U.S. Imperiled Species are Most Vulnerable to Habitat Loss on Private Lands
By Adam J Eichenwald, Michael J Evans, Jacob W Malcom
First published: 02 March 2020
Abstract
To stem the ongoing loss of biodiversity, conservation practitioners must distinguish between effective and ineffective approaches for protecting species habitats. Using Google Earth Engine and 31 years of Landsat images, we quantified changes in the habitats of 24 vertebrates listed under the US Endangered Species Act (ESA) and on the International Union for Conservation of Nature (IUCN) Red List across categories of land ownership (eg federal, state, private) in the continental US that are subject to different conservation‐focused legal restrictions. These estimates exclude changes attributable to agricultural conversion and burned areas. The imperiled species we evaluated lost the least amount of habitat (3.6%) on federal lands, whereas losses on private lands without conservation easements were more than twice as high (8.1%). Differences in annual percent loss before and after ESA listing, and between ESA‐listed and Red List species, indicate that the ESA limited habitat loss and was most effective on federal lands. These results underscore the importance of federal lands in protecting habitat for imperiled species and highlight the need to improve habitat protection on private lands for long‐term conservation.
Habitat destruction is the primary driver of global biodiversity loss, and reducing or reversing habitat loss is therefore a critical goal of conservation (Betts et al. 2017). Minimizing habitat destruction is particularly important for the conservation of imperiled species, because habitat loss negatively affects the population sizes (Donovan and Flather 2002) and reproductive success (Kurki et al. 2000) of at‐risk species. Given the current global biodiversity crisis (Betts et al. 2017), it is critical to identify drivers and effective mechanisms for preventing future losses of both habitats and species. Rigorous analyses that evaluate the effectiveness of conservation laws are therefore needed to identify successes and shortcomings in efforts to maximize habitat protection and inform policy decisions.
The US has some of the strongest laws in the world protecting imperiled species, but federal and non‐federal entities are often regulated differently. The Endangered Species Act (ESA) provides a prominent example. The ESA prohibits the take (eg harming, killing, harassment) of protected animals but implements this protection in different ways for federal actors (eg US Forest Service, Army Corps of Engineers, Department of Transportation) and non‐federal actors (eg state wildlife agencies, private citizens, corporations). Incidental take of protected animals as a result of federal actions is authorized under Section 7, which is widely used (Malcom and Li 2015; Evans et al. 2019). In contrast, non‐federal entities receive authorization for incidental take of listed animals through Section 10, a mechanism that is difficult to enforce (Carter 1991). Both processes are initiated by the regulated party, but Section 7 consultation is standard practice for federal agencies, and federal decisions that trigger consultation are publicly visible. In contrast, it is not known how often private entities avoid the Section 10 process. Furthermore, federal land management agencies have specific legal and regulatory conservation mandates (eg Federal Land Planning and Management Act, National Forest Management Act) that protect imperiled species. Protections for imperiled species may therefore be stronger on federally owned lands.
The question remains as to whether differences in conservation law implementation between federal and non‐federal contexts result in observable differences in protections for imperiled species habitat. Weaker protections on non‐federal land would severely limit conservation because so much biodiversity exists outside of protected areas (Jenkins et al. 2015). However, scientists have lacked the data and tools to evaluate patterns of imperiled species habitat loss at large scales. Instead, research has largely focused on small‐scale case studies (Smith et al. 2018). The results of studies of limited geographic or taxonomic scope are difficult to extrapolate beyond the study area, and while informative, are insufficient for informing policy debates concerning effective approaches to biodiversity conservation at the national scale. Large‐scale data and analyses are needed to inform policy and enable oversight of land management agencies (Trainor et al. 2013).
Advances in remote sensing and cloud computing have made it possible to rapidly analyze global satellite imagery datasets, facilitating analysis of habitat loss over larger scales and longer time periods (Song et al. 2018). We used satellite imagery to report on 31 years of habitat loss for 16 ESA‐listed species and eight species listed on the International Union for Conservation of Nature (IUCN) Red List (IUCN 2018) in different land ownership classes in the continental US, excluding agricultural land. Unlike ESA‐listed species, species on the Red List are not legally protected. Differences in protections among species and between federal and non‐federal lands create an opportunity to evaluate how effectively conservation policies in the US protect imperiled species habitat, and to identify strengths and weaknesses in their implementation. Our goal was to quantify differences in habitat protection for imperiled species between federal and non‐federal lands, and to evaluate the degree to which the ESA contributes to habitat protection.
Methods
Study species and area
We measured habitat loss for 24 vertebrates in the continental US from among those listed under the ESA and on the Red List (Figure 1). We selected species with ranges modeled by the US Geological Survey (USGS) Gap Analysis Program (GAP) that contained both federal and non‐federally owned land. GAP models provide a methodologically consistent representation of species geographic distribution at 30‐m resolution; these models are built using a national wildlife habitat relational database consisting of habitat associations described in the published literature (including detailed land cover, elevation, and hydrological characteristics). We used these models to delineate the areas where each species was most likely to be found. The GAP provides maps of suitable habitat for over 2000 listed and unlisted species in the US, most of which have limited ranges that were less useful for our goals (USGS 2018). The 24 species that we selected were those with ranges that encompassed both federal and non‐federally owned land, that relied on habitats that could be detected in satellite imagery, and that collectively inhabited all major ecoregions in the continental US.
Figure 1
Examples of listed species included in our analysis. (a) Gray wolf (Canis lupus), (b) aplomado falcon (Falco femoralis), (c) Florida scrub jay (Aphelocoma coerulescens), and (d) Utah prairie dog (Cynomys parvidens).
D Avery; CC BY 2.0P Burian; CC BY‐SA 4.0 J Gallagher; CC BY 2.0B Thaller; CC BY 2.0
For ESA‐listed species, we also considered an additional delineation of species range consisting of the legally recognized inhabited areas documented and provided by the US Fish and Wildlife Service (FWS; “administrative range”, https://ecos.fws.gov). These administrative ranges vary in how they are designed. Many only represent presence/absence of the species at county‐level resolution, whereas others may be delineated using readily available geological or ecological features, such as watersheds. As GAP distributions are not legally recognized, we included both range representations in order to compare habitat loss within all areas in which a given species may be ecologically likely to occur (ie GAP range) to the subset of areas receiving federal protections (ie administrative range). This comparison helped to further evaluate the impact of federal protections on habitat loss because it defined the explicit areas where ESA protections are legally applicable.
We defined land ownership classes using the US Protected Areas Database v2.0 (PAD; https://on.doi.gov/3bzB0JY) and classified lands as federal, state, non‐governmental organization (NGO), protected private, or non‐protected private. Tribal lands were intentionally excluded from our analyses, as they are legally governed by separate nations. In aggregate, these species represented multiple ecoregions that covered 49% of the continental US (WebTable 1).
Remote‐sensing analysis
We used the Google Earth Engine implementation of the LandTrendr algorithm (Kennedy et al. 2018) to measure loss of imperiled species habitat between 1986 and 2018 from Landsat imagery by identifying sudden changes in the trend of the normalized difference vegetation index (NDVI) at image pixels over time. NDVI is an index of vegetation intensity, and we defined habitat loss as an area where vegetation degraded quickly over a short period of time. We standardized thresholds for identifying breakpoints among habitat types using the mean and standard deviation within each focal species range (Yang et al. 2018). Disturbances smaller than 450 m2 were discarded to reduce oversensitivity. Landscape metrics like fragmentation or connectivity were not considered in our analyses. Because agricultural crop rotations can confound change detection by creating false positives, we masked LandTrendr output with the US Department of Agriculture 30‐m resolution Cropland Data Layer from 2017 (Boryan et al. 2011), excluding agriculture conversions prior to 2017. This created a conservative bias to loss estimates.
LandTrendr is also likely to detect habitat loss due to fire. However, because we could not distinguish between natural and prescribed burns, fire would confound our assessment of legal and regulatory protections; as such, inclusion in our analysis of burned areas up to the year they were burned could potentially introduce a confounding variable. We therefore eliminated burned areas using data from the Monitoring Trends in Burn Severity program (MTBS 2018). The percentage of land excluded due to fire and agriculture per species is shown in WebTable 2.
Habitat loss analysis
We calculated annual percent habitat loss as the proportion of pixels within a species’ range in each land ownership category showing habitat loss in each year. Total pixels within the range remained constant across years. To account for background trends in habitat loss, we standardized annual percent loss estimates by subtracting the mean annual percent loss across all species and land ownership types in a given year from species‐specific annual percent loss estimates; we referred to this measurement as the “adjusted annual loss”. We also calculated total percent loss as the proportion of pixels that were disturbed over the entire period and that did not recover back to their original NDVI values (total percent loss); this was calculated by using the LandTrendr algorithm to identify breakpoints that resulted in NDVI gain rather than loss, then utilizing these breakpoints to exclude pixels that were disturbed but subsequently demonstrated an increase in NDVI values in subsequent years. We excluded years corresponding to the beginning of our data collection (1986) and transitions between Landsat satellites (2001) due to extreme peaks occurring in these periods.
To identify important predictors of habitat loss, we fit linear mixed models estimating adjusted annual loss per species as a function of land ownership (“Zone”), time (“Year”), listing status (“Status”), and whether the GAP or FWS range is considered (“Range”). All models included a random intercept per species to account for correlation in repeated measures. Unless otherwise specified, we considered losses for ESA‐listed species that occurred within the FWS range when the species was listed. Models were estimated in a Bayesian framework using the R package rstanarm (R Core Development Team 2012; Goodrich et al. 2018). We used default priors and sampled 1000 iterations of four Markov chains following a 1000‐iteration burn‐in period. Chain convergence for all parameters was assessed using the &error;statistic, with &error; < 1.1 indicating convergence (Gelman and Rubin 1992). The Watanabe‐Akaike information criterion (WAIC) was used to measure support among competing models (Watanabe 2010); the sum of WAIC weights (ωi) among models that included a given variable was then used to measure variable importance, with ∑ωi > 0.90 signifying important variation in loss as a function of a given variable (Watanabe 2013).
To test for differences in habitat loss among land ownership classes, we fit a candidate set of linear mixed models that included an intercept‐only (null) model and all combinations of the effects of Year, Zone, and Status. This candidate set was fit to adjusted annual loss data from ESA‐listed and Red List species within FWS and GAP ranges, respectively. To test for differences in habitat loss inside and outside administrative ranges, we also fit an identical model set with Range substituted for Status to adjusted annual loss data for ESA‐listed species from within GAP distribution ranges inside and outside of the FWS range. Because Red List species do not have administrative ranges, the candidate sets were fit to different data and evaluated separately. We also tested for differences in percent loss between land ownership types, between ESA‐listed and Red List species, and between ranges, using the maximum probability of effects (MPE) and 95% credible intervals around pairwise contrasts.
As a post hoc analysis, we estimated differences in habitat loss trends over time among the land ownership classes. We considered four trends in adjusted annual loss, representing four simple potential patterns of change: linear, logarithmic, exponential, and quadratic. To identify the appropriate trend form, we fit a set of four mixed models of adjusted annual loss data as a function of year separately for each ownership class to account for potential differences in the trend form among classes. We identified the most supported relationship as the model receiving the lowest WAIC score. Trends among land ownership types were compared in terms of the form of the trend that was most supported, and whether slopes were positive, negative, or effectively zero. We conducted separate analyses for ESA‐listed and Red List species.
Finally, we evaluated three predictions to test the hypothesis that the ESA was the mechanism responsible for observed differences in habitat loss. If the data were consistent with the hypothesis, we expected to find: (1) lower adjusted annual losses within ESA‐listed species ranges after they were listed than before they were listed; (2) no differences in total percent loss among land owned by different federal agencies, as differences would indicate agency‐specific regulations as the mechanism driving loss reduction; and (3) lower total percent loss for ESA‐listed species than for Red List species. We tested the first expectation by estimating the effect of ESA listing on adjusted annual loss using a linear mixed model with a random intercept per species and fixed effects on the interaction between an ESA listing indicator variable and Zone. To test the second expectation, we used a linear mixed model estimating differences in total percent losses within the FWS range of listed species on federal lands as a function of the federal agency that owned the land. Finally, the effect of listing status on total percent losses in GAP ranges for Red List species and in FWS ranges for ESA‐listed species was used to test the third expectation. For ESA‐listed species, we considered only losses that occurred after the species was listed. In all analyses, we inferred differences among groups and meaningful trends if the relevant coefficient had an MPE > 95% and a 95% credible interval around estimated effect sizes that did not include zero.
Results
Variations in adjusted annual habitat losses for imperiled species were explained by land ownership types, listing status (ESA versus Red List), the range considered (ecological versus administrative), and across years. Linear mixed effects models containing a three‐way interaction between Zone, Year, and Status or Range received the most support (ΔWAIC > 13.9; WebTable 3). Model parameter estimates are available in WebTable 4. Among all land types, the smallest proportion of imperiled species habitat loss occurred on federal lands (sample mean ± standard deviation [ ± SD] = 3.6% ± 3.8). This reduction in habitat was significantly smaller than the reductions that occurred on all other land types (difference between groups [Δ] ≥ 6.2%, MPE ≥ 0.97). Species lost the most habitat on non‐protected lands ( ± SD = 8.6% ± 6.3). Total losses on both non‐protected and protected private lands were significantly greater than total losses on all other land ownership types (Δ ≥ 15.0%, MPE = 1.00; Figure 2). No difference (Δ ≤ 1.5%, MPE = 0.69) in net habitat loss was detected between NGO ( ± SD = 4.5% ± 3.8) and State ( ± SD = 4.6% ± 3.6) lands. Finally, total losses were higher within ecological ranges than administrative ranges (Δ ≥ 14.4%, MPE = 1.00; Figure 2).
Figure 2
Rates of habitat loss within imperiled species ranges were lowest on federal lands and highest on non‐protected private lands. Box plots show the distribution of total habitat loss among imperiled species from 1986 to 2017. Percentages differed for US Endangered Species Act (ESA) listed species, depending on whether the potential/ecological range (black) or administrative range (blue) was considered. Total percent losses within International Union for Conservation of Nature (IUCN) Red List species ranges (orange) were not significantly greater than those for ESA listed species according to our pre‐specified Bayesian criteria, but Red List species lost marginally more habitat overall. Horizontal lines within boxes depict median values, boxes represent the interquartile range (25th–75th percentiles), and whiskers (vertical lines) represent 1.5×interquartile range.
Rates of habitat loss within imperiled species ranges were lowest on federal lands and highest on non‐protected private lands. Box plots show the distribution of total habitat loss among imperiled species from 1986 to 2017. Percentages differed for US Endangered Species Act (ESA) listed species, depending on whether the potential/ecological range (black) or administrative range (blue) was considered. Total percent losses within International Union for Conservation of Nature (IUCN) Red List species ranges (orange) were not significantly greater than those for ESA listed species according to our pre‐specified Bayesian criteria, but Red List species lost marginally more habitat overall. Horizontal lines within boxes depict median values, boxes represent the interquartile range (25th–75th percentiles), and whiskers (vertical lines) represent 1.5×interquartile range.
Trends in annual adjusted losses over time also differed among land ownership types (Figure 3). A linear decline was the best supported model of annual habitat losses on NGO lands for both ESA‐listed and Red List species, indicating consistently declining loss rates (WebTable 4). Annual habitat losses increased logarithmically on non‐protected private lands for ESA‐listed and Red List species, indicating increasing annual losses that stabilized over time (Figure 3). Among Red List species, there was no change in annual losses over time on state or protected private lands, whereas annual loss of habitat for ESA‐listed species on state and protected private lands increased over time (WebTable 5). On federal lands, quadratic trends indicated decreasing annual loss rates from 1986 to 2005 for ESA‐listed species, and from 1986 to 2008 for Red List species, after which annual loss has been increasing (Figure 3).
Figure 3
Between 1986 and 2018, federal and non‐federal lands exhibited different trends in annual rates of habitat loss within imperiled species ranges. (a and b) Data points show the unweighted annual total percent habitat loss across species within each land management zone, adjusted by the mean loss in a given year. Trends over time were similar between (c) ESA‐listed species and (d) Red List species. Trend lines were estimated from the marginal relationship between year and adjusted annual loss accounting for random intercepts per species.
Between 1986 and 2018, federal and non‐federal lands exhibited different trends in annual rates of habitat loss within imperiled species ranges. (a and b) Data points show the unweighted annual total percent habitat loss across species within each land management zone, adjusted by the mean loss in a given year. Trends over time were similar between (c) ESA‐listed species and (d) Red List species. Trend lines were estimated from the marginal relationship between year and adjusted annual loss accounting for random intercepts per species.
Species experienced significantly less habitat loss after they were protected under the ESA compared to before (Δ = 25.0%, MPE = 1.00; Figure 4). Total habitat loss was consistent among the six agencies managing federal lands in this study. Pairwise differences in total loss between agencies did not differ from zero (MPE < 0.95), with the exception that losses on National Oceanic and Atmospheric Administration lands (7,405,233.3 km2, all on beaches) were significantly lower than those on Bureau of Reclamation (2,896,025,400 km2) and Department of Defense (519,493.5 km2) lands (Δ ≥ 6.5%, MPE ≥ 0.96). ESA‐listed species lost marginally less habitat than Red List species overall (Δ = –2.62%, MPE = 0.67), a difference that was most pronounced on federal lands (Δ = –4.57%, MPE = 0.75).
Figure 4
Species lost less habitat within their ranges when they were ESA listed. (a) Box plots show the distribution of annual habitat losses for each ESA‐listed species as the percentage of species habitat within administrative ranges lost each year before and after listing. (b) Data for habitat losses before listing were unavailable for species listed prior to 1986. Vertical lines within boxes depict median values, boxes represent the interquartile range (25th–75th percentiles), whiskers (horizontal lines) represent 1.5×interquartile range, and solid circles depict outliers.
Discussion
Critically evaluating approaches to conserving habitat is imperative to conserving imperiled species, yet historically scientists have been able to conduct research only at small scales. We used satellite data to examine habitat losses within various land ownership classes across the US to assess the efficacy of habitat preservation under different types of land ownership. Habitat losses for imperiled species were lowest on federal lands and highest on protected private and non‐protected lands. Moreover, the results of our analysis indicated that federal lands provide protections for imperiled species habitat, and also identified important shortcomings of protections outside of federally owned areas.
We found evidence that the ESA likely contributed to habitat protections on federal lands. Species lost less habitat after they were ESA listed than before, and because habitat loss was consistent across lands managed by different federal agencies, it is unlikely that agency‐specific regulations (eg National Forest Management Act) were primarily responsible for limiting these losses. This finding provides further support for the hypothesis that legal protections under the ESA contributed to reduced habitat loss. The ESA is the only law in the US that provides non‐discretionary habitat protection across federal lands; although the National Environmental Policy Act (NEPA) requires federal agencies to report on the environmental impacts of planned actions, unlike the ESA, it does not contain any requirements that these impacts be avoided, minimized, or mitigated. Therefore, while federal laws other than the ESA almost certainly provide some degree of protection for imperiled species, our results are consistent with the hypothesis that the ESA provides unique protections across federal lands.
There are several reasons why the implementation and effectiveness of conservation laws are likely reduced on private lands in comparison to federal lands. To begin with, individual property rights are highly protected in the US (Knight 1999), and conservation laws like the ESA include exemptions for private actors (eg no Section 7 protection for critical habitat on private lands, absent involvement of federal agencies or funding). Even where laws do apply, a lack of visibility and the voluntary nature of conservation initiatives on private lands mean that regulations may still fail to provide protections to habitat (eg oil development within the geographic range of the lesser prairie chicken [Tympanuchus pallidicinctus]; Melstrom 2017). In addition to the lack of oversight over private actors, landowners may engage in preemptive habitat destruction to avoid perceived ESA land‐use restrictions (Lueck and Michael 2003; Brook et al. 2003), although such behavior does not always occur (Melstrom 2017). Thus, inefficient protections outside of federal lands will undermine past, present, and future conservation work.
Uneven protections are problematic to imperiled species conservation for three reasons. First, species do not only inhabit federally managed areas. In the US, there is considerable misalignment between biodiversity and federally protected areas (Jenkins et al. 2015): for more than half of the ESA‐listed species, more than 80% of their geographic range is on private land (FWS 2009). Even though the highest rates of habitat loss were found to occur on private lands, our results likely underestimate losses on private lands, given that we did not account for losses due to agricultural expansion. The cropland data layer indicated that between the years 2008 and 2017, 127,100,181.6 km2 (4.4%) of imperiled species habitat on private lands were converted to agriculture, whereas only 838,440 km2 (0.1%) of federal lands were similarly converted, rates consistent with national‐level agricultural conversion trends (Lark et al. 2015). Furthermore, high and increasing losses on state lands were an indication that current state protections are also inadequate. Relying on federal protections alone is therefore insufficient to conserve imperiled species.
Second, species within protected ranges can still be threatened by habitat loss in adjacent areas. Isolated islands of protected habitat surrounded by development reduce the value of protected areas (Radeloff et al. 2010) and create extinction debts (ie delayed extinction following habitat loss) for species that might not be immediately recognized to be at risk (Tilman et al. 1994). In addition, although habitats change over time naturally, anthropogenic climate change will drive drastic transformations in ecosystems (Jones et al. 2009). Model projections suggest that some vertebrate ranges will move out of protected areas in response to future climate change – US national parks are expected to lose up to 20% of their mammal species due to these species’ ranges shifting onto other land ownership types (Burns et al. 2003). These interactions highlight the need for conservation laws that consider landscapes from a holistic perspective.
Finally, uneven protections likely limit species recovery. Protecting species only within administrative ranges may constrain them to their current geographic distributions, instead of allowing for movement or expansion to potential future ranges (Carroll et al. 2010). Greater habitat losses within ecological ranges compared to administrative ranges are not surprising given that ESA interventions and other federal regulations are usually limited to these administrative areas, potentially reflecting an inability to protect unoccupied suitable habitat. Our results highlight that failure to extend protections to unoccupied habitat may impede recovery, especially for wide‐ranging species.
This study examined a limited sample of the 1600 ESA‐listed species and does not include land covered by croplands or areas burned by fire. As such, our results must be interpreted with caution, particularly given that our comparisons of habitat losses before and after ESA listing were limited to eight species. Furthermore, we selected species for which habitat loss is a major threat, and our results cannot be used to infer the efficacy of legal protections against other threats. In addition, USGS GAP models were created in 2001, which may have resulted in the exclusion of habitat losses that occurred over the period 1986–2001. This exclusion presumably was unbiased by land ownership types and therefore should not affect the results of our relative comparisons across land types. However, exclusion of habitat losses prior to 2001 would affect analyses comparing habitat losses before and after ESA listing, leading to an underestimation of habitat loss prior to listing. The reduction in habitat loss following species listing identified by our analysis may therefore have been even greater than our results indicate (Figure 4).
Importantly, our method is applicable to any species for which a discrete range can be identified, and not merely those modeled by GAP habitat distributions. We chose GAP models for consistency among species’ ranges, but the LandTrendr algorithm can be applied to measure habitat loss within any designated range for a given species.
Measurement and evaluation are critical for successful biodiversity protection. The results of our analyses demonstrate that federal protections have successfully reduced habitat loss for imperiled species in the US, but they also identify gaps that compromise wildlife conservation. To successfully recover imperiled species, current differences in legal protections between federal and private lands must be minimized through increased enforcement, greater conservation incentives, and/or compliance assistance. All listed species experienced a net loss of habitat, even on federal lands, which may be symptomatic of systemic conservation challenges (Laurance 2010). Furthermore, greater habitat loss among Red List species indicates that imperiled species without legal protection remain at risk. As species can become functionally extinct before losing 30% of their population (Saterberg et al. 2013), legally protecting species only once they face immediate threats may still result in extinctions (Rohlf 1991). As human well‐being is intimately interconnected with biodiversity (IBPES 2019), failure to protect imperiled species will have consequences beyond species extinctions.
Acknowledgements
We thank J Miller, M Evansen, and M Moskwik for assistance in formulating project goals and analyses. J Miller and JM Reed provided reviews and feedback that improved the manuscript.
