Including Condition into Ecological Maps Changes Everything—A Study of Ecological Condition in the Conterminous United States

Abstract

In 2021, the Biden administration signed an executive order to protect 30% of American lands by 2030. Accomplishing this ambitious goal in the U.S. requires understanding the relative contribution of public and private lands toward supporting biodiversity. New approaches are needed because existing approaches focus on quantity of habitat without incorporating quality. To fill this need, we developed a 30 m resolution national habitat condition index (HCI) that integrates quality and quantity measures of habitat. We hypothesized that including an evaluation of the quality of habitat at landscape scales, both in conservation-focused preserves and working lands would provide a better assessment of the value of geographies for conservation. We divided the conterminous U.S. by major land cover type and into natural and cultivated lands and then spatially mapped multiple anthropogenic stressors, proximity to aquatic habitat, and vegetation departure from expected natural disturbance regimes. Each map layer was then scored for site impact and distance decay and combined into a final national index. Field observations providing scored relative ecological conditions were used for HCI calibration and validation at both CONUS and regional scales. Finally, we evaluate lands by management (conservation versus working lands) and ownership (public versus private) testing the value of these lands for conservation. While we found regional differences across CONUS, functional habitat was largely independent of protection status: working lands provide clear habitat and other values. These results are relevant for guiding strategies to achieve the U.S. 30 by 30 goals. Where similar data exist in other countries, analogous modeling could be used to meet their national conservation commitments.

1. Introduction

Conserving biodiversity, particularly in the face of global climate change, is critical for maintaining the natural systems that support humanity in, and beyond, the 21st Century. A recent Executive Order [1] by U.S. President Biden is an acknowledgement of this global need. This Order prescribes conserving at least 30% of U.S. lands and freshwater and 30% of U.S. ocean areas by 2030 (henceforth “30 by 30”). To achieve this ambitious goal (current land in conservation in the United States is approximately 12% [2]), cost will be a limiting factor. The majority of lands not in conservation in the United States are used for production of goods as farms, ranches, and production forests (henceforth, “working lands”). A conversion of enough of these working lands from their current uses to preserves wholly dedicated to biodiversity may not be feasible at a scale necessary to achieve 30 by 30 due both to cost of acquisition and opportunity costs of lost production. Costs aside, simply creating preserves may also not be adequate in achieving conservation goals either. U.S. and global efforts have resulted in highly variable protection across ecosystems and species [3]. Environmental degradation within preserves, even as well protected as National Parks and National Wilderness Areas in the US, are well-documented [4,5,6,7]. The biodiversity value of working lands is also well-documented and could potentially contribute to “30 × 30” goals. Managed forests, rangelands, even the unused margins of croplands, regularly support native species and ecological processes, often providing critical habitat corridors [8,9,10,11].

Spatial datasets have been developed to support efforts to determine locations for targeted conservation action, designate reserves, identify habitat corridors, and estimate populations based on available habitat [12,13]. They are also used to estimate habitat loss through conversion to other uses [14], but many landscape-scale analyses are based primarily on the extent of and trends in land cover (i.e., quantity and location of habitat) and do not account for the condition of the land within land cover types [15,16]. As a result, we may have a clear estimate of how much habitat of various types exists across landscapes, but often lack an understanding of the condition of what was lost or what remains. To determine the success of 30 by 30, agencies and environmental organizations need a spatial tool that measures the quality, not just the quantity of the preserved lands with respect to biodiversity.

Incorporating habitat quality data into ecosystem assessment and planning efforts should improve upon current approaches by providing additional insight into whether a given location can support focal species and healthy ecosystem function. While an area may be designated on a map as being a particular habitat type, the capacity for any such area to support species is a function of habitat quality and landscape context [17]. Quality measures also establish a common currency that enables comparisons of condition across the range of a given ecosystem type [18] or across multiple land cover types, ownership boundaries, and geographic regions. This type of comparison is widely recognized as critical for smaller-scale conservation applications (e.g., biodiversity offset programs) [19]. For example, Australia’s Bush Broker/Bush Tender programs and Habitat Exchanges in the U.S. rely upon assessing the relative value of habitats or ecosystems as compared to an ideal for that system [15,20]. Under these programs, condition is assessed as the combination of both relative quality and quantity rather than as a binary measure (i.e., an area of land “is” or “is not” habitat). These more complete assessments of condition can be used to make more complex conservation decisions, for example, to attain a species population goal by conserving either more land of lower habitat quality or less land of higher quality [21,22]. However, these small-scale tools are generally developed for site-level conservation or mitigation programs and typically lack a landscape-scale component [19,23]. Given this limitation, site-level assessments can rarely address questions regarding the value of each parcel of land to long-term viability of a species or ecosystem.

To address the limitations of area-only approaches, we used quantity and quality-based data sets to develop a habitat condition assessment of the conterminous U.S. (CONUS). The Habitat Condition Index (HCI) is intended to help conservationists and planners assess the ecological condition in which they work. The HCI provides a measure of the generalized, landscape-scale condition of habitat, or amount of functional habitat (quality × quantify) for a wide range of species, and is intended to supplement, not replace, other measures of habitat quality that are typically focused on smaller spatial scales (e.g., habitat suitability indices). The landscape-scale variables incorporated into the HCI have been demonstrated to influence general habitat conditions, rather than identify habitat quality for individual species [24,25,26]. Where more specific habitat models are desired, species- and region-specific variables should be included (e.g., specific environment and resource availability). The HCI can provide the landscape context base layer for these finer-scale efforts.

Our aim was to build the HCI from datasets that both expand on and simplify prior efforts to evaluate the ecological condition of landscapes [27]. Because habitat assessments require timely data, we chose datasets that are open access and periodically updated. We incorporated datasets on fragmentation, distance from development, vegetation departure from natural composition and structure, and proximity to aquatic systems as factors that influence habitat quality. Fragmentation can alter environmental conditions of adjacent habitat [28,29], impair species migration or gene flow [30,31,32,33], reduce the habitat quality of the patch itself due to exposure to other land cover types that may be unsuitable for species [26,30], increase predation rates on species within the patch [34,35], and alter disturbance frequency [36]. Similarly, developed land cover types, including urban and built structures, roads, and agriculture have well-documented negative impacts upon native species and adjacent ecosystems [37,38,39]. Anthropogenic conversion can interact with ecological processes in negative ways, resulting in habitat degradation [40] and biodiversity loss [41,42]. As with other analyses of this type [43,44], we excluded high intensity agricultural production (e.g., row crops) and developed lands from the HCI. The decision to exclude row crops from other working lands (used for food or fiber production) resulted from insufficient data on which to develop a habitat quality measure for these high intensity agricultural uses. Developed lands we assume to provide little significant habitat for native species.

We included additional datasets that reflect vegetation structure and composition, because it affects habitat quality for plants and wildlife within an ecosystem. For example, fire suppression allows open forest stands to close, changing availability of appropriate sites for establishment, foraging, and nesting or bedding [45,46,47]. Many species depend on both upland and wetland sites for some portions of their life cycles [48,49]. A growing body of literature supports the hypothesis that streams subsidize food webs and energy transfer of terrestrial ecosystems, and that the magnitude of these subsidies can have large positive impacts on habitat quality [50,51,52].

To assess the accuracy of the HCI tool, we conducted a thorough validation. Following validation of the tool, we used the HCI to test five hypotheses at the CONUS scale:

• Areal estimates of habitat area significantly overestimate the capacity of land to support biodiversity, because they may assume the quality of natural habitat areas is pristine (or 100% quality which is a 100 HCI score). By measuring ‘functional’ area of habitat (i.e., area × HCI value), the HCI values more accurately represent the relative capacity of the land to support healthy, thriving ecosystems and the species they support when compared to a hypothetical “pristine” landscape. Thus, comparing areal estimates of habitat with HCI scores on those same lands tests the degree of degradation of the natural landscape across the CONUS.
• Lands designated for biodiversity conservation alone will have higher average landscape condition (HCI values) than lands designated for multiple use. This hypothesis examines the differences between various levels of conservation status as indicators of the effectiveness of land protection policies.
• HCI values of natural and semi-natural lands do not correlate with ownership (public versus privately owned). This hypothesis tests whether management influences habitat quality.
• Privately-owned natural and semi-natural working lands (i.e., lands in range or silvicultural production), regardless of management or use, comprise a large percentage of total functional habitat of the CONUS. This hypothesis tests the assumption that privately-owned lands may be critical for wildlife conservation if public lands are not sufficiently large or of high enough quality [53,54,55,56,57].
• The sum of HCI-weighted values for currently protected and unprotected lands in the CONUS are greater than that needed to conserve 30% of the U.S. This hypothesis tests whether the 30 by 30 goal is theoretically possible.

This work establishes a replicable system for the production and validation of national-scale maps that could be employed in the U.S. to document the ecological condition of terrestrial ecosystems. This same method can be periodically updated to enable tracking of change in condition over time. Methods described in this manuscript may be readily adapted elsewhere given available datasets for other countries. Finally, this work may provide a critical tool for measuring the role of landscapes under different types of management, from conservation-focused preserves to agriculture in supporting native species and ecosystems.

2. Materials and Methods

Our methodology consists of performing spatial analyses on several national-scale datasets and merging them mathematically to create one map describing a generalized index of habitat condition (HCI). We then determined the correlation between HCI and both land management (degree of conservation versus use for human purposes such as sylviculture and grazing) and land ownership (public vs. private). The HCI was developed and tested in four steps: (1) identifying and assembling the component data layers, (2) calibration, (3) validation, and (4) hypothesis testing (Figure 1).

2.1. Base Data Layer Selection and Spatial Analysis

Four data sets were created as potential predictors of habitat condition: (1) habitat fragmentation, (2) anthropogenic influence, (3) vegetation departure from pre-European conditions, and (4) proximity to aquatic resources (distance to water bodies). To produce each of these input layers we used existing spatial datasets (henceforth, “base layers”) with a native resolution of the raster datasets of 30 m. The source and relevant qualitative information for each base layer is described below in each respective section. When vector-based layers (e.g., roads) were required as inputs to developing a base map, we converted the native dataset to 30 m rasters. We then compiled these input layers to derive the composite HCI map (Figure 1).

2.1.1. Habitat Fragmentation

We used the National Land Cover Dataset [58] to derive the base fragmentation layer. The NLCD dataset describes the land cover of the United States, defining 20 classes, 16 of which are natural land cover types (e.g., Grassland/Herbaceous and Evergreen Forest) and four that depict various human-created urban or built-up land cover types of varying density. We chose to use the 2011 NLCD dataset as it represented the most applicable land cover data available across the entire CONUS at the time of this research. Figure 2 depicts 10 of the 20 NLCD classes used subsequently for summarizing analysis results.

Fragmentation was defined as discontinuities in land cover types that produce distinct, separate patches. We estimated fragmentation levels using the Graphical User Interface for the Description of Objects and their Shapes Toolbox [59], a freeware program designed to assess fragmentation and connectivity of raster-based maps of landscapes. Within the GUIDOS Toolbox, we utilized morphological spatial pattern analysis (MSPA, [60]) which conducts segmentation on a binary image to identify mutually exclusive morphometric feature classes describing the shape, connectivity, and spatial arrangement of image objects on a categorical map. A binary map of natural/non-natural was created by combining NLCD land cover classes. The non-natural class was created using the four developed classes and the cultivated lands. The MSPA process identifies core and non-core areas of patches, in this case, based upon a three-pixel (90 m) edge width. Following the methods outlined in [59] we adapted a modified index weighting scheme to incorporate additional fragmentation categories (perforation, bridge, loop, and branch) provided by GUIDOS Toolbox. The resultant input layer describes the relative amount of fragmentation of the binary natural land cover map we created.

2.1.2. Anthropogenic Influence

To create the anthropogenic influence input layer, we applied methods proposed by Hak and Comer [44] that provide a repeatable, empirically validated set of distance decay functions expressing the ecological impact of each stressor. Transportation input layers were derived from TIGER roads data (https://www.census.gov/cgi-bin/geo/shapefiles/index.php; accessed on 18 February 2019) and the urban development layers using NLCD development classifications. We augmented TIGER road data by merging it with data provided by Open Street Map (https://www.openstreetmap.org; accessed on 18 February 2019) that provided a more complete set of unpaved roads. For each anthropogenic layer (e.g., roads or development) identified in S1 Table S1 in the Supplementary Materials, a Euclidean distance was calculated extending away from each feature. These Euclidean distances were used as inputs into distance decay functions developed by Hak and Comer [44] based on empirically validated observation points. The result is a combined layer that represents the relative impact of anthropogenic sources that extend beyond their location into the surrounding landscape.

2.1.3. Vegetation Departure from Pre-European Conditions

To determine relative vegetation degradation, we used the vegetation departure index included within the LANDFIRE models for 2014 [61]. The LANDFIRE vegetation departure index models the relative change in vegetated conditions from historical pre-European settlement conditions for several hundred natural vegetation types occurring across CONUS. Reference conditions represent simulated historical vegetation composition and structure resulting from historical disturbance occurrence and severity [62]. To characterize current conditions, LANDFIRE generates successional class maps representing the current successional state of vegetation as determined by comparing LANDFIRE existing vegetation data products (current vegetation type, cover and height) with the defined successional composition and structure rules outlined by experts using the Vegetation Dynamics Development Tool [63]. Current conditions can then be compared to reference conditions to determine a measure of departure. All spatial data are generated at 30 m pixel resolution.

2.1.4. Proximity to Aquatic Resources

We applied methods developed by [52], who found that a negative power function best matches the observed relationship between terrestrial habitat condition and distance to aquatic habitats across a wide range of ecosystem types. We applied a normalized aquatic subsidy curve [52] by including a weighted addition to the HCI value of any pixel within one kilometer of a river, lake, or wetland.

2.2. Calibration

The evidence supports each of the component layers of the HCI as predictors of landscape-scale habitat quality when analyzed separately. To ascertain whether the HCI is parsimonious, as well as determine the relative contribution of each layer in a composite product (i.e., the coefficient for each layer), we conducted a model calibration exercise. We created a Bayesian framework that tests each possible combination of predictor variables fit to a set of real-world field assessments of landscape condition distributed throughout the CONUS. For response variables to indicate real-world landscape condition, we utilized 32,377 point locations from natural community occurrences derived from Nature Heritage Program datasets which rank landscape condition from “A” (excellent ecological condition) to “D” (poor ecological condition) [64]. We reclassified these points into a binary response variable where ranks A and B were considered good ecological condition and ranks C and D, poor condition. We normalized each of the predictor layers (z-transformation) to assure that no data layer in the multivariate has outsized predictive power simply due to differences in absolute values that might span orders of magnitude. We assumed there may be regional variation in which predictor layers best describe landscape condition. For this reason, we incorporated a five-ecoregion map (Figure 3) into our model, allowing parameter values to vary by-region for each.

Using the BRMS package in R (https://cran.r-project.org/web/packages/brms; accessed on 12 December 2019), we then tested each possible combination of variables, both with and without eco-regionally varying parameter values (a total of 32 models). The general goodness of fit of the models were compared using Watanabe-Akaike Information Criteria (WAIC) and Leave-One-Out Cross-Validation (LOO) [65]. Model rank and posterior model probability (PMP, the probability that a given model is the best, most parsimonious model given the data) were then determined using Bayesian bootstrap stabilized pseudo-Bayesian model averaging (BB-pseudo-BMA, Ref. [66]).

Based upon model fit rank, which penalized overfitting, the best fit model included each of the four base layers with regionally varying parameters. The slope for each predictor of the chosen model can be found in Supplemental S3 (Supplementary Materials). Tables ranking and describing how well each tested model predicted the Natural Heritage occurrence response data are in Supplemental S1 (Supplementary Materials). We calculated the CONUS-scale HCI and regionally varying HCI maps based upon the parameters of the best model in the CONUS and regional analyses, respectively. To do this, we multiplied the parameter values for each of the HCI components from the respective best-fit models to each pixel of these component maps using the Raster Calculator function in ArcMap. Finally, we rescaled these datasets to a one to one-hundred scale to create an easy to interpret map of quality.

2.3. Creation of the HCI

We next created the HCI map based upon the parameter values obtained from the calibration. Since we found that a map with regionally varying parameters best fit the validation points, we created different HCI maps for each of the ecoregions utilizing best-fit parameters for each (see Supplemental S1 (Supplementary Materials)). Each of the HCI components was multiplied by the appropriate parameter value for a given region and then summed. The resulting five regional HCI maps were then combined into one map for all of CONUS (Figure 4). Finally, this map was rescaled using the linear rescale function in ArcGIS to generate a distribution in which the lowest HCI value is one and the highest is 100.

2.4. Validation

We tested the predictive power of the HCI dataset using the remaining 10% of the ecological integrity point location data (n = 3597). Again, we used the Natural Heritage ratings of A and B as actual high ecological condition and C and D as low ecological condition sites. Using the “predict” function in the “stats” package in R (https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/predict; accessed on 18 December 2019), we generated predicted habitat quality values for each validation point. This yielded a probability that each of the validation points should be classified as one (high quality). For example, a point might be rated as 0.8233 meaning it has an 82.33% probability of being high quality (see Supplemental (Supplementary Materials) for more detail). Because logistic-based analyses calculate the probability that a given point is classified as either a zero or one, the predict function assigns a probability that each of the 3597 locations were high or low quality. We simplified this output by classifying any point predicted at 50% or greater likelihood of being a good quality site as a one. Those points below this threshold were classified as zero. Finally, we compare observed and predicted classes using a confusion matrix allowing for not only an overall assessment of model fit, but a clear picture of potential model biases (i.e., errors in the model tend to mis-classify low quality data points as high quality or vice versa).

Specific data source information and the subsequent geoprocessing necessary to derive the HCI are illustrated in Figure 1. Additional details regarding the parameters for creating each of the base map layers, the Bayesian model selection, the weighting process using calibration point data and the validation of the final HCI map are provided in the Supplementary Materials.

2.5. Hypothesis Testing

The first hypothesis is that the total ‘functional’ hectares of habitat (i.e., area × HCI value), are significantly lower than the total area of habitat. We generated separate HCI maps for each of the “natural” land cover types (see Figure 6 for results for each of these “natural” land cover types) in the 2011 NLCD using the “Extract by Attributes” function in ArcGIS. For each land cover class (e.g., “Evergreen Forest) in each region (e.g., “Western Forest”), we summed the total HCI value of all pixels (See Equation (1)). Finally, we determined the percent functionality of land cover class by dividing the summed HCI value by the total number of pixels in each given land cover class.

$${\mathrm{Functionality}}_{\mathrm{ClassX}, \mathrm{RegionY}}={\displaystyle \sum }_{\mathrm{i} =1}^{\mathrm{n}}{\mathrm{HCI}}_{\mathrm{i}}$$

(1)

To test our second hypothesis, that protected areas designated solely for biodiversity protection would have higher quality habitat than those designated for multiple-uses (“working lands”), we used the Protected Areas Dataset of the United States (PAD-US (https://doi.org/10.5066/F7G73BSZ; accessed on 18 February 2019)) spatial data layer. These data provide a CONUS-wide map of areas with a Gap Analysis Project (GAP) status of 1 through 4 (greatest to least restrictive conservation management on public lands or private lands bound by conservation easement) (Figure 5). For this analysis, we created a fifth GAP status, which represents private working lands (i.e., lands in range or silvicultural production) without conservation easements. Given that legal protection of the land increases with lower GAP Status numbers (e.g., GAP Status 1 is most protected), a confirmation of our hypothesis would be that each sequentially higher GAP Status would have, on average, lower HCI values. We tested the hypothesis that better-protected lands would have higher HCI values by employing ANOVA and Tukey Highly Significant Difference analyses on an equalized stratified random sample (200,000 points in each Status from 1 to 5).

We used the same analyses used for the second hypothesis but included GAP status 5 to test our third and fourth hypotheses: that working land habitat quality would be independent of public or private land ownership and that privately-owned natural and semi-natural working lands comprise a large percentage of total functional habitat of the CONUS, respectively. We determined the total area of all natural and semi-natural lands based on the NLCD dataset and then compared the functional areas of the non-working lands (GAP Status 1 and 2) and the working lands (Gap Status 3 and 5) using the zonal statistics function in ArcGIS.

To test our fifth hypothesis, we calculated the HCI values of all areas of natural and semi-natural lands. We then divided that area by the total CONUS land area to determine whether the “30 by 30” objective could feasibly be attained by augmenting existing protected areas with unprotected area scoring as high for the HCI.

3. Results

Overall, we find that 30% of the natural and semi-natural habitat in the conterminous US has an HCI value below 50, indicating substantial degradation (Figure 4). Results also show variation in the habitat condition across ecoregions, land cover classes, and private versus public ownership. In general, the western ecoregions (Western Forest and Desert) exhibit the highest HCI values and the eastern ecoregions (Southeast Forest and the East and Northeast Forest) the lowest.

While each land cover class we tested had areas of high and low HCI values, the relative distribution of HCI values within each class was markedly different, indicating that some land cover types are in much better condition than others at the landscape scale. The majority of land area of evergreen forest, shrub/scrub, grassland, and barren (often high elevation exposed rock or snow and ice) had HCI values of 80 or higher, while the majority of the land areas of deciduous forest, mixed forest, pasture, and woody wetlands had HCI scores of 60 or below (Figure 6).

Hypothesis Results

Hypothesis 1 (H1).

Areal estimates of habitat area significantly overestimate the capacity of land to support biodiversity, because they assume the quality of natural habitat areas is pristine (or 100% quality which is a 100 HCI score).

The results from our HCI model support our first hypothesis that the area of ‘functional’ habitat (i.e., area × quality), is considerably less than the total area of habitats as designated by the NLCD. Natural and semi-natural land cover covers 6,044,800 km² (76%) of the CONUS, whereas the total functional area of these same lands is only 4,253,800 km², or 54% (Figure 7). We also find considerable variation in the proportion of functional hectares across CONUS regions. In several ecoregions, functional habitat was less than 20% of total habitat area, indicating most of these lands may be severely degraded. Ecoregions with functional habitat more than 20% less than total habitat are Central Plains (26.4%), Southeast Forest (41.1%) and Eastern and Northeastern Forest (48.4%) (Figure 7).

Hypothesis 2 (H2).

Lands designated for biodiversity conservation alone will have higher average landscape condition (Hci values) than lands designated for multiple use.

We reject this hypothesis, because we did not find that protection-focused management regimes (i.e., where extractive or high impact human use is prohibited) are always associated with higher HCI values (Figure 8). We had hypothesized that public lands with GAP status of 1 and 2 (designated exclusively for biodiversity conservation—below designated as “non-working” lands) would have the highest average HCI values. However, while GAP status did influence HCI value (F = 1,057,733, p = 2 × 10⁻¹⁶), GAP 1 (mean HCI = 85, SD = 10) and 3 (mean HCI = 82, SD = 10) had significantly (p adj < 0.00000 in Tukey HSD test) higher mean values than GAP 2 (mean HCI = 75, SD = 14) and 4 (mean HCI = 74, SD = 17) (Figure 6). GAP 5 lands (private, unprotected lands) showed the lowest HCI scores (mean HCI = 65. SD = 18) as expected.

Hypothesis 3 (H3).

Hci values of natural and semi-natural lands will not vary based on ownership (public versus privately owned).

We reject this hypothesis, because the average HCI value of working lands managed by private landowners (GAP Status 5 forests and grasslands, Figure 9a) was different than the average HCI values of land managed for multiple purposes by public agencies (GAP Status 3 and 4, Figure 9b). The average HCI value on private working lands (forests and grasslands) (mean = 52.645, SD = 14.646) was significantly lower (F = 154105, p = <2 × ⁻¹⁶) than that on public working lands (mean = 57.965, SD = 19.715).

Hypothesis 4 (H4).

Privately-owned natural and semi-natural working lands (i.e., lands in range or silvicultural production), regardless of management or use, comprise a large percentage of total functional habitat of the conus.

As hypothesized, privately-owned working lands contribute significantly to the total functional habitat across CONUS. Natural and semi-natural working lands, classified by NLCD as forests or grassland/pasture that are in either public (GAP status 3 and 4) or private (GAP status 5) ownership, represented 54% of the total functional habitat within the CONUS study area (Figure 7). Private lands (Figure 7) represent over 31.7% of this habitat. This contribution of private land to the total functional habitat varies by ecoregion as the percent of land in private holdings varies (Figure 7). Most private in-use working lands in the CONUS are maintained as grassland/pasture (21.2%), or forest (20.6%) of the total CONUS land area respectively.

Functional Forest working (Gap 3–5) and non-working (Gap 1–2) habitat broken down by ecoregion, (Figure 9a) indicates that 87% of forests are on working lands. This relationship reflects the relatively small proportion of lands designated as Gap 1–2. These relative proportions vary regionally, with Western Forests and Northeastern Forests including the highest overall proportions of functional forest area, followed by the Southeastern Forests (Figure 10a). When public vs. private land ownership is considered (Figure 10b), public lands (mostly National Forests), retain most functional forest area in the Western Forests and Desert ecoregions, while private lands support proportionally more functional forest area in the Great Plains, Southeastern Forests, and Northeastern Forests.

Similar to forests, the majority (91%) of functional grassland habitat is on working lands (Gap 3–5) (Figure 10a). Again, this is primarily due to the relatively small proportion of grasslands designated as Gap 1–2. These relative proportions vary regionally that differ from forests, with the Great Plains and Desert ecoregions including the highest overall proportions of functional grassland area, followed by the Western Forests ecoregion (Figure 9a). Functional grassland habitat found within “non-working” protected areas is concentrated in the Desert and Western Forest ecoregion, and almost none is found in the Great Plains and eastern ecoregions.

When public vs. private land ownership is considered (Figure 9b), private lands support the most functional grassland area (65%), a relationship dominated by ownership patterns in the Great Plains, with lower amounts in Southeastern Forests and Northeastern Forests. Public lands (mostly BLM lands and National Forests), retain most functional grassland area in the Desert and Western Forest ecoregions, respectively.

Hypothesis 5 (H5).

The sum of Hci-weighted values for currently protected and unprotected lands in the conus are greater than that needed to conserve 30% of the biodiversity.

The total area of natural and semi-natural lands scoring high according to the HCI and falling within GAP 1–5 lands (4,253,800 km²) divided by the total CONUS land area (8,080,463 km²) is 52.6%, supporting the hypothesis. Therefore, the “30 by 30” objective for U.S. lands could feasibly be attained with additional conservation attention on GAP 3–5 lands scoring as high by the HCI.

4. Discussion

The validation of the HCI by field observations indicates that this tool reflects landscape-scale habitat patterns, and therefore could provide a common currency that enables comparisons of condition across the range of a given ecosystem type, or across multiple land cover types, ownership boundaries, and geographic regions. The results from hypothesis testing demonstrate that: (1) the amount of functional habitat is smaller than total habitat area would suggest; (2) patterns of functionality vary across the U.S. by ecoregion and by major vegetation category (i.e., forests vs. grasslands, etc.), and do not directly correspond to management designations as described by GAP protection status; (3) working lands (public and private) contribute significantly to functional habitat across the U.S.; (4) while private lands are providing functional habitat hectares, improving habitat quality on private working lands would have significant impacts on the overall numbers of functional habitat area across CONUS and within regions; and (5) the “30 by 30” objective for U.S. lands could feasibly be attained with additional conservation attention on GAP 3–5 lands scoring as high for the HCI.

The first hypothesis test suggests that many estimates of extant habitat across CONUS are significant over-estimates, because the habitat is likely too fragmented or degraded to support many sensitive species associated with these ecosystems. As such, while the total hectares reflect potential habitat, heavily degraded areas may provide little, if any habitat for native species. The HCI can provide guidance for conservation decision-makers on where improvements to the HCI value is possible (e.g., where reducing the vegetation departure value would improve the overall value), and how to improve functionality in those places.

These results also suggest that we should not make assumptions regarding the functionality of habitat based on protected status of land, but instead measure functionality directly using metrics that include quality, like the HCI. HCI scores were not always higher under more restrictive management regimes. While we do find that GAP Status 1 lands have a higher mean HCI value than privately owned working lands, publicly owned working lands (GAP Status 3) had slightly higher mean HCI value than GAP Status 1 (82 vs. 75 respectively). Interpretation of these patterns is complex, as the different GAP status types represent both different proportions of the CONUS, and of regions within the CONUS (Figure 5) GAP Status 3 lands represent the highest proportion of total public land area of CONUS, with the vast majority found west of the Great Plains. These inconsistencies between more biologically-based habitat condition and protected status likely apply everywhere GAP status is used as a proxy for condition.

The combined finding that while habitat quality is better on average on public than on private working lands, but the latter contribute significantly to functional habitat, indicates that conservation strategies aimed at maintaining or increasing the habitat value of working lands may be of great benefit to many species and ecosystem goals within the CONUS. The value of utilizing the HCI, and a functional habitat approach in-general, as a lens for landscape analysis may be most apparent in the northeastern and southeastern U.S, in which significant recovery of forest habitat in the mid-19th century following two centuries of widespread clearing [67]. In this region, while reforestation has been widespread, habitat fragmentation remains high, particularly on private lands (e.g., [68]). The NLCD map shows that over 890,000 km² of various forest types exist, yet the average condition of forest is only 52.85% representing roughly 470,000 km² of functional habitat. In contrast the western forest ecoregion contains much less area (61% of the northeastern region), yet nearly the same amount of functional habitat (97% of northeastern region) because it is in better condition (85%). This information could be useful for decision-makers considering restoration and preservation strategies in the two regions.

Conservation strategies on lands with average functionality in western forests may lead to relatively good outcomes particularly because the majority of those lands are publicly owned, reducing the threat from development; the best conservation strategy in the region would be to increase federal management attention on habitat restoration. Conversely, protection of lands in the east may benefit from much greater targeting to assure that the highest quality forests receive focus. Because the lands in the east are more fragmented and subject to greater anthropogenic influence from roads and human structures, our results predict that apparently healthy forest will support fewer and smaller populations of at-risk species unless fragmentation is reduced. This result suggests that land protection and restoration strategies designed to reduce fragmentation (e.g., road removal and revegetation to restore native composition and structure) could create beneficial outcomes. These types of restoration may be wholly consistent with maintained production values on those same properties; similar lessons about fragmentation reduction on working lands will likely apply to geographies well beyond CONUS [30].

While the regional variation in mean HCI score may largely reflect land use patterns across the continental US, intra-regional variation may be far more informative for conservation planning and action. This could be particularly relevant to advancing the “30 by 30” goal. While, as noted above, this goal appears to be attainable nationally with additional conservation attention on GAP 3–5 lands, particularly focusing on high-scored HCI lands with minimal costs to conserve or improve, this strategy becomes more challenging when one segments the national landscape by region, by land cover class, or by more rigorous forms of mapping ecosystem diversity (e.g., [14]). Within each region, we see high variability in HCI within each land cover class, providing examples of particularly high quality or degraded landscapes. Therefore, achieving appropriate levels of ecological representation will likely require conserving some larger areas with lower HCI scores to be equivalent to higher scoring areas that are simply not present for that type of ecosystem.

Because the HCI is a landscape-scale tool for interpreting broad habitat condition, it should not be used to identify small parcels of land for conservation action. However, the HCI may be used as a tool for broader-scaled analysis (e.g., counties, states, ecoregions, etc.) to determine which of the component elements of the HCI may be particularly affecting the quality of the natural landscape, and thus a good target for actions aiming at restoration or recovery.

Three existing landscape-scale tools exist that most closely align with the purpose and methods deployed here with the HCI. Two of those existing tools, including the NatureServe landscape condition model [44] and LANDFIRE vegetation departure model [61], have been integrated either indirectly or directly into the HCI. The representation of land use categories, along with relative weightings and distance decay functions of the landscape condition model were used in the development of HCI, and the LANDFIRE vegetation departure model was used directly. In each case, those models aim to represent important components of ecological condition in terrestrial ecosystems, while the HCI aims to bring those components together for a more comprehensive representation. A third general model to quantify ecological integrity of landscapes [43] most closely aligns with the NatureServe landscape condition model aiming to map the degree of human modification. That model integrated a larger number of sources of ecosystem stress than the NatureServe model and—in contrast to the NatureServe model that used field observations in calibration and validation—it used photo interpreted samples to estimate human modification and then associated those results with numerous land cover classes. While the national map outputs of these several models appear to be similar, important differences emerge—reflecting differences in model inputs and methods—when viewed at more local scales.

Data on most HCI inputs (habitat fragmentation, anthropogenic influence, proximity to aquatic resources) may be found in countries around the world. Common land cover and land use data may not always be available at the same spatial and thematic resolutions as used here, but they do exist in forms that are feasible for application of the HCI methodology. The Vegetation Departure data set used here is unique to the United States, so parallel data sets, most necessary in temperate latitudes with fire-dependent terrestrial ecosystems, are unlikely to be available in comparable form. However, analogous models can likely be developed in other geographies and would be valuable for both incorporating landscape condition into 30 × 30 planning and evaluating the generality of HCI results in the U.S.

Our results illustrate that conservation strategies should include working lands, both publicly and privately-owned, as part of a comprehensive strategy to conserve habitat and connectivity. Not only are working lands spatially intermingled and socially connected (https://www.blm.gov/programs/natural-resources/rangelands-and-grazing/livestock-grazing; accessed on 10 June 2021), but HCI values suggest they can provide valuable habitat. While HCI values are generally higher on lands with more-restrictive protections (GAP 1–2), working lands represent the majority of the functional habitat within the CONUS. Despite habitat degradation on working lands, the absolute area of these lands more than offsets the lower HCI values.

The value of working lands may be even greater when the surrounding context of potential future landscapes is considered. The HCI treats working lands and non-working lands of the same type (e.g., forests) as contiguous patches of the same type with respect to fragmentation. That is, a patch is considered fragmented when it borders upon a dissimilar (in this example, non-forest) patch. We recognize that mosaics of landscape patches are often associated with high biodiversity, and sometimes represent a natural state (e.g., savannah woodlands). However, as the landscape types indicated by the NLCD are generally broad and the cell size large, our fragmentation algorithm generally is accounting for transitions between larger patches of dissimilar landscapes rather than fine scale mosaiced habitats. With the habitat quality-based methods employed in the HCI, the conversion of working lands to dissimilar land cover creates loss of habitat in both the patch that is converted and in all adjacent patches of the same land cover type. For example, if a patch of working forest is converted to a parking lot, not only would the HCI measure a loss of habitat associated with the now parking lot, but all adjacent forested areas would also have reduced functional habitat due to fragmentation and proximity to build land cover. Because most landscapes consist of a matrix of both working and non-working lands, conversion of patches of working lands would have profound effects on the greater landscape.

One of the reasons for creating the HCI was to evaluate the quality of lands with differing levels of protection and varying amounts of public use. Refinements to the underlying datasets would improve both the HCI model and its application. For example, the NLCD database could be substituted to better represent the gradient from cultural to ruderal, semi-natural and natural land cover classes as is now depicted by LANDFIRE (https://landfire.cr.usgs.gov; accessed on 23 August 2021). We had also hoped to reflect the variation of habitat quality across croplands as influenced by adjacent land uses but had insufficient data on which to base a model layer. As land use shifts become necessary for other purposes [69], such data could inform where change should be prioritized and incentivized.

5. Conclusions

In this article, we developed a new spatial analysis tool to estimate the functional condition of landscapes in the CONUS. We demonstrate the utility of this tool for conservationists by testing several hypotheses relevant to the goals of “30 × 30”. Using the HCI to estimate functional habitat illustrates how human activities are impacting the natural landscapes of the CONUS and demonstrates the need to weigh lands by functionality rather than just geographic area of habitat for conservation purposes. We identify considerable variation in landscape condition by cover type and across regions. Our results show that the CONUS has experienced a 43% loss of habitat (a combination of land cover conversion and degradation of remaining natural lands) since European settlement. At a finer scale, our work shows that habitat quality of remaining natural lands has also been further impacted by human activities and can indicate where management may have the greatest impact.

Optimistically, this work also clearly demonstrates that lands outside of publicly designated protected areas in the United States may have immense value as natural habitat. While managing and improving our designated areas (GAP status 1 and 2) should remain a high priority into the future, conserving and improving the natural habitat conditions on our working lands should greatly improve conservationists ability to achieve large-scale conservation goals.

The HCI may also be a useful component of analyses that may integrate other landscape components necessary for more in-depth analysis. Such analyses might conceivably include habitat delineation and evaluation for known species of concern, climate change future scenario modeling for species ranges and habitat suitability, water conservation and reapportionment, Native American lands conservation, and any number of working lands conservation valuation projects.

Ultimately by assessing the condition of the landscape, our work indicates that relatively degraded lands have considerable value as habitat. Further, the HCI product itself may be useful as a tool for conservation action, serving as a means to evaluate which areas of land may hold the most potential as natural habitat based on quality. Even modest improvements in the condition of natural and semi-natural lands outside of public protected areas may be a critical component of conservation in the 21st Century [1].