hereafter; Figure 2). Historical photos were gathered from various collections of government agencies, universities, museums and library collections where photos have a known location, either due to GPS information or unique features and descriptions (Webb et al., Reference Webb, Boyer, Turner and Bullock2007a; Boyer et al., Reference Boyer, Webb and Turner2010). Some locations had historical repeat photos (two or more), while most stake locations only had a single historical photo. Though the intended subject matter of the original photos is unknown, it was not assumed that the photographer was purposefully targeting vegetation and were taken by a range of observers, including geologists, tourists, tour guides, NPS staff and professional photographers in southeastern Utah. Exceptions include some relatively recent photos taken by NPS with the intent to monitor road recovery or riparian areas.
Normalized air temperature (temp) (a), normalized annual precipitation (precip) (b) and photo count by first year of each repeat-photo series for study period (c). (a) Red indicates normalized temperatures above the mean (warmer), blue is below the mean (colder), (b) blue indicates normalized precipitation amounts above the mean (wetter) and red is below the mean (drier), (c) blue bounded grey box is the time frame when the last repeat photo was taken for all photo stake locations. Some first photos are within that time frame and would also have had a repeat photo taken. Climate data from PRISM (PRISM Climate Group, Oregon State University, http://prism.oregonstate.edu), accessed October 26, 2022. Mean values presented based study period (1910–2016).
Figure 2. Long description
A multi-panel figure with three vertically stacked bar charts sharing a common x-axis representing the Year of first photo from 1910 to 2016.
* Panel a: Normalized Mean Air Temp. The y-axis ranges from negative 2 to 2. Blue bars below the zero-line indicate colder than average years, dominant from 1910 to 1930. Red bars above the zero-line indicate warmer than average years, becoming increasingly frequent and sustained from the mid-1990s through 2016, with several peaks reaching or exceeding 2.
* Panel b: Normalized Annual Precip. The y-axis ranges from negative 2 to 2. Blue bars above the zero-line indicate wetter years, and red bars below indicate drier years. The data shows high inter-annual variability. Notable wet peaks occur around 1927, 1941, 1957, 1968, and 2015. Significant dry periods are visible in the early 1930s, early 1950s, and late 1980s.
* Panel c: Photo count. The y-axis ranges from 0 to 100. Dark grey bars show the frequency of first photos taken each year. Most years have counts below 20, with a major outlier peak exceeding 100 photos around 1965. A light grey box bounded by blue vertical lines highlights the period from approximately 2002 to 2015, representing the time frame when the last repeat photos were taken.
A contemporary repeat photo was obtained at each stake between 2002 and 2016, with efforts made to revisit as many stakes as possible in NPS managed lands and stakes outside of the parks as opportunities arose. During the repeat photo campaign, teams navigated to the stake GPS coordinates and then photographers used the original photo(s) to fine-tune the location and camera orientation and angle. Several stakes have photos taken in more than one orientation capturing different parts of the landscape from the same location. For these, the different orientations were repeated as well and cataloged as sub- stakes (same GPS location information), denoted with a unique identifier (that is s6036a, s6036b, s6036c) and interpreted independently. For more technical information on collecting and matching historical photographs, refer to Boyer et al. (Reference Boyer, Webb and Turner2010). Photos can be found in Duniway et al. (Reference Duniway, Belnap, Webb, Miller, Van Scoyoc and Bishop2026).
Photo interpretation
To broadly capture information on vegetation types occurring in each photo and general patterns of change, we did a class-based vegetation type and change classification by photo interpretation. For each set of photographs taken at each stake location, vegetation patches were visually classified using the SWReGAP classifications (Lowry et al., Reference Lowry, Ramsey, Thomas, Schrupp, Sajwaj, Kirby, Waller, Schrader, Falzarano, Langs, Manis, Wallace, Schulz, Comer, Pohs, Rieth, Velasquez, Wolk, Kepner, Boykin, O’Brien, Bradford, Thompson and Prior-Magee2007; Table A1). The three main vegetation classes used to identify vegetation patches in this analysis are (1) Colorado Plateau Blackbrush-Mormon-tea Shrubland (CP); (2) Inter-Mountain Basins Semi-Desert Grassland and Big Sagebrush Shrubland (IM) and (3) Colorado Plateau Pinyon-Juniper Woodland/Shrubland (PJ), composed of Pinus edulis and/or Juniperus spp. (likely just Juniperus osteosperma) dominating the tree canopy (USGS National Gap Analysis Program, 2005; Table A1). For each vegetation patch that was classified in the first (historical) photo, interpretations of change were made for the same classified vegetation patch in the repeat photo (Figure 3; Table 1). All photos from a stake in a series were included unless the resolution of the photo was too poor to distinguish individuals, there was no vegetation present in the photo, or the vegetation in the photos was predominantly riparian. To quantify change, we interpreted each patch type in every repeat photo pair (one interpreter for each stake, calibrated across two main interpreters) for the overall net change between the historical photo (taken between 1910–2012) and the final contemporary photo (taken between 2002–2016), repeating the net change for each photo at stakes with more than one historic photo. We used this net change approach, instead of incremental change because we did not have enough stakes with multiple repeat photo pairs to integrate incremental change into the analysis. This approach allows us to interpret change on a broad scale of compounding influences over many years. Change metrics (vegetation response variables) for interpretation on each patch type for each repeat photo pair included: Individual Counts– increase/decrease/stable, Individual Size – larger/smaller/stable, Patch shift – yes/no, Patch size – increase/decrease/stable, Live Plant Cover – increase/decrease/stable (Table 1; Bishop et al., Reference Bishop, Kelly, Cheng, Burch, Mikenas, Knight and Duniway2026). In each interpreted photo, most patches should have had discernable boundaries, and the interpreter should have been able to visually identify the exhibited species characteristic of the SWReGAP class including grasses, shrubs, and trees. If the patch in the image was too far away or out of focus (blurry) to distinguish characteristic species or change metrics, then it was not included in the analysis. All interpretations are considered within SWReGAP classes for a photo pair (that is all individual counts in the CP Blackbrush vegetation class) or with respect to the class boundary or area as per the delineated imaginary polygon (refer to Table 1 for descriptions of each change metric, notes on how metrics differ from one another and the specific questions that were answered by the interpreter for each change class; Webb et al., Reference Webb, Boyer, Turner and Bullock2007a).
Conceptual graphic to illustrate metrics used for repeat photo interpretation analysis with covariates. Interpretations of change were made for the same classified vegetation patch in the repeat photo for individual counts, individual size, patch size, patch shift and live plant cover (described in Table A2). Graphics indicate placement, size, and number within each interpreted patch. Plant artwork credit to Michael A. Bishop.
Figure 3. Long description
The graphic displays five rows of comparative plant illustrations.
* Row 1: Individual Counts. Increase shows one tree and two shrubs in the Historical frame becoming two trees and three shrubs in the Repeat frame. Decrease shows one tree and two shrubs becoming only two shrubs.
* Row 2: Individual Size. Increase shows a small tree and two shrubs becoming a significantly larger tree and larger shrubs. Decrease shows the tree and shrubs becoming much smaller in the Repeat frame.
* Row 3: Patch Size Change. Increase shows a small irregular purple boundary containing three plants expanding into a much larger black boundary with more plants. Decrease shows the boundary shrinking and containing fewer plants.
* Row 4: Patch Shift. YES shows the Repeat boundary moved to the right of the original Historical position. NO shows the Repeat boundary remaining in the same spatial position as the Historical boundary.
* Row 5: Live Plant Cover. Increase shows four sparse plants in the Historical frame becoming five denser, larger plants in the Repeat frame. Decrease shows the plants becoming smaller, sparser, and lighter in color.
Photo interpretation metrics of change
Table 1. Long description
The table consists of two columns: Metric of change and Descriptions and questions addressed.
* Patch size: Evaluates if the area occupied by a vegetation class within a virtual bounding polygon has changed. It asks if the area expanded, remained the same, or contracted.
* Patch shift: Interprets if the edges of vegetation classes shifted due to range expansion, contraction, or disturbance. It asks if the boundary changed (Yes/No).
* Individual plant counts: Interprets the net change in the number of individuals. It asks if the number increased, remained the same, or decreased (+/-/0).
* Individual plant size: Interprets change in the size of individual plants, assessing growth or dieback. It asks if individuals are larger, smaller, or the same (+/-/0).
* Plant cover: Measures change in live cover within the patch due to growth, recruitment, dieback, or mortality. It asks if the density of live vegetation cover increased, decreased, or stayed the same (+/-/0), and if the leaf area became more lush or sparse.
Note: A distinction to note between patch shift and patch size is that patch shift refers to the area covered by the delineated imaginary polygon and not just the outer boundary and whether that changed. patch size is also not an interpretation of live cover, density, or sizes of individuals. because photos often capture multiple patches of the same class, we recorded estimated net change in the photo.
Statistical analysis
To address our objective of understanding what factors may explain vegetation and patch patterns of changes observed, we used information from the photos as well as overlay of stake location with spatial data. Explanatory information we used from the photos themselves was the SWReGAP class membership (veg.class, or hereafter vegetation class in text), the change of invasives (change.invasives) and human disturbance (human.disturbance; e.g. road or foot traffic, buildings, etc.) from photo interpretation, and number of years between photos (span; Table A2). Spatial data we extracted were a suite of geomorphic (landform), topographic (topographic wetness index [TWI], aspect, slope, and curvature), climate (mean annual precipitation and mean annual temperature averaged across the time span in between the historic and most recent photo). Number of years since grazed by domestic livestock [yrs.grazed] was extracted from spatial maps indicating the year domestic livestock were removed or if it had remained open to grazing (Table A3). To summarize spatial data for each photo field, we used the Viewshed Analysis tool in ArcGIS Pro for each stake (ESRI ArcGIS Pro; Spatial Analyst Toolbox). This included recorded information from each recent final photo such as the camera diagonal lens length, focal length and the height of the camera during photo capture. Within each viewshed, mean values were extracted for all topographic and climate variables.
We modeled vegetation change responses using the following formula: Response ~ vegetation class + landform + change in invasives + human disturbance + change in disturbance + TWI + aspect + slope + curvature + span (years between photos) + years since grazed + mean annual precipitation + mean annual temperature + two-way interactions between vegetation class and: human disturbance, change in disturbance, TWI, aspect, slope, curvature, span, years since grazed, mean annual precipitation and mean annual temperature. We assessed pairwise correlations among continuous predictor variables to identify and avoid multicollinearity, selecting mean annual temperature and mean annual precipitation as representative metrics from highly correlated seasonal and annual climate variables. No pairwise correlations among numeric predictors exceeded 0.60 in absolute value. We modeled binary or yes/no responses using logistic regression or more specifically, a generalized linear mixed model with a binomial response distribution. For ordered responses with three levels (smaller/fewer/decrease, stable, larger/more/increase), we used a cumulative link or ordered logistic regression model (Agresti, Reference Agresti2002; McCullagh and Nelder, Reference McCullagh and Nelder2019). All models included a random effect for park group that accounts for within-group spatial correlation. For the ordered logit models, we assumed proportional odds such that the only difference among the ordered response levels is modeled with a different latent intercept for each response level but common estimated slopes and treatment levels across predictors. We confirmed the proportional odds assumption using the graphical method described in Harrell (Reference Harrell2015) that plots binary logit predictions for each response level versus the remaining levels for each predictor. We observed no departure from parallelism that would warrant a more complex multinomial approach with separate parameters estimates for each response level. For responses that may marginally violate the proportional odds assumption, we note that the estimates still represent averages across the response levels. Because we modeled each response, we can have significant results for either direction (increase/decrease/stay the same or yes/no). All modeling was performed using the R environment (R Development Core Team, 2024). Binomial GLM’s were estimated using function glmmTMB in R package glmmTMB from the base R program (McGillycuddy et al., Reference McGillycuddy, Warton, Popovic and Bolker2025). Ordered logit models were fit using function clmm in R package ordinal (Christensen, Reference Christensen2023). Effect estimates and comparisons were generated using package emmeans (Lenth, Reference Lenth2017) and plots were generated using package ggeffects (Lüdecke, Reference Lüdecke2018). Models were assessed based on Accuracy and Cohen’s Kappa (Cohen, Reference Cohen1960). Fit metrics were computed using package caret (Kuhn, Reference Kuhn2008), function confusionMatrix. Pairwise comparisons were adjusted using Tukey’s honest significant difference method (Tukey, Reference Tukey1949).
Results
Overall, photo interpretations were made on 296 stakes with a range of 1 to 3 repeat photos (2–4 total photos) per stake, for a total of 338 photo pairs analyzed (some photo pairs were a different cardinal direction at the same stake location), with 102 years of change documented (Figure 4). The mean time span is 50 years with a range varying from 1 to 102 years between photo pairs. Of the 338 photo pairs, 111 had only one vegetation class, 162 photo pairs had two vegetation classes and 65 photo pairs had all three vegetation classes. Photo interpretation and associated spatial data are available in Bishop et al. (Reference Bishop, Kelly, Cheng, Burch, Mikenas, Knight and Duniway2026).
Historic repeat-photo examples. Top: stake 6659, Horse Canyon, Canyonlands National Park, 1965 [left, archive collection Brigham Young University, Provo, UT] and 2015 [right, photographer C. Shelz, National Park Service]. Middle: stake 6036b, Happy Canyon, Canyonlands National Park, 1931 [left, photographer A. Baker] and 2012 [right, photographer M. Gertler]. Bottom: stake 6663, Salt Creek, Canyonlands National Park, 1965 [left, archive collection Brigham Young University, Provo, UT] and 2015 [right, photographer US Geological Survey].
Figure 4. Long description
A six-panel collection of photos arranged in three rows of two.
* Top row (stake s 6 6 5 9): The 1965 photo on the left shows a tall sandstone pillar and a large alcove with two vehicles and people in the foreground. The 2015 photo on the right shows the same rock formations with significantly denser green shrubbery and trees at the base, and the vehicles are gone.
* Middle row (stake s 6 0 3 6 b): The 1931 black and white photo on the left shows a wide canyon vista with scattered desert scrub. The 2012 color photo on the right shows the same vista with a dramatic increase in the density and size of dark green pinyon-juniper woodland across the foreground and middle ground.
* Bottom row (stake s 6 6 6 3): The 1965 photo on the left shows a sandstone cliff with a small natural arch (pothole) near the top and sparse vegetation in the sandy foreground. The 2015 photo on the right shows the same cliff face, but the foreground is now heavily populated with tall grasses and thick shrubs, obscuring much of the sandy soil seen previously.
Due to the large number of results and an effort to keep the text concise, we present in the results model significance of α = 0.05, but present model results of significance α = 0.1 in Tables 2 and 3 and discussion due to the nature of the study with coarse resolution data. Model fits for the five change response variables (Individual Counts, Individual Size, Patch Shift, Patch Size, Live Plant Cover; Table 2) were fair to substantial agreement with accuracy and kappa statistics ranging from 0.59–0.72 and 0.27–0.44, respectively (Table 2). Span, or the time from the initial photo to the last photo, was the only temporal covariate that was considered statistically significant across all five-vegetation response variable models. We found northness was significant across four of the five response variables as main effects and one interaction, mean annual temperature by vegetation class (Table 2). Variables that were not significant in any models were topographic variables (slope, TWI, curvature) and their interactions with vegetation class, and mean temperature as a singular main effect (model accuracy and test statics in Table 2 and 3).
Statistical results of metrics of change from generalized linear models (GLM), including degrees of freedom (Df), likelihood ratio test chi-square (LR Chisq) and significance of chi-square test (Pr > Chisq)
Table 2. Long description
A complex table presenting statistical results of metrics of change from generalized linear models.
Header Structure:
The table is organized into 12 columns. The first column lists Covariates and Spatiotemporal covariates. The second column is D r (degrees of freedom). The remaining 10 columns are grouped under Vegetation response, with two columns (Chisq and Prob.) for each of the following five metrics: Individual count, Individual size, Patch shift, Patch size, and Live plant cover.
Data Rows:
- Vegetation class: D r 2. Significant for Individual size (Chisq 11.533, Prob. 0.003).
- Time span: D r 1. Significant across all five metrics (Prob. values ranging from 0.003 to 0.042).
- Years since last grazed: D r 1. Significant for Individual count, Patch size, and Live plant cover.
- Change in invasives: D r 1. Significant for Individual size (Prob. 0.037).
- Presence of human disturbance: D r 1. Significant for Patch shift (Prob. 0.008) and Patch size (Prob. 0.039).
- Climate (M A P): D r 1. Significant for Individual count (Prob. 0.009).
- Geomorphology (landform): D r 3. Significant for Live plant cover (Prob. 0.030).
- Geomorphology (northness): D r 1. Significant for Individual size, Patch size, and Live plant cover.
- Vegetation class interactions: Significant interactions are noted for veg.class:yrs.grazed (Individual count, Patch size, Live plant cover), veg.class:change.disturbance (Individual count, Patch size), and veg.class:M A T (Individual count, Individual size, Live plant cover).
Overall Statistics:
- Accuracy: Ranges from 0.59 to 0.7219 across the five metrics.
- Kappa: Ranges from 0.2735 to 0.4423.
- P-Value [ACC > NIR]: All metrics show high significance, with values as low as 3.338 E minus 16.
Note: Bottom rows show model accuracy and performance results including kappa values (CI, confidence interval). dark orange shaded cells are spatiotemporal explanatory variables that are significant and shared with 4 or more vegetation response variables; medium orange are spatiotemporal explanatory variables that are significant and shared with 2–3 vegetation response variables and light orange have at least 1 significant vegetation response variable. dark green shading with any metric with p < 0.05, light green is p < 0.1. Significance level indicated by asterisk (*p < 0.05; **p < 0.01; ***p < 0.001). TWI, topographic wetness index.
This summary is to provide a graphical representation of the probability of directional change of the response change metrics based on the ‘marginal mean’ or the average response of a predictor while holding the remaining predictors in the model constant, i.e, at their mean reference level
Table 3. Long description
The table is organized into three main regional sections: C P Blackbrush, I M Sagebrush, and P J Woodland. The top header row lists response variables: Individual counts, Species richness, and Shannon diversity. The first column lists explanatory variables including Time since last grazed, Elevation, Slope, Aspect, Temperature, and Precipitation.
Directional symbols and color coding indicate trends:
* Upward arrows indicate an increase, downward arrows indicate a decrease, and horizontal dashes indicate stability.
* Bold arrows signify a high probability greater than 0.60.
* Green shading indicates a pure increase, red indicates a pure decrease, and yellow indicates stability.
* Light green indicates stable and increase, light pink indicates stable and decrease, and light blue indicates a mix of stable, increase, and decrease.
* Grey cells indicate non-significant variables with p-values greater than 0.05.
Key data trends:
* In C P Blackbrush, Individual counts show a bold upward arrow for Time since last grazed and a bold downward arrow for Temperature.
* In I M Sagebrush, Species richness shows a bold upward arrow for Elevation and a bold downward arrow for Temperature.
* In P J Woodland, Shannon diversity shows a bold upward arrow for Precipitation and a bold downward arrow for Temperature.
Note: The arrows represent whether the probability was higher and significant for the response variable (top row) to increase or get larger (↑), ‘no change’ or ‘stable’ (−), decrease or get smaller (↓). Color shading represents whether the change was all in one direction (e.g. only increased = green shading, only decreased = red shading, ‘stable’ = yellow), or shared directions (e.g.‘stable’ AND increase = light green shading, ‘stable AND decrease = light pink shading, ‘stable’ AND increase AND decrease = light blue). The explanatory variables in the first column are presented as descriptions of either continuous or categorical. For continuous variables such as temperature (temp) and precipitation (precip) these represent that the probabilities were higher for generally higher or lower temperature, but there are no defined cutoffs. For example, in CP Blackbrush, individual counts had a higher probability of increasing in number of individuals at stakes with a longer time since last grazed. Bold font for arrows (↑ or ↓) indicates highest probability of greater than 0.60; Grey cell shading indicates explanatory variables were not significant in the model (p ≤0.05). CP, Colorado Plateau; IM, Inter-Mountain; PJ, Pinyon-Juniper.
For each metric of change, the calculated probabilities of direction (increase, decrease, stable/no change) had varied responses where confidence intervals may overlap for more than one direction. Here, we report the probabilities of significant explanatory variables that were more singular in direction, where the mixed direction probabilities (probabilities were high for more than one direction) are only reported in Table 3.
Individual count
The number of individuals showed variation across explanatory variables, some within each vegetation class (accuracy 0.59, kappa 0.356; refer to Table 2 for model results). Across vegetation classes, we found individuals tended to have stable-increased (meaning stable and increased number of individuals had probabilities with overlapping confidence intervals) numbers of individuals with longer time span (older initial photos) with fairly stable numbers of individuals with newer initial photos (Tables 2 and 3).
In the CP Blackbrush community, greater probability for an increased number of individuals occurred in areas that had a longer time since grazed, recovered or the same disturbance and increasing temperatures (Table 3). There was greater probability of decreased number of individuals with more mean annual precipitation (Table 3).
In PJ communities, we calculated greater probability of increased number of individuals with more recent grazing and decreased number of individuals with more time since last grazed (Table 3). Whereas, in the IM grasslands, we calculated higher probabilities of decreasing individual counts in areas that had been more recently grazed and increased number of individuals with longer time since grazing, with mixed probabilities of overlapping confidence intervals for other covariates (Table 3).
Individual size
The ordered logit model for individual size showed variation across vegetation classes for how individual plant sizes have changed across time (accuracy 0.5957, kappa 0.2735; refer to Table 2 for full model accuracy results). Overall, probability was above 0.6 for individual sizes increasing across vegetation classes in alluvial and valley landforms and where more invasives were present as compared to when no change in or fewer invasives were seen (Tables 2 and 3). Individuals in the CP Blackbrush community increased in size, across all significant explanatory variables, whereas the PJ and IM Grassland communities had more varied responses.
Individuals in PJ communities had higher probabilities to increase in size in areas with lower mean annual temperature (Tables 2 and 3) with mixed probabilities of overlapping confidence intervals of stable increase for other significant covariates (Table 3).
The probabilities for individual sizes to change in the IM Grassland communities with respect to explanatory variables was less clear (Tables 2 and 3). Results suggest slight association of individual size with temperature, with greater probability of stable increase with higher mean temperatures and stable decrease with lower mean temperatures. We found stable sizes with regards to more northern and southern aspects.
Patch shift
Span, time since grazed, human disturbance and change in disturbance, precipitation and northness resulted in a statistically significant (p ≤ 0.1) probability for a patch shift, with accuracy of 0.7219 and kappa of 0.4423 (Table 2). We found longer spans of time associated with greater probability of patch shift but probability for either a patch shift or no patch shift was relatively stable the shorter the photo time span. The presence of human disturbance in the repeat photograph, along with recovery or same disturbance and more precipitation indicated a higher probability for a patch shift in the CP Blackbrush and IM Grassland communities (Table 3). There was higher probability of no patch shift for PJ communities with more precipitation (Table 3).
Patch size
The change in patch size model fit had an overall accuracy of 0.6272 and kappa of 0.3586 (Table 2). Probability for the patch size to be stable increasing was greater with longer time span, while shorter spans remained stable in patch size.
In CP blackbrush communities, increased patch size had higher probability with more time since last grazed, recovery or same disturbance, and more southerly aspects and decreased patch size with less time since grazed (Table 3).
The PJ community patch size had higher probabilities with lower temperatures and shorter time since last grazed, and relatively stable across explanatory factor significant interactions (Table 3). Whereas, IM Grasslands showed higher probabilities of greater patch sizes with more time since grazed and smaller patch sizes with shorter time since grazed, similar to CP Blackbrush (Table 3).
Live plant cover
The change in live plant cover model fit had accuracy of 0.6288 and kappa of 0.3318 and was affected by main effects of landform, and northerly aspects and interactions between vegetation class and time span, years since last grazed and mean annual temperature (Table 2 and 3). There was higher probability of having more dense live plant cover in more southerly aspects, alluvial, plateau benches and valley landforms but neutral responses for dunes (Table 3).
CP Blackbrush vegetation patches had a higher probability of more dense live plant cover with longer time since grazed, longer time span and increasing temperatures (Table 3). In contrast with the CP Blackbrush community, the PJ community had higher probabilities of less dense live plant cover with longer time since grazed and higher mean annual temperatures (Table 3). More dense live cover for PJ vegetation patches were more likely in areas with lower mean annual temperatures and longer time span since the initial photos.
Within the IM Grassland community covariates all had overlapping confidence intervals with all directional change indicating more variability and less certainty for directions of change within this vegetation community (Table 3).
Discussion
Perennial vegetation in drylands typically is slow to respond to drivers of change, with woody species in particular responding to disturbance and climate over decades (Brown et al., Reference Brown, Rodriguez Buritica, Goldberg, Reichenbacher, Venable, Webb and Wilder2024; others). We document vegetation change from 1910 to 2016 on the Colorado Plateau, capturing major shifts in land management and climate (Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017) and discuss here the possible mechanisms for the response by vegetation to those major drivers of change and how landscape factors may mediate change.
Time and plant community changes
An impetus of this and other repeat photograph studies is to understand how vegetation communities are changing with time (e.g. Bowers et al., Reference Bowers, Webb and Rondeau1995). Among the community metrics of change we analyzed here, we found that while timespan was significant for all but change in individual sizes, there were few interactions with vegetation communities. Previous work suggests that woody species infilling or expanding following change in disturbance (e.g. grazing pressure, wildfire, increased invasion, harvesting; reviewed in Miller et al., Reference Miller, Chambers, Evers, Williams, Snyder, Roundy and Pierson2019) or natural demographic and population growth processes (Shriver et al., Reference Shriver, Pletcher, Biondi, Urza and Weisberg2025). However, on the Plateau, pinyon pines and more recently juniper, have shown vulnerability to drought (Allen et al., Reference Allen, Macalady, Chenchouni, Bachelet, McDowell, Vennetier, Kitzberger, Rigling, Breshears, Hogg, Gonzalez, Fensham, Zhang, Castro, Demidova, Lim, Allard, Running, Semerci and Cobb2010; Kannenberg et al., Reference Kannenberg, Driscoll, Szejner, Anderegg and Ehleringer2021b; Redmond et al., Reference Redmond, Urza and Weisberg2023) and perennial grass species, but not shrubs, have been shown to be highly sensitive to extreme drought (Winkler et al., Reference Winkler, Belnap, Hoover, Reed and Duniway2019; Hoover et al., Reference Hoover, Pfennigwerth and Duniway2021; Kannenberg et al., Reference Kannenberg, Driscoll, Malesky and Anderegg2021a; Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022). Our analysis of time may be further complicated due to the final photographs being captured after the initiation of the 2000s ongoing southwestern megadrought (Williams et al., Reference Williams, Cook and Smerdon2022) and before the 2019 juniper dieback (Kannenberg et al., Reference Kannenberg, Driscoll, Szejner, Anderegg and Ehleringer2021b). The greater live plant cover with older initial photos in blackbrush and PJ communities and not in grasslands support this previous research. However, we attribute the lack of significant interactions of community with time for the other metrics to the relatively coarse nature of the photo interpretation variables and complex nature of Plateau vegetation temporal dynamics seen at longer time scales.
Environmental and anthropogenic drivers of community change
Water availability
Dryland ecosystems are primarily limited by water availability (Noy-Meir, Reference Noy-Meir1973), with expectations of greater water inputs from precipitation or decreased evapotranspiration demands from cooler temperatures typically leading to persistence, growth and/or expansion, particularly for woody species (Schlaepfer et al., Reference Schlaepfer, Bradford, Lauenroth, Munson, Tietjen, Hall, Wilson, Duniway, Jia, Pyke, Lkhagva and Jamiyansharav2017; Tietjen et al., Reference Tietjen, Schlaepfer, Bradford, Lauenroth, Hall, Duniway, Hochstrasser, Jia, Munson, Pyke and Wilson2017; Winkler et al., Reference Winkler, Belnap, Hoover, Reed and Duniway2019). The results from the repeat photographs analyzed here support the importance of precipitation and temperature as drivers of Plateau vegetation patch dynamics (Duniway et al., Reference Duniway, Finger‐Higgens, Geiger, Hoover, Pfennigwerth, Knight, Van Scoyoc, Miller and Belnap2023), though the directional impacts on the vegetation community metrics evaluated were less clear. The study period captured important climatic events including the influential 1950s regional drought (Herbel et al., Reference Herbel, Ares and Wright1972; Browning et al., Reference Browning, Duniway, Laliberte and Rango2012) and the turn-of-the-century southwestern megadrought (Williams et al., Reference Williams, Cook and Smerdon2022). Many of our initial photos were taken after the 1950s drought by NPS and other government officials seeking to establish Canyonlands National Park (Quintano, Reference Quintano2014) and, thus, many of the patches we document here were likely recovering from that previous drought period. The slow recovery of plant communities to the 1950s drought may also have caused some conflicting results in precipitation and temperature results. For example, reductions in grassland extent in the Chihuahuan Desert due to 1950s drought persisted even after the drought conditions abated (Browning et al., Reference Browning, Duniway, Laliberte and Rango2012). Additionally, several studies have identified the importance of seasonal drought or precipitation for drylands (Hoover and Rogers, Reference Hoover and Rogers2016; Hao et al., Reference Hao, Zhang, Biederman, Li, Cui, Xue, Du and Wang2018; Esch et al., Reference Esch, Lipson and Cleland2019; Currier and Sala, Reference Currier and Sala2022), including Plateau grasslands and shrublands (Munson et al., Reference Munson, Long, Wallace and Webb2016; Hoover et al., Reference Hoover, Pfennigwerth and Duniway2021). In drylands of the southwestern United States, increased winter precipitation especially has been linked to growth and expansion of deep-rooted woody species (Browning et al., Reference Browning, Duniway, Laliberte and Rango2012; Munson et al., Reference Munson, Long, Wallace and Webb2016). An opportunity therefore presents itself to analyze historical photo interpretations with higher resolution than the annual climate periods analyzed here. Seasonal precipitation and temperature data could provide a more nuanced understanding of long-term Plateau community assembly with changing seasonal dynamics projected with climate change (IPCC, 2023).
Landform and topographic settings
Soils and geomorphology provide important controls on dryland ecosystem dynamics on the Plateau and drylands globally (Munson et al., Reference Munson, Long, Wallace and Webb2016; Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022). Our results support these past studies, with the significant association detected between long-term vegetation change and geomorphology (aspect and landform). For example, all communities in our repeat photographs showed increased individual size and live cover in alluvial and valley landforms, more so than other landforms. Alluvial and valley soils in this region are typically deep with higher available water holding capacity (Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022), which may explain why we see evidence of increased productivity for communities in these landforms (individual sizes and live cover). With aspect, we found that CP Blackbrush communities appear to grow better in more southerly aspects (increase patch size and greater evidence for increase individual size). This is potentially due to the drought tolerant nature of blackbrush allowing it to benefit from these aspects providing an increased number of warm growing days in the spring and fall (Munson et al., Reference Munson, Long, Wallace and Webb2016). The lack of clear grassland and PJ community associations with aspect (no clear directional association) somewhat contradicts experimental work and observational work suggesting Plateau grasslands and PJ are vulnerable to warming and drying conditions (Wertin et al., Reference Wertin, Reed and Belnap2015; Winkler et al., Reference Winkler, Belnap, Hoover, Reed and Duniway2019; Kannenberg et al., Reference Kannenberg, Driscoll, Malesky and Anderegg2021a) that should be magnified on southern aspects and some of which are observable during the study’s final photos (Figure 2c). This contradictory response of IM grasslands, and lack of clear metric responses of other vegetation classes may be due in part to a mismatch in scale between the 30-m resolution elevation dataset used here and the complex topography of the study area. However, we reason that the established communities in our repeat photos were determined initially by the soil geomorphic setting (Comstock and Ehleringer, Reference Comstock and Ehleringer1992; Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022) and change drivers such as climate and disturbance may affect the directional changes of the community moving forward, with complex interactions between drivers of change and soil geomorphic setting.
Livestock removal
Livestock management evolved from minimal federal oversight to implementation of the Taylor Grazing Act (1934), establishment of National Monuments and Parks including livestock removal (1964–2000; McPherson, Reference McPherson1995) and increased federal grazing oversight under the Federal Land Policy and Management Act (1976) and related legislation (McGinty et al., Reference McGinty, Baldwin and Banner2009). These changes, along with broader economic forces, shifted livestock composition from sheep and cattle to primarily cattle while also decreasing overall numbers of both cattle and sheep on the landscape (Thomas, Reference Thomas1973; McPherson, Reference McPherson1995; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). Our analysis of repeat photography focused primarily on the presence or absence of livestock and photographs of before and after livestock were removed from pastures within the national monuments and parks. Our analysis suggests that more time since livestock removal, regardless of livestock type, led to observable changes in the plant community. We observed recovery of plants in the CP Blackbrush and IM grassland communities, driven by increasing numbers of individuals, increasing patch size, and increasing cover within patches, while our observations suggest the PJ shrubland/woodland communities have some evidence of fewer individuals, smaller patch sizes and cover as the area recovers from livestock grazing (Table 3). Individual counts and increasing patch size in IM grassland communities indicate that time since grazed may provide the opportunity for perennial grasses (which are common forage species) to establish and infill with the release of grazing pressure, which has been observed in other studies conducted in the project area (Miller, Reference Miller2008; Miller et al., Reference Miller, Belote, Bowker and Garman2011). Longer time since grazed also led to increased establishment and infill in CP blackbrush communities. While blackbrush is not typically desired browse for cattle, cattle will eat new shoots and the herds of sheep common in the early study period (e.g. < 1965) would have likely utilized blackbrush communities (West, Reference West and Goodall1983; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). There were modest patch size changes in PJ shrublands/woodlands and an overall increase in the number of individuals with more recent or ongoing grazing pressure and fewer individuals with more time with livestock removed. Various studies link grazing to increased woody cover with the hypothesized mechanism being selective herbivory decreasing competition and increased niche space for unpalatable woody species (Roques et al., Reference Roques, O’Connor and Watkinson2001; Valone et al., Reference Valone, Meyer, Brown and Chew2002; Gibbens et al., Reference Gibbens, McNeely, Havstad, Beck and Nolen2005), reduction in grasses and browse (Harris et al. 2003; Caracciolo et al., Reference Caracciolo, Istanbulluoglu, Noto and Collins2016) and subsequent reduction of fine fuels and associated effects on fire occurrence (Madany and West, Reference Madany and West1983; Romme et al., Reference Romme, Allen, Bailey, Baker, Bestelmeyer, Brown, Eisenhart, Floyd, Huffman, Jacobs, Miller, Muldavin, Swetnam, Tausch and Weisberg2009). The changes with time since grazing we document here for the Plateau are likely due to many of the mechanisms and interactions among them.
Disturbance change
Disturbances, such as those caused by livestock, development of roads and trails, off-highway vehicles and energy exploration can be increasing concerns in the southwestern United States and Plateau (Cordell et al., Reference Cordell, Betz, Green and Owens2005; Duniway and Herrick, Reference Duniway and Herrick2011; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). Tourism on federal lands increased substantially since the mid-1970s, accelerating after the late 1990s (Leaver, Reference Leaver2014; Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). The impacts of off-highway vehicles, roads and highways on vegetation, run-off and erosion processes and wildlife has been found at the landscape scale (e.g. Switalski, Reference Switalski2018; Duniway et al., Reference Duniway, Pfennigwerth, Fick, Nauman, Belnap and Barger2019) but is less well-documented at the vegetation-patch scale as evaluated in our repeat photo interpretation. Similarly, it is less well-documented for non-motorized recreational trails and lightly used off-highway vehicle roads captured in these repeat photographs. Given the proliferation and increased use of recreational hiking, mountain biking and off-highway vehicle trails regionally (Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017), these activities have been a catalyst for change in the areas that are now more accessible to the public (Taff et al., Reference Taff, Benfield, Miller, D’Antonio and Schwartz2019) and often, photos taken by tourists are a source of historical and repeat photography in this and other studies (Kull, Reference Kull2005; Webb et al., Reference Webb, Boyer, Turner and Bullock2007a; Hammond et al., Reference Hammond, Stone and Stone2020). Our change in disturbance indicator was important for explaining vegetation change and results suggest the different responses of the three vegetation communities to change in disturbance. However, we had less robust evidence for human disturbance, with ~180 stakes having evidence of human disturbance, mostly related to recreational activities (e.g. vehicular traffic, hiking/foot traffic), presence of human disturbance between photo pairs was associated with patch shifts for both CP Blackbrush and IM grasslands, but not PJ shrubland/woodlands. The smaller vegetation and often more sparse patches in the CP Blackbrush and IM grasslands likely provides opportunities for human activities to fragment and break up patches more than the larger stature and denser PJ woodlands (Jentsch and White, Reference Jentsch and White2019). While tourism has been a catalyst for land conservation on the Plateau (Leaver, Reference Leaver2014; Mitchell and Gallaway, Reference Mitchell and Gallaway2019), and a source for the historical repeat photos used here, in our results the non-human disturbances (e.g. livestock, erosion processes) appear to be more explanatory for vegetation change than roads, trails and recreation.
Broader implications and methodological limitations
The approach we used here to interpret change between photos, the photos themselves, as well as the explanatory variables all present limitations. There may be compositional changes within the vegetation classes used here that may not be reflected in our change analysis. For example, there are common grassland shrubs such as Ephedra sp. that are resilient to drought (Hoover et al., Reference Hoover, Pfennigwerth and Duniway2021) and shifts toward more shrubs within the IM grassland community may be underestimated. Photo quality varied across stakes and time, which is common as historic photos are often taken with various cameras (Duniway et al., Reference Duniway, Belnap, Webb, Miller, Van Scoyoc and Bishop2026) but may have resulted in an increased ability to detect change based on the differences in photo quality. Additionally, the time of photo capture may have varied in season. Another limitation is we did not obtain precise GPS coordinates of all the vegetation patches captured in the repeat photos, which is likely beneficial for relatively fine scaled (30-m) topographic variables and thus relied on approximations based on camera set up information. Thus, the potential dissonance in analyzing approximations with very broad scale gridded historical weather data would also not capture fine-scale variation due to local topography. However, with changes in technology and increased use of automated image processing there is potential to capture more fine-scale variation which was not available or practical within the timeframe the interpretations were initially done. Finally, livestock use within pastures can be very heterogeneous in response to management (e.g. water developments and mineral supplements) and pasture characteristics (e.g. available forage, rough topography that impedes livestock movement), none of which were accounted for in this study. However, given the limitations, the varied response among areas with different grazing legacies, ongoing human disturbance and varying dominance of invasive species highlights the continued effects of anthropogenic disturbances on ecosystems, despite the protection provided by national parks for most locations analyzed (308 of 338 photo pairs). Additionally, while we only included a few variables associated with geomorphic setting, this work broadly supports previous studies emphasizing the importance of soils and topography in governing ecosystem dynamics (Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017; Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022). We can further validate the value of repeat photography for quantifying vegetation and landscape change in time scales and with details that are not possible with other methods. These insights can be beneficial for managerial and conservation perspectives, as well as broader public appreciation for the dynamic nature of drylands.
Conclusion
Our century-long analysis reveals complex vegetation dynamics on the Colorado Plateau, with variable responses between and within vegetation classes. Recovery trajectories following livestock removal demonstrate the range of grazing legacies on the Plateau, particularly in blackbrush and grassland communities. Climate variability, including the 1950s drought, continues shaping community dynamics in complex ways that can be further evaluated with a more nuanced seasonal analysis to understand possible drought mortality and other mechanisms for change (Browning et al., Reference Browning, Duniway, Laliberte and Rango2012; Hoover et al., Reference Hoover, Pfennigwerth and Duniway2021; Currier and Sala, Reference Currier and Sala2022). Physical landscape settings, especially landform and aspect, provide fundamental controls on community responses to environmental drivers (Duniway et al., Reference Duniway, Benson, Nauman, Knight, Bradford, Munson, Witwicki, Livensperger, Van Scoyoc, Fisk, Thoma and Miller2022). These findings underscore the use of long-term observational data for understanding dryland dynamics and highlight the benefits of considering multiple interacting drivers in ecosystem management (Copeland et al., Reference Copeland, Bradford, Duniway and Schuster2017). As climate change intensifies and recreational use increases, these patterns provide baseline information for anticipating future changes and informing adaptive management across this ecologically significant landscape.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/dry.2026.10019.
Data availability statement
Photo interpretations and explanatory data available from Bishop et al. (Reference Bishop, Kelly, Cheng, Burch, Mikenas, Knight and Duniway2026) and photos used in this analysis are available from Duniway et al. (Reference Duniway, Belnap, Webb, Miller, Van Scoyoc and Bishop2026).
Acknowledgments
We are grateful to Sam Burch for assistance with geospatial analysis; Peekay Briggs providing access to the photo collections at the National Park Service; Charlie Schelz for helping with repeat photo capture and Meg Ripley for creating the study area map. We thank Michael A. Bishop for artwork used for the creation of Figure 3 and the graphical abstract. The graphical abstract was built from Mr. Bishop’s individual plant artwork and the use of generative fill and other editing in Adobe Photoshop by authors. Any use of trade, firm or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Department of Agriculture.
Author contribution
Conceptualization and Funding Acquisition: JB, RHW, MEM, MCD; Data Curation: TBBB, AK, EC, JM; Formal Analysis: TBBB, LSB; Investigation: TBBB, AK, EC, MEM, JB, RHW, MVS, JM; Methodology: TBBB, AK, LSB, MCD; Project administration: TBBB, MCD; Supervision: MCD; Visualization: TBBB; Writing – Original Draft: TBBB, MCD; Writing – Reviewing and Editing: All.
Financial support
This work was supported by the National Park Service (agreement # P14PG00334), Canyonlands Natural History Association, and US Geological Survey Ecosystems Land Change Science Program and Water Resources Research Act Program.
Competing interests
The authors declare none.
Open access
Comments
Dear Editor:
I am pleased to submit our manuscript entitled “A picture is worth a thousand plants: Using historic and repeat photography to quantify long-term change in drylands on the Colorado Plateau” for consideration for publication in Drylands.
This study analyzes over a century of vegetation change on the Colorado Plateau using historic and repeat photography from 1910-2016. We provide unique insights into long-term ecological dynamics in this globally important dryland region and discusses how vegetation communities respond differently to interacting factors such as grazing cessation, climate variation, and geomorphic settings.
We believe this manuscript is particularly well-suited for Drylands for the following reasons:
• Long temporal scope in dryland research: Our study leverages a uncommon long-term dataset spanning over 100 years, providing insights into vegetation dynamics at timescales rarely captured in ecological research. This temporal depth allows us to document recovery patterns following land-use changes and climate variations that would be difficult to detect in shorter-term studies.
• Integration of historical data with modern geospatial analysis: A main unique aspect to this work is the combination of historic repeat photography with GIS and landscape analysis to quantify ecological change across large spatial and temporal scales—an approach that can be applied in drylands worldwide where conventional long-term monitoring data are scarce.
• Direct applications for dryland management under climate change: By identifying how different plant communities respond to multiple and interacting environmental factors across varied landforms and soil conditions, our findings provide land managers with important context for anticipating ecosystem responses to future climate scenarios, particularly relevant as drought conditions intensify across southwestern drylands.
This manuscript represents original work that has not been published elsewhere and is not under consideration by another journal. All authors have approved the manuscript and agree with its submission to Drylands. We have no conflicts of interest to declare.
Thank you for your consideration of our manuscript. We look forward to your response.
Sincerely,
Tara B. B. Bishop, PhD
Utah Valley University Department of Earth Science Email: tbishop@uvu.edu
On behalf of all co-authors: Anne Kelly, Evelyn Cheng, Mark E. Miller, Jayne Belnap, Robert H. Webb, Matt Van Scoyoc, L. Scott Baggett, Jessica Mikenas, and Michael C. Duniway