Journal list menu

Volume 15, Issue 3 e4779
Open Access

Spatial extent drives patterns of relative climate change sensitivity for freshwater fishes of the United States

Samuel C. Silknetter

Corresponding Author

Samuel C. Silknetter

Department of Biological Sciences, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA


Samuel C. Silknetter

Email: [email protected]

Search for more papers by this author
Abigail L. Benson

Abigail L. Benson

U.S. Geological Survey, Science Analytics and Synthesis, Denver, Colorado, USA

Search for more papers by this author
Jennifer A. Smith

Jennifer A. Smith

Department of Biological Sciences, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA

Department of Integrative Biology, The University of Texas at San Antonio, San Antonio, Texas, USA

Search for more papers by this author
Meryl C. Mims

Meryl C. Mims

Department of Biological Sciences, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA

Search for more papers by this author
First published: 14 March 2024

Handling Editor: Wyatt F. Cross


Assessing the sensitivity of freshwater species to climate change is an essential component of prioritizing conservation efforts for threatened freshwater ecosystems and organisms. Sensitivity to climate change can be systematically evaluated for multiple species using geographic attributes such as range size and climate niche breadth, and using species traits associated with climate change sensitivity. These systematic evaluations produce relative rankings of species sensitivity to aid conservation prioritization and to identify relatively sensitive species that may otherwise be understudied or overlooked. Due in part to biogeographic constraints, species assemblages change across regions and spatial extents; yet, the degree to which spatial factors influence relative rankings of species sensitivity is unclear. The spatial extent of multispecies analyses may alter relative rankings of species climate sensitivity; alternatively, relative climate sensitivity may be conserved among spatial scales, resulting in consistent identification of sensitive species among regions and spatial extents. We investigated how spatial extent influences our understanding of relative climate sensitivity for 137 native freshwater fishes of the United States that were representative of taxonomic, trait, and geographic diversity. Using publicly available occurrence data from the Global Biodiversity Information Facility, we calculated a systematic, geographically derived index of climate change sensitivity for study species at national and regional extents, including within four major hydrologic subregions of the United States. We examined the effects of spatial extent on the relative ranking of climate sensitivity among species, and we explored relationships among climate sensitivity, species traits, and conservation status at regional and national extents. We found that climate sensitivity rankings of species were influenced by spatial extent in some specific instances, but that relative rankings were largely conserved across spatial scales. However, correlations among geographically derived climate sensitivity rankings and species traits associated with climate sensitivity were variable across scales and regions, suggesting that links between geographic rarity and species traits may be scale-dependent in some cases. Finally, we found few associations between climate sensitivity and current conservation status among species. Systematic approaches to quantifying climate sensitivity may offer an opportunity to identify sensitive but overlooked species for pre-listing actions such as monitoring or conservation agreements.


A major challenge of the current biodiversity crisis is to determine how best to prioritize a limited pool of conservation resources for the maintenance of biological diversity. To meet this challenge, multispecies frameworks are used that assess vulnerability to climate change by ranking species relative to one another (Foden et al., 2013; Wheatley et al., 2017). Approaches for assessing species vulnerability to climate change vary widely, as do the methods that these approaches employ (Pacifici et al., 2015). Despite an increased understanding of climate change's impacts on species (Foden et al., 2019), challenges remain for vulnerability assessments to shape effective conservation actions (Butt et al., 2016; de los Ríos et al., 2018). One impediment to conservation planning has been the high data requirements and computational intensity required for many assessment approaches, such as models of bioenergetics (Briscoe et al., 2016) or the mechanistic niche (i.e., Kearney & Porter, 2009). These impediments are particularly pronounced for species with limited locality or life history data (Bland et al., 2015, 2017; Roberts et al., 2016). Therefore, new approaches must be developed that can assess relative climate change vulnerability across multiple species with a wide range of data availability.

Species vulnerability to climate change can be characterized with three dimensions: extrinsic climate exposure as well as the intrinsic factors of sensitivity to climate variability and adaptive capacity (Dawson et al., 2011; Foden et al., 2013; IPCC, 2007). The sensitivity and adaptive capacity dimensions of vulnerability, collectively termed intrinsic sensitivity throughout the remainder of this manuscript, can be assessed across multiple species despite wide-ranging data availability. The intrinsic dimensions of vulnerability include characteristics that contribute to a species' ability to persist and/or capacity to adapt to climate change (Foden et al., 2013). From a management perspective, intrinsic sensitivity assessments can help identify at-risk yet poorly studied species for which additional information may be beneficial in informing conservation status listings or other protections. For example, determining the specific effects of climate exposure on a given species is an endeavor complicated by numerous geographical, biological, and ecological factors (Foden et al., 2019). As an intermediate step in the conservation planning process, assessments of intrinsic climate sensitivity can be used by managers to identify species at greatest relative risk, thus providing target species for more complex climate change vulnerability assessments.

Intrinsic sensitivity is linked to both geographical and biological attributes of a species (Beever et al., 2016; Foden et al., 2013; Pacifici et al., 2017; Pearson et al., 2014; Thurman et al., 2020, 2022), and thus, analyses may be conducted using both rarity and traits-based classifications (Mims et al., 2018). Geographic characteristics of a species range, such as native distribution or extent of endemism, can be derived from occurrence data and reflect the presence of an organism at a given place and time. For example, a species' area of occupancy (AOO) is the summed area of occupied sample units (Hartley & Kunin, 2003). This is an important component of geographic rarity (Rabinowitz, 1981) and can be calculated using individual occurrence data (Estrada et al., 2015; Maggini et al., 2014; Smith et al., 2020). Occurrence data can also inform climate niche breadth, or the range of climatic conditions that a species occupies (Ralston et al., 2016). Species with a small AOO or a narrow climate niche are predicted to be more at risk for both geographic rarity and extinction than species that are more common (Slatyer et al., 2013). This prediction is based on several hypotheses, including (1) species with smaller ranges have lower adaptive capacity (Hossain et al., 2018; Thurman et al., 2020) and (2) species with narrow climate breadth are more sensitive to climatic changes than those that tolerate a wider breadth of climate conditions (Mims et al., 2018; Rinnan & Lawler, 2019).

Many biological traits are associated with climate change vulnerability and are linked to intrinsic climate sensitivity of species. Although the driving mechanisms of trait-vulnerability relationships can be complicated, many of the observed relationships are consistent across taxonomic groups. For example, a large body size allows species to disperse across greater distances (Jenkins et al., 2007), providing the capacity to track suitable climate conditions (Thurman et al., 2020). Diet breadth may influence species vulnerability during periods of extreme climatic changes. During drought conditions, omnivorous fishes were less vulnerable than diet specialists, suggesting diet generalists are buffered from the most severe climate impacts (Chessman, 2013). Habitat breadth may vary widely among species, yet, in most cases, habitat or microhabitat specialists are less tolerant of environmental or climatic changes than species with broad habitat use (Foden et al., 2008). A recent study by Smith et al. (2019) illustrates how species traits can influence different dimensions of vulnerability. Traits limiting a species ability to persist in a location, such as poor ability to regulate temperature, may lead to high sensitivity, whereas poor dispersal ability may prevent a species from avoiding harmful climate effects due to low adaptive capacity (Smith et al., 2019).

Recent work by Foden et al. (2019) has outlined the utility of traits for assessing climate change vulnerability. A traits-based approach is founded on the association between biological traits and impacts from climate change and can be used to rank species relative sensitivity and adaptive capacity (Foden et al., 2019). As the availability of trait databases for a range of taxa continues to grow, so too does the potential of traits-based analysis to inform intrinsic sensitivity and vulnerability of species to climate change (Estrada et al., 2016; Gonzalez-Suarez et al., 2013; Poff et al., 2010; Rocha-Ortega et al., 2020). Combining traits-based measures of climate sensitivity with measures that are geographically derived may increase the utility of climate vulnerability assessments by providing biological explanations that underpin geographic responses and predictions of intrinsic sensitivity (e.g., Garcia et al., 2014). However, the linkages between geographic ranges, species traits, and climate change vulnerability are complex and influenced by taxonomic and geographic scales of assessment (MacLean & Beissinger, 2017). Different assessment approaches can be considered redundant if they provide consistent predictions of intrinsic climate sensitivity or complementary if the predictions of these approaches are inconsistent. What remains is the question of whether, or to what degree, the linkages between geographic and traits-based approaches are consistent across regions and spatial extents.

Geographically derived and traits-based measures of climate sensitivity may vary spatially with analytical approach. For example, measures of geographic rarity such as AOO can be constrained by the limits of spatial grain or extent of a given study (Hartley & Kunin, 2003; Smith et al., 2020), and regionally distinct species pools and novel life history traits may arise from unique biogeography (Mims et al., 2010; Olden et al., 2010). The effects of spatial scale have also been documented for climate-driven extinction risk in desert fishes (Fagan et al., 2005) and climate sensitivity in montane mammals (Rinnan & Lawler, 2019). For freshwater fishes of the United States, previous work at a regional extent (Pacific Northwest) found that geographically derived predictions of relative climate sensitivity were not correlated with life history traits (Mims et al., 2018). However, at a national extent, body size of freshwater fishes positively correlated with range size, and both were associated with reduced climate sensitivity (Giam & Olden, 2018). The misalignment of results from geographically derived and traits-based assessments of climate sensitivity at different extents suggests that simultaneously conducting assessments at multiple extents is required for revealing when scale-dependencies are present and under what scenarios (Mims et al., 2018).

In this study, we performed an intrinsic sensitivity analysis at multiple spatial scales to generate geographically derived indices of rarity and climate sensitivity (RCS) for a subset of freshwater fishes native to the conterminous United States (N = 137). Our study species were selected to represent taxonomic, trait, and geographic diversity of freshwater fishes in the United States. Using publicly available occurrence data to quantify species AOO and climate niche breadth for each species, we addressed three objectives. First, we used a geographically derived sensitivity index to provide relative rankings of intrinsic sensitivity at both national and regional extents, with regions delineated by watershed. We then evaluated correlations of the sensitivity index, as rank values, among extents and regions. Second, we characterized the composition and dispersion of six biological traits with links to climate change vulnerability at national and regional extents and determined whether these traits correlated with geographically derived sensitivity index values. Though we only used a subset of species from each region, we hypothesized that geographically derived sensitivity may be decoupled from traits-based sensitivity in regions like the Western United States where biogeographic processes have shaped unique species assemblages. Third, we determined how sensitivity index values relate to conservation status at state and federal levels. If status listings capture geographic rarity, we would expect strong positive correlations between conservation status and intrinsic sensitivity index values. However, observations of high intrinsic sensitivity for unlisted species may point to species whose vulnerability to climate change is currently underestimated. Used together, these objectives can help to understand how intrinsic sensitivity analysis is influenced by spatial extent and in turn be used to support the identification of species warranting further conservation assessment.


Candidate species selection and attributes

Our goal for this assessment was to select a subset of freshwater fishes native to the conterminous United States that represented a range of taxonomy and geography and that met minimal data requirements (Figure 1). This study followed work by Smith et al. (2020) that investigated geographic rarity of freshwater fishes and retained their methodology for selecting a representative pool of >150 “candidate species.” First, we used data from Mims et al. (2010) to identify all freshwater fishes native to the United States (Figure 1A). We excluded most species documented as introduced outside of their native range by the U.S. Geological Survey (USGS) Nonindigenous Aquatic Species (NAS) database (USGS, 2018), though some NAS-listed fishes were included to maintain regional representation. For these NAS species, non-native occurrences were identified via native range distributions from the NatureServe database's Explorer tool (NatureServe, 2010) and removed, leaving only native occurrence records. We included at least one species from as many families and genera as possible to maximize taxonomic coverage and included additional species in proportion to genera richness to maintain taxonomic representation. We used the count of eight-digit hydrologic unit code (HUC) subbasins (HUC-8), as reported in the IchthyMaps database (Frimpong et al., 2015, 2016), to ensure that species selected had range sizes representative of the variability observed for all freshwater fishes of the United States. In addition to the abovementioned methods retained from Smith et al. (2020), our study objectives required candidate species to meet trait data requirements (described in detail below) and to have conservation status information (including a “Not Listed” designation) reported in the U.S. Endangered Species Act and the USGS National Species of Greatest Conservation Need list (compiled from 2015 State Wildlife Action Plans, USGS, 2015).

Details are in the caption following the image
(A) We selected candidate species to represent the range of taxonomic and geographic diversity observed in native freshwater fishes of the United States. Selection was made using publicly available data sources, including the U.S. Geological Survey's Nonindigenous Aquatic Species (NAS) database. Conservation status listings were accessed for all candidate species through state and federal sources. (B) We obtained occurrence records from the Global Biodiversity Information Facility ( for all candidate species and filtered records to ensure quality of records, included only records from native ranges, and ensured an adequate number of species for each region. HUC, hydrologic unit code.

For regional analysis, we delineated four regions within the United States based on watersheds with distinct biogeography (Hocutt & Wiley, 1986; Mayden, 1988). The resulting regions are of variable size and species richness (Figure 2) but reflect documented patterns of native freshwater fish diversity (Sheldon, 1988; Warren & Burr, 1994) that may drive regional patterns of intrinsic climate sensitivity. Each region consisted of multiple USGS two-digit hydrologic units (HUC-2; Figure 2). The West region included the Colorado, Great Basin, California, and Pacific Northwest HUCs. The Gulf region included the Rio Grande and Texas-Gulf HUCs. The Mississippi region (abbreviated to “Miss” in Tables 2–4) included the six HUCs within the Mississippi River drainage. Finally, the East region included the New England, Mid Atlantic, and South Atlantic-Gulf HUCs. The regional analysis excluded the Great Lakes and the Souris-Red-Rainy HUCs because the majority of occurrences were located outside of the conterminous United States. Furthermore, the large lakes present in these HUCs (i.e., Laurentian Great Lakes) are not subdivided at the HUC-12 subbasin level, resulting in large AOO estimates regardless of the spatial variation in occurrence records.

Details are in the caption following the image
The full study extent of the conterminous (lower 48) United States with four geographic regions, delineated by major hydrologic units (HUC-2 watersheds, U.S. Geological Survey). West includes Colorado, Great Basin, California, and Pacific Northwest HUCs; Gulf includes Rio Grande and Texas-Gulf HUCs; Mississippi includes the six HUCs within the Mississippi River drainage; and East includes New England, Mid Atlantic, and South Atlantic-Gulf HUCs. We excluded the Great Lakes and the Souris-Red-Rainy HUCs (gray) from regional analyses because the majority of both drainages are outside of the conterminous United States. Sample size and percent regional representation (based on regional totals calculated from NatureServe) reflect the final list of study species after the application of all filters.

Occurrence data filtering

We obtained occurrence records from the Global Biodiversity Information Facility (, data accessed August 25, 2021) for candidate species using the package rgbif (version 2.2.0, Chamberlain et al., 2020) in R version 4.0.3 (R Core Team, 2021); all analyses were conducted in this R version except where noted. We then applied a series of filtering steps to occurrence records (Figure 1B). First, we used the occ_issues() function in rgbif to remove occurrence data with missing or invalid coordinates, contradictory or improbable sampling dates (in the future or pre-1700), or mismatches between coordinates and country of origin. Next, we clipped occurrence records to the study extent (conterminous United States, Figure 2), and we removed occurrence records located in estuaries (EPA, 2017; estuary mapping and other spatial data are detailed in Appendix S1: Table S1). Smith et al. (2020) incorporated expert review of occurrence data in their species selection process and determined that regardless of NAS status, an additional filtering step was required to exclude non-native or improbable occurrences; we remedied this issue by spatially filtering records using native range distributions from the NatureServe database's Explorer tool (NatureServe, 2010). We required a minimum of 50 occurrence records to retain candidate species in national analyses; however, this threshold was reduced to 25 occurrences for species only occurring in the West to retain representation of several genera found only in that region and with fewer than 50 occurrence records. For regional analyses, we required a minimum of 10 records per region (following Mims et al., 2018).

Rarity and climate sensitivity index

To systematically assess intrinsic sensitivity, we used a RCS index (Mims et al., 2018) that incorporates AOO and climate niche breadth (measured as climate sensitivity, CS). Prior work using a highly similar pool of study species found that estimates of range size were strongly correlated across analytical approaches and data sources with no detectable bias of taxonomy (Smith et al., 2020); we found similarly high correlations for AOO calculated using both occupied HUC-12 subbasins (watersheds) and occurrence records buffered with a 1-km radius. We opt to report AOO methods and results for watersheds because it represents a spatial grain that is biologically relevant for aquatic taxa. Additionally, HUC-12 watersheds are coarse enough to encompass stream segments and smaller habitat systems (Frissell et al., 1986) yet represent finer spatial resolution than landscapes (Angermeier et al., 2002) or HUC-8 grains (McManamay & Frimpong, 2015) that are commonly used for continental analyses. The map overlay function over() in package sp (version 1.4-1, Bivand et al., 2013) was used to associate native occurrence records with watersheds. A watershed was considered occupied when one or more records occurred anywhere within the subbasin boundary, with data filters (Figure 1B) applied to reduce the likelihood of false-positive occurrences to the extent practical. We then summed the cumulative area (in square kilometers) of all occupied watersheds for each species. AOO values were scaled from 0 to 1, then subtracted from 1, resulting in a metric where values near 0 indicated species with large geographic extents whereas values near 1 indicated species with small extents (geographically rare).

Climate sensitivity, our measure of climate niche breadth, was calculated using the Parameter-elevation Relationships on Independent Slopes Model (PRISM) AN81m spatial climate dataset from 1895 to 2017 (Appendix S1: Table S1; PRISM Climate Group, 2020). Using a weighted average extract function from the package raster (version 3.0-7, Hijmans, 2019), yearly PRISM climate data were upscaled from a spatial grain of 4-km2 raster cells to HUC-12 watersheds (Silknetter et al., 2023). We then used these data to build a 30-year moving temporal window of climate conditions. We selected a 30-year duration to match the United Nations World Meteorological Organization's 30-year climate normals, which are commonly used as a reference period average for computing climate anomalies (Arguez & Vose, 2011). We included the climate variables of annual precipitation, annual means of maximum and minimum daily average air temperature, maximum daily air temperature in August, and minimum daily air temperature in January (Appendix S1: Table S2). The moving window was calculated using the appropriate function (i.e., mean for annual precipitation and minimum and maximum for temperature variables) for the year of the observation and 29 years prior such that each HUC-12 watershed had an annual value per climate variable that captured medium-term variation in climate trends. For each study species, we then used occurrence records to estimate their putative, contemporary ranges, excluding occurrence records prior to 1924 that lacked 30 years of climate data (unavailable before 1895) and records after 2017 that lacked stable PRISM climate data. We calculated the SD of each moving window climate variable across all occupied watersheds and scaled those values from 0 to 1 using the approach from Mims et al. (2018). We then calculated the mean of the five scaled climate variable SDs and subtracted it from 1 such that values near 1 indicate species with low niche breadth (high climate sensitivity) and values near 0 indicate species with a high niche breadth (low climate sensitivity). Finally, we combined AOO and CS (see Data Availability Statement for a link to externally archived files) into the RCS index by assigning equal weights to both, with RCS scores ranging from 0 to 1 for which high scores indicate rare species with a small AOO and/or a narrow climate niche.

Regional RCS analyses

We calculated regional (RCSR) scores for each of the four regions and compared those with national scores (RCSN) (Figure 3). To do this, we clipped occurrence data to the spatial extent of each region and recalculated AOO and CS scores. Regional scores were rescaled from 0 to 1 such that RCSR scores are relative to other study species in the same region. We used univariate one-way ANOVA and Spearman rank correlation coefficients to evaluate differences among RCSN and RCSR scores.

Details are in the caption following the image
Rarity and climate sensitivity (RCS) scores for 137 study species of freshwater fishes. All species have a single national score (RCSN), as well as scores for each region with a minimum of 10 occurrences (RCSR). Species are sorted in decreasing order of intrinsic risk at the national scale: (A) the most sensitive species and (B) the least sensitive species. For the most highly sensitive species, RCS scores are highly conserved across regions and scales. As intrinsic risk decreases, variation in RCS scores generally increases.

Relating RCS to traits and conservation status

We assessed the RCS index's relationship with species traits to determine whether these attributes provided redundant or complementary information about intrinsic sensitivity. We selected six focal traits (described below) from the FishTraits database (Frimpong & Angermeier, 2009) and hypothesized potential links of each trait to intrinsic climate sensitivity. Many life history and ecological traits are known to be highly correlated with one another, representing energetic trade-offs in life history investments (Cortés, 2000; Stearns, 1989). Further, traits related to demographic processes have been used to distinguish generalized life history strategies employed by multiple taxa (Pianka, 1970), including fishes (Winemiller & Rose, 1992), and links between traits and climate change sensitivity may depend upon trade-offs and interactions among traits. Evidence suggests that considerations of individual traits may be useful for understanding mechanisms that drive species responses to global changes (Williams et al., 2010).

We included four life history traits with hypothesized links to intrinsic climate sensitivity (Table 1) that were widely available for our study species in the FishTraits database: Body size (log-transformed maximum total length, in centimeters)—larger organisms tend to exhibit greater active dispersal (Jenkins et al., 2007) that allows individuals to shift-in-space to track suitable bioclimatic conditions (Thurman et al., 2020). Alternatively, large body size may prevent fishes from accessing colder water and diverse microhabitats provided by small streams that might otherwise serve as a climate refuge in dendritic river networks (Giam & Olden, 2018). Longevity (lifespan in the wild, in years)—long-lived species can increase survivorship during climate extremes by reducing or suspending reproductive effort (Martin & Mouton, 2020). Maturation age (age at reproductive maturity for females, in years)—early maturation allows opportunistic or early successional species to rapidly recolonize habitats following severe physical disturbances (Schlosser, 1990). Fecundity (maximum reported, measured as count)—the production of abundant eggs may result in greater surviving offspring in suitable environmental conditions (Winemiller & Rose, 1993). Alternatively, the production of fewer eggs may increase survivorship under harsh environmental conditions (Bears et al., 2009). We also included two ecological traits with hypothesized relationships with intrinsic sensitivity: Diet breadth (count of trophic ecology traits employed; Appendix S1: Table S3)—omnivorous or diet generalist species are buffered from the effects of extreme climatic changes like drought (Chessman, 2013). Habitat breadth (count of habitat preference traits employed; Appendix S1: Table S4)—habitat generalists have greater realized niche space (Brown, 1984) and may tolerate a greater level of climatic change than habitat specialists (Foden et al., 2008).

TABLE 1. Hypothesized relationships among traits and intrinsic climate sensitivity.
Trait (units) Measure and scoringa Hypothesized relationship with intrinsic climate sensitivity Supporting references
Body size (cm) Log-transformed maximum total length (−) Larger taxa have greater dispersal potential for tracking suitable habitat Jenkins et al. (2007) and Thurman et al. (2020)
(+) Larger taxa are unable to access small-stream microhabitats Knouft and Page (2003)
Longevity (years) Longevity based on life in the wild (−) Long-lived taxa increase survival by suspending reproductive effort Martin and Mouton (2020)
Maturation age (years) Age at reproductive maturity for females (+) Early maturation allows for rapid habitat recolonization following severe physical disturbances Schlosser (1990)
Fecundity (count) Maximum reported fecundity (+) Production of fewer eggs may increase survivorship under harsh environmental conditions Bears et al. (2009)
(−) Production of abundant eggs results in more offspring in suitable environmental conditions Winemiller and Rose (1993)
Diet breadthb Count of trophic ecology traits (−) Diet generalists are buffered from effects of extreme climatic changes Chessman (2013)
Habitat breadthb Count of habitat preference traits (−) Habitat generalists tolerate more climatic changes than specialists Brown (1984) and Foden et al. (2008)
  • a All traits are available and explicitly defined in the FishTraits database (Frimpong & Angermeier, 2009).
  • b Binary coding was used for diet breadth (n = 11) and habitat breadth (n = 25) traits (Appendix S1: Tables S3 and S4, respectively); counts of binary responses are used to calculate species-level trait breadth.

We tested correlations between species traits and national and regional RCS scores using univariate Spearman rank correlation coefficients and performed a principal coordinates analysis (PCoA) of Gower distance based on a trait matrix for all species with ≥50% trait data available. The PCoA approach was used to visualize dissimilarities by finding the principal coordinates (axes) that best describe the trait data in a low-dimensional space (Gower, 1966); we then evaluated the significance of eigenvectors using Monte Carlo randomization (1000 permutations). We conducted a permutational multivariate analysis of variance (PERMANOVA) and an analysis of multivariate homogeneity of group dispersions (PERMDISP2) to evaluate differences in regional position (centroid) and dispersion of species traits, respectively. Analyses were conducted using package vegan (version 2.5-6, Oksanen et al., 2019) in R version 3.6.3 (R Core Team, 2021).

We also used Kruskal–Wallis and Dunn's tests to examine whether predictions of intrinsic climate sensitivity using RCS scores differed significantly by conservation status. We assigned a single status classification to each species based on the greatest level of protection reported at the national extent. In descending order, these included federally “Endangered” and “Threatened” species, “State Listed” Species of Greatest Conservation Need (SGCN), and species that were “Not Listed.”


Rarity and climate sensitivity

Following species selection and occurrence data filtering, our final list of study species included 137 freshwater fish native to the conterminous United States (see Data Availability Statement for a link to externally archived files). These study species include approximately 18% of all native U.S. freshwater fishes (Mims et al., 2010) and between 31 and 66 species in each region (Figure 2). Mean RCS scores (±SD) were skewed toward higher values (RCSN  = 0.79 ± 0.16; RCSR 0.66 <  < 0.75; Table 2, Figures 3 and 4). Spearman's rank tests revealed that RCSN and RCSR were significantly correlated for all regions (0.54 < rho < 0.95; Table 3), and RCS scores were consistent across regions and scales for the most highly sensitive species (RCSN > 0.9; Figure 3). However, we observed increased variation in RCS scores by region and extent as intrinsic sensitivity decreased, particularly for species with occurrence records in multiple regions. Several species in the West had significantly higher RCSN than RCSR scores (Figure 3), indicating that they appear more sensitive at a national scale but less sensitive (more common) when limiting the study scope to the West only. In addition, mean intrinsic sensitivity was higher at the national extent than at the regional extent across all study species in the West (ANOVA, F = 12.61; p < 0.001; Table 2). In the other regions, RCSN and RCSR were not significantly different (Table 2, Figure 4).

TABLE 2. Rarity and climate sensitivity (RCS) scores vary across national and regional spatial extents.
F df p
East 0.75 ± 0.19 0.75 ± 0.20 0.0040 1, 130 0.95
Gulf 0.71 ± 0.20 0.71 ± 0.18 0.001 1, 60 0.97
Miss 0.72 ± 0.18 0.74 ± 0.17 0.027 1, 130 0.87
West 0.81 ± 0.09 0.66 ± 0.22 12.61 1, 66 <0.001
  • Note: National (RCSN) and regional (RCSR) scores, reported as mean ± SD, for each study extent. One-way ANOVA tests to compare variation between RCSN and RCSR scores for each region. Significant statistics are bolded.
Details are in the caption following the image
Distribution of national (RCSN, left side in gray) and regional (RCSR, right side in color as in Figure 2) rarity and climate sensitivity (RCS) scores for each regional species pool. Individual species scores are depicted as horizontal black lines. Higher scores indicate increasing intrinsic climate risk.
TABLE 3. Correlations between species traits and RCS scores illuminate possible links between traits-based and geographic measures of intrinsic climate sensitivity at different spatial scales.
Region and trait East Gulf Miss West RCSN Diet Habitat Body size Maturation Longevity Fecundity
East 0.64 −0.17 −0.14 −0.19 −0.08 −0.19 −0.06
Gulf 0.54 −0.26 −0.20 −0.29 0.00 −0.23 −0.27
Miss 0.79 −0.23 0.08 −0.36 −0.33 −0.46 −0.42
West 0.95 −0.15 −0.34 −0.43 −0.28 −0.25 −0.36
RCSN <0.01 <0.01 <0.01 <0.01 −0.16 −0.19 −0.47 −0.31 −0.43 −0.38
Diet 0.17 0.17 0.07 0.41 0.07 0.33 0.07 0.04 0.09 0.22
Habitat 0.26 0.28 0.50 0.05 0.03 <0.01 0.11 −0.03 −0.01 0.06
Body size 0.13 0.12 <0.01 0.01 <0.01 0.44 0.20 0.85 0.89 0.81
Maturation 0.54 0.98 <0.01 0.11 <0.01 0.62 0.78 <0.01 0.88 0.74
Longevity 0.13 0.24 <0.01 0.18 <0.01 0.33 0.95 <0.01 <0.01 0.82
Fecundity 0.64 0.16 <0.01 0.05 <0.01 0.01 0.48 <0.01 <0.01 <0.01
  • Note: Spearman rank correlation coefficients are presented in the top triangular matrix, with corresponding p-values in the bottom matrix. Differences in study species pools prevented analysis of correlations of RCSR scores among regions. Significant statistics appear in boldface.

RCS, species traits, and conservation status

Four study species (Anguilla rostrata, Cottus princeps, Cyprinella lepida, and Notropis chlorocephalus) with <50% trait data available were excluded from the PCoA. For the remaining 133 species, the first principal coordinate (PC) explained 57.8% of trait variation (Figure 5a) and primarily distinguished fish species based on the positively correlated life history traits of body size, longevity, and maturation age (Table 3; Appendix S1: Table S5). The second PC explained 31.2% of the variation and was driven primarily by habitat and diet breadth, which were also positively correlated. Fecundity was associated with both PCs and had strong positive correlations with all other traits except habitat breadth. National RCS scores were negatively correlated with all six traits investigated (Table 3); correlations were strong (p < 0.01) for all four life history traits and weaker but still meaningful for habitat breadth (p = 0.03) and diet breadth (p = 0.07). However, at a regional extent, significant negative correlations only emerged between traits and RCS scores in the Mississippi and West regions. For the Mississippi, results followed the national trend with strong negative correlations (p < 0.01) between all four life history traits and RCSR scores. In the West, negative correlations were identified between RCSR and body size (p < 0.01), fecundity (p = 0.05), and habitat breadth (p = 0.05). Taken together, we found support for hypothesized negative relationships between intrinsic sensitivity and the traits of longevity, diet breadth, and habitat breadth but found no support for the positive relationship predicted for maturation age (Table 1). Trait dispersion did not vary by spatial extent, but trait composition of study species in the West was significantly different from other regions (PERMANOVA, R2 = 0.032, p = 0.006; Table 4, Figure 5b–e).

Details are in the caption following the image
(A) Principal coordinates analysis (PCoA) of trait composition for all study species with trait eigenvectors scaled by percent variance explained. Point size in (A) increases with rarity and climate sensitivity scores so that the largest points correspond to species with the greatest intrinsic climate sensitivity. (B–E) PCoA of trait composition for each of the regional study species pools with an ellipse corresponding to a 95% CI. Colors in all panels refer to regions as in Figure 2.
TABLE 4. Multivariate analyses of study species trait composition and trait dispersion revealed few differences between national and regional extents.
R2 p F df p
East 0.008 0.31 0.080 1, 131 0.78
Gulf 0.008 0.37 0.002 1, 131 0.97
Miss 0.010 0.26 <0.001 1, 131 0.98
West 0.032 0.006 0.2 1, 131 0.66
  • Note: Permutational multivariate analysis of variance (PERMANOVA) and multivariate dispersion analysis (PERMDISP2) were used to evaluate differences in the centroids and dispersions of study species traits, respectively. Significant statistics appear in boldface.

At the national extent, we found that variability in RCS scores differed among conservation status (Kruskal–Wallis test, χ2 = 12.66, p = 0.005; Appendix S1: Figure S1). Post hoc Dunn's tests indicated significantly higher RCS scores for endangered species than state listed species, but we found no other differences among conservation status categories (Appendix S1: Table S6). At the regional extent, we identified eight species with high RCS scores (RCSR > 0.9) in at least one region with no conservation listing at any level (Figure 6). This includes Elassoma okefenokee and Morone chrysops in the East and Erimyzon tenuis, Lythrurus bellus, Nocomis effusus, Noturus albater, N. elegans, and N. leptacanthus in the Mississippi. For the West RCSR scores, the mean variation of scores within conservation listing categories was higher than that observed nationally (West RCSR: 0.60 <  < 0.80; RCSN: 0.79 <  < 0.87) (Figure 6D).

Details are in the caption following the image
National (RCSN, left side in gray) and regional (RCSR, right side in color as in Figure 2) rarity and climate sensitivity (RCS) scores for each conservation status. Species RCS scores are represented by laterally jittered points, with outliers duplicated in black. The box reflects the median and interquartile range (IQR), and whiskers extend to values no further than 1.5× IQR from the box; values beyond the whiskers are outliers.


Relative climate sensitivity is generally, but not always, conserved across spatial scales

Geographically derived climate sensitivity rankings were largely conserved across spatial scales for 137 species representative of freshwater fish diversity in the United States. Comparison of species-specific RCS scores across spatial extents revealed two distinct patterns. First, species with high RCS scores (i.e., the most sensitive species) were consistently identified at national and regional extents. These species often had occurrences limited to a single region, such as the darters Etheostoma fonticola and E. bellum. Species with the smallest geographic areas are expected to have high climate sensitivity (Estrada et al., 2015; Mims et al., 2018; Slatyer et al., 2013) and are good targets for additional investigation, such as assessments of range shifts in response to climate change (Carosi et al., 2019; Moyle et al., 2013; Radinger & Wolter, 2015).

We also found that despite general agreement among RCS values at different spatial extents, regional RCS scores were highly variable for some species that occurred in more than one region. For example, the silver chub (Macrhybopsis storeriana) is common throughout much of the lower Mississippi drainage but found only in a few drainages of the East and Gulf regions, resulting in a relatively low RCS score for the Mississippi region (RCSR = 0.59) and nationally (RCSN = 0.64) but high values in both the East (RCSR = 0.84) and Gulf (RCSR = 0.96). These departures from that general trend of conserved RCS scores are in line with other studies suggesting that spatial extents can influence vulnerability assessments (Fagan et al., 2005; Rinnan & Lawler, 2019). By conducting assessments at multiple spatial scales, managers can consider whether species identified as sensitive in their region are also sensitive at a larger scale as well. Such context may be important, particularly for partnerships among federal and state agencies that span spatial scales and may influence conservation decisions and allocation of resources for specific species (Olden et al., 2010).

Despite AOO and CS being scaled from 0 to 1, relative climate sensitivity was generally skewed toward high values (Table 2, Figure 3). The skewed pattern was driven by a few relatively widespread study species with much larger areas of occupancy and considerably more variation in climate sensitivity than species with relatively smaller ranges, which tended to dominate our list of study species. Although our study included only a subset of all freshwater fishes of the United States, our findings are also consistent with generally observed patterns of few common species and many rare ones across communities and assemblages (Siqueira et al., 2012).

We found little variation in mean RCS scores for the national, East, Gulf, and Mississippi regions, suggesting general consistency in relative sensitivity across these regions. This consistency may be due to shared pre-Pleistocene geological history (Mayden, 1988), similar environmental filters (e.g., Poff, 1997), and some shared study species in this study. Conversely, RCS scores for species in the West region were lower than scores in other regions or nationally. We attribute the lower RCS scores in the West to the largely unique subset of native fishes used in our analysis; of the 34 species we included from the West region, only the Mexican stoneroller (Campostoma ornatum), three-spined stickleback (Gasterosteus aculeatus), and the Gila trout (Oncorhynchus gilae) occurred in another region. For the 31 species with occurrences restricted to the West, geographic rarity was greater at the national extent (mean AOON = 0.97) than at the regional extent (mean AOOR = 0.75). Owing to the relative nature of our rarity and climate sensitivity index, rankings remain informative despite differences in regional species assemblages.

Correlations among geographic rarity and species traits vary by spatial scale

Nationally, we found that body size, maturation age, longevity, and fecundity were correlated with one another, following established work examining life history trade-offs among freshwater fishes (Lamouroux et al., 2002; Mims et al., 2010; Poff & Allan, 1995). The ecological traits of habitat breadth and diet breadth were also correlated with one another, consistent with previously observed patterns (Slatyer et al., 2013). Correlations between RCS and traits were strongest for the four life history traits examined, indicating that large-bodied, late-maturing, long-lived, and highly fecund species are the least likely to have small areas of occupancy and/or narrow climate niche breadths. Species exhibiting these traits generally correspond to a periodic strategist life history (Winemiller & Rose, 1992) that has evolved to take advantage of large, hydrologically connected rivers that provide predictable resource availability and environmental conditions (Winemiller, 2005). Our results suggest that for periodic strategist fishes, the potential costs associated with large body size (lack of access to small streams—Giam & Olden, 2018) and high fecundity (low egg survivorship—Winemiller & Rose, 1993) are superseded by the benefits of these traits when assessing intrinsic climate sensitivity. However, an important consideration is that realized climate vulnerability may be greater than our geographically derived measure of intrinsic sensitivity when climate exposure or anthropogenic impacts are considered. For example, we retained historical records and excluded occurrences outside of documented native ranges to eliminate potential bias from recent distributional changes (LeMoine et al., 2020), yet doing so results in measures of sensitivity that do not account for recent habitat fragmentation (such as dams; Grill et al., 2019). This limitation emphasizes the need for sensitivity-only assessments, like the RCS index, to be followed by detailed studies that can consider extrinsic factors that may lead to climate vulnerability.

Larger species tend to benefit from increased dispersal ability that allows them to track environmental conditions (Thurman et al., 2020), and increased lifelong reproductive output is advantageous, particularly given strong correlations between fecundity and longevity found in this study (Winemiller & Rose, 1993). For periodic strategists, the traits of large body size and high fecundity were associated with reduced climate sensitivity and greater geographic ranges at a national extent. Notably, traits that increase species-occupied area are consistently shown to reduce vulnerability across a wide range of taxa (Pearson et al., 2014). For freshwater fishes specifically, traits associated with increased range sizes have been used to predict both increased rarity (Pritt & Frimpong, 2010) and decreased extinction risk (Giam & Olden, 2018). Taken together, these findings suggest redundancy between geographically derived and traits-based measures of intrinsic climate sensitivity for freshwater fishes at a national extent.

In contrast, correlations among regional RCS scores and traits varied, indicating that these approaches may provide complementary information regarding intrinsic climate sensitivity. The Mississippi region most closely aligned with correlations observed at the national extent, which we attribute to three key factors: the Mississippi region was the largest in the present study, it represents a single hydrologically connected watershed, and along with the East, it had the highest study species richness. For body size and fecundity, hypothesized relationships with sensitivity are likely to vary with environmental context (Bears et al., 2009; Giam & Olden, 2018; Knouft & Page, 2003), and these traits may have reduced potential to explain broad patterns of sensitivity at regional to local scales. In the East, there are a number of independent river basins that flow from the Appalachian Mountains to the Atlantic coast, and the lack of hydrologic connectivity reduces the benefits of increased dispersal capacity. This may lead to scenarios where alternate mechanisms predict climate sensitivity, such as small-bodied fishes having greater access to thermal refugia in small streams (Knouft & Page, 2003). The lack of correlations observed in the East and Gulf suggests that the effects of traits on relative intrinsic sensitivity are not fully captured by geographically derived measures.

The West was the only region where trait composition differed from the national extent. We attribute this result to the traits of body size and to a lesser degree habitat breadth and fecundity, which were negatively correlated with regional RCS scores. We recognize that our use of a subset of study species provides an incomplete measure of trait diversity within regions and specifically note that some large-bodied fishes in the West region were excluded during species selection due to data deficiencies. However, freshwater fish traits are known to vary across the United States due to biogeographical constraints (Mims et al., 2010), and the West has some unique geologic features (endorheic basins and stream capture) that may explain the lack of study species that are long-lived, large-bodied, or late-maturing. Using a comparable sample size, Mims et al. (2018) found no relationship between traits (including body size, longevity, maturity, fecundity, and parental investment) and geographically derived intrinsic sensitivity for 54 freshwater fishes of the U.S. Pacific Northwest, an area that falls entirely within our West region. While traits associated with generalized life history strategies (Pianka, 1970; Winemiller & Rose, 1992) were correlated with geographically derived sensitivity at the national scale, this pattern did not hold true in the West and suggests that individual traits may complement geographic assessments of intrinsic climate risk at regional scales (Williams et al., 2010). Taken together, these results demonstrate the potential for relationships between geographic-derived and traits-based assessments of intrinsic sensitivity to vary across spatial extents and regions. Considering these approaches together may provide a more comprehensive and well-informed view of climate sensitivity, particularly when conducted across distinct biogeographic regions.

Intrinsic climate risk is not captured by conservation status for many species

Our findings suggest that sensitivity does not strongly differ among most conservation status categories and that intrinsic sensitivity to climate change may not be fully captured by current listing criteria (Delach et al., 2019). A mismatch between sensitivity to climate change and protected status has been documented for other freshwater taxa (fishes, Pritt & Frimpong, 2010; crayfish, Hossain et al., 2018), as well as for broader taxonomic groups (snakes, Reed & Shine, 2002; birds, Gardali et al., 2012; trees, Fremout et al., 2020; anurans, DuBose et al., 2023). This mismatch may result from conservation listing criteria that include many factors and are not limited to intrinsic sensitivity (Smith-Hicks & Morrison, 2021). Our subset of study species may not proportionally represent status listings for all freshwater fishes of the United States, as our use of data filters may exclude rare or hard to sample taxa with few occurrences in GBIF. For example, Jelks et al. (2008) documented that approximately 39% of freshwater fishes in North America were imperiled, whereas less than 12% of our study species were ESA-listed as threatened or endangered.

Furthermore, intrinsic climate sensitivity does not incorporate exposure to threats, such as climate change (Foden et al., 2013; Williams et al., 2008), and exposure is a leading driver of the listing process (Smith et al., 2018). The inclusion of climate exposure in making status determinations may help explain why some species with low measures of intrinsic sensitivity were state or federally listed (Butt et al., 2016; Summers et al., 2012). Nevertheless, systematic approaches that rank intrinsic sensitivity, such as ours, are valuable in identifying species with high relative sensitivity but no status at either the state or federal level. Examples include the elegant madtom (Noturus elegans, RCSN = 0.954) and redtail chub (Nocomis effusus, RCSN = 0.947), two species found in the Mississippi drainage that have small areas of occupancy. For at least some of the unlisted species, a lack of data or detailed understanding of the species biology may be leading to cases of vulnerable yet overlooked species (Bland et al., 2015). Some of these unlisted but highly sensitive species may not warrant state or federal listing owing to factors not assessed in this study (i.e., population abundance), but others are likely facing climate risks and have been overlooked in the listing process.

Conclusions and future directions

Relative comparisons of intrinsic sensitivity, particularly when leveraging publicly available and opportunistically collected data, offer a promising means of systematically assessing intrinsic climate risk across multiple species and scales. Yet occurrence data are imperfect, and the use of geographically derived measures of climate sensitivity may present challenges and trade-offs relative to other approaches. By comparing the RCS approach with species traits, we can better understand how these analyses are affected by spatial scale and support the identification of at-risk species. Our findings reveal several key insights into how spatial scale and study extent influence analyses and understanding of intrinsic sensitivity of freshwater fishes to climate change. Systematic assessments of relative intrinsic sensitivity consistently identified the most sensitive species across extents, yet some of these species lacked any conservation status listings. Correlations among geographically derived climate sensitivity rankings and species traits associated with climate sensitivity were variable across extents and regions, suggesting that geographically derived and traits-based approaches are redundant at a national extent but may be complementary at some regional extents. As the availability of species occurrence and trait data increases, so too does their ability to improve systematic, multispecies assessments of intrinsic sensitivity to climate change (Foden et al., 2019; Pacifici et al., 2017). By providing an understanding of how relative intrinsic sensitivity is influenced by spatial extents and regions, we better inform this assessment approach and support its continued use for considering species climate risk.


We thank the GBIF network for making data available for public acquisition and D. Noesgaard (GBIF) for assistance in creating a derived occurrence dataset. We also thank D. Wieferich (USGS) for the provision of updated hydrologic unit mapping, and T. DuBose for assistance with analyses. This study was funded in part by the U.S. Geological Survey's Science, Analytics, Synthesis, and Research Program (agreement numbers G17AC00235 and G19AC00388). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.


    The authors declare no conflicts of interest.


    Extracted climate data (Silknetter et al., 2023) are available from the U.S. Geological Survey's ScienceBase: Occurrence data (, 2021) are available from the Global Biodiversity Information Facility: Taxonomy, derived climate sensitivity measures, and trait/conservation status data for all 137 study species, as well as R scripts and associated input data for this work (Silknets, 2023), are available from Zenodo: