Contemporary climate change is rapidly affecting species’ ranges and distributions. While there is a general trend of poleward movement, there are exceptions. For example, despite the North American least weasel’s (Mustela nivalis) elusive nature, mammalogists in the mid-1900s noted the potential southward extension of its range into the central US, opposite the general poleward paradigm. Historically, a record of a species in a new location was sufficient for documenting a range expansion; yet, such observations can be biased by the extent of regional sampling. To investigate this unusual southward range shift, we use historical specimen records to statistically assess the absence of least weasels in the central United States before the 1960s. We include records of similarly trapped mammals as a measure of sampling effort, to better distinguish between perceived absence and geographic sampling bias. We then use ecological niche modelling to measure the association between changing climate and least weasel range dynamics. Our results provide evidence of a significant expansion at the southern periphery of their range, consistent with the 1960s timeline of hypothesised dispersal in historical field notes. Comparison of historical and contemporary niche models shows a significant increase in suitability at the south-western leading edge during that same time frame. Our findings underscore the importance of natural history specimen records for understanding species’ responses to climate change and provide methods for a more robust validation of suspected changes in range dynamics, particularly for rare or elusive species.

Climate change, ecological niche modelling, natural history museum, range expansion, sampling effort, target-group

Changing climate is causing global shifts in the geographic distributions of species (Walther et al. 2002; Parmesan and Yohe 2003; Benson and Cummins 2011). Shifts poleward or upward elevation, for example, have been documented for many taxa, including both endo- and ectotherms (Canning-Clode and Carlton 2017). Such shifts are consistent with the retention of an ancestral thermal niche, whereby a species may have a narrow window of thermal tolerance (Chen et al. 2011; Prost et al. 2013). On shorter timescales, it may be easier for populations to relocate than to adapt in situ (Donoghue 2008; Moritz and Agudo 2013). For example, Chen et al. (2011) reported an average distributional shift away from the Equator at a rate of 16.9 km per decade in hundreds of endotherms and ectotherms, coincident with the regional pace and direction of warming (Loarie et al. 2009). Exceptions to that trend are thus ecologically and behaviourally novel and warrant further investigation. Temperature is a common driving factor of distributional change; however, consideration of additional climatic variables and species-specific ecological needs (Crimmins et al. 2011; Keith et al. 2011) can reveal more nuance in the rate and direction of anticipated change (VanDerWal et al. 2012; Williams and Blois 2018).

Detection of a species in a place where it had previously been undocumented has historically been used as conclusive evidence of range expansion (Benedict et al. 2000; Albert et al. 2004; Kowalczyk et al. 2015; Barnes and Hoffman 2023). That framework, however, conflates non-detection with absence, which may lead to erroneous conclusions about species range dynamics (Gu and Swihart 2004; Mackenzie 2005). Especially in the case of rare or difficult to sample taxa, locations of perceived absence are more likely to represent areas of non-detection (MacKenzie et al. 2005; Belmont et al. 2022). For instance, consider a species that occurs at the base of a mountain and was not found when the mountain top was sampled one hundred years ago. Modern sampling of the mountain top later yields a record of the species: has the species expanded its range in the last one hundred years or was the species simply not observed during prior sampling? Occupancy modelling, species-accumulation and sampling-completeness analyses are powerful tools for studying distributional changes over time and space, but require repeated, high resolution sampling with a focused design (Chao et al. 2009; Costa et al. 2016; Henriksen et al. 2019). Anderson (2003), however, presented an approach for measuring the significance of observed absences by using occurrences from primary biodiversity data of a focal species and those of target-group background species (that is, species collected using methods similar to those that target the focal species). By measuring the relative likelihood of having found the focal species given the observed sampling effort (e.g. via the number of target-group background occurrences) in one location, it is possible to determine the significance of non-detection in another ecologically equivalent location. Where Anderson’s method originally measured the probability of true absence across space, we aim to measure absence across time. By considering historical and modern sampling effort in a particular location, we are able to modify the method to measure the significance of an observed absence between time periods, rather than locations. By validating a historical absence, we can more confidently identify a modern range expansion and, thus, the underlying ecological and environmental factors that may have facilitated a change (Anderson 2003).

Ecological niche models (ENMs) are a set of tools widely used to characterise species distributions, suitability landscapes and potential range shifts that may be due to environmental change (Peterson et al. 2011). ENMs can help contextualise range shifts by identifying environmental conditions in which a species can persist and predicting how changes in such conditions may alter a species’ geographic range (Peterson 2003; Lenoir and Svenning 2014). By assuming that a species’ ecological needs and limits are consistent before and after a range shift, we can estimate how regional environmental suitability for the species has changed after the shift at the edge of expansion (Sequeira et al. 2018). A positive change in suitability at former range limits may indicate that the environmental variables modelled are associated with the expansion. These ENM-based analyses are particularly useful for studying under-sampled species, as they are relatively robust to low input occurrences (particularly the Maxent model; Pearson et al. (2007); Wisz et al. (2008)). Additionally, modern survey and resurvey initiatives (e.g. Grinnell Resurvey Project; Moritz et al. (2008)) that contribute to the growth of primary biodiversity infrastructure, in the form of specimens in natural history museums and associated digital databases, will only increase the efficacy of these approaches moving forward.

One such under-sampled species is the North American least weasel (Mustela nivalis), the world’s smallest carnivore, which has a Holarctic distribution spanning Eurasia to North America (Abramov and Baryshnikov 2000). There are conflicting hypotheses surrounding the range dynamics of North American least weasels over the last century, one emerging from historical literature and the other from recent ENMs. The first hypothesis argues that least weasels may be expanding their range southwards, opposite to the general distributional trends for mammals (Ray et al. 2012; Prost et al. 2013; Roy-Dufresne et al. 2013; Williams and Blois 2018). In the early 1900s, Nebraska was the southern limit of the known range for least weasels (subspecies M. n. campestris; Swenk (1926); Allen (1933)). By 1955, Eugene Raymond Hall (E. R. Hall), renowned mustelid biologist and then Director of the Natural History Museum at the University of Kansas (Findley and Jones 1989), suspected that M. nivalis “almost certainly occurs in north-central Kansas” (Hall 1981). Still, Hall’s graduate student J. Knox Jones, another reputable mammalogist (Findley et al. 1996), declared the species absent from the western and southernmost counties of Nebraska (Jones 1964). The long history of North American mustelid research in the Great Plains (Hall 1951; Choate et al. 1979) makes their non-detection in Kansas at the time especially noteworthy. If caught, a least weasel would have almost certainly been collected and preserved. Despite decades of regular sampling across the region, the species was not documented in Kansas until 1965 (Jones and Cortner 1965). Since then, M. nivalis has become “...relatively abundant in northern Kansas and has dispersed southward” (Choate et al. 1979; Bailey and Terman 1986; Benedict et al. 2000; Barnes and Hoffman 2023). Opposite to these observations, a recent study by Cheeseman et al. (2024) found a 54% decrease in suitable habitat for least weasels within the contiguous United States during the 1900s. Additionally, they found evidence of substantial range fragmentation, specifically at range edges in Kansas and Nebraska. We aim to empirically assess the probability of true absence versus non-detection of least weasels in the Great Plains to resolve these contradictory reports.

Least weasels are notoriously difficult to sample (Easterla 1970; King and Powell 2006), resulting in major sampling gaps across large portions of their range (e.g. Canada and the northern Great Plains). For example, despite their Holarctic distribution, only 10% of preserved least weasel specimens recorded in the Global Biodiversity Information Facility (GBIF.org) database were collected in North America (Fig. 1; GBIF 2022a). Further, like many small mammals, least weasels are not reliably identified from photographs (Kays et al. 2022). These, coupled with the inherent biases of opportunistic and observation-based datasets like iNaturalist (Arazy and Malkinson 2021), reinforce the importance of continued specimen-based, expert-led, hands-on surveys and resurveys to document changing species distributions. Thanks to the historical and continued collection and preservation of mammalian specimens in Kansas (~ 50,000 specimens since the year 1900; GBIF (2024)), this is an ideal system to test for potential change in the geographic range of the otherwise elusive least weasel in North America.

Here, we assess the purported southward range expansion of North American least weasels in the mid-20^th century by accounting for sampling effort. We hypothesise that the southern expansion of least weasels described by historical experts in the field represents a true range extension. We further hypothesise that changes in climatic conditions are correlated with and may partially explain that expansion. We use primary biodiversity specimen records of least weasels and an index of regional sampling effort created with specimen records of similarly-sized and similarly-collected small mammals to test whether least weasels were absent from the central United States prior to the 1960s. We then use ENMs to test whether observed change in the geographic range of least weasels is correlated with changes in climatic conditions.

Figure 1. 

The top panel shows all preserved specimen records of least weasels (orange) and background Mustelidae and Rodentia (blue) taxa, downloaded from the Global Biodiversity Information Facility and collected within the geographic area simulated as accessible to least weasels. The bottom panel shows the workflow for testing significance of absence through time within discrete hexagon shaped cells. Additional detail provided in Suppl. material 1: fig. S1.

Specimen-backed, North American least weasel occurrences with coordinates were downloaded from GBIF (GBIF 2022a). An important element in demonstrating absence from a site is the sampling effort that could have yielded records of the species of interest (Anderson 2003). As the smallest species in the order Carnivora, least weasels are most frequently caught in Sherman live-traps or large (rat-sized) snap traps (Easterla 1970), the same traps used to survey rodents and other small mustelids (e.g. M. richardsonii). Thus, we used rodent and mustelid specimen records as an index of regional sampling effort of mammals of similar size and ecology. We downloaded specimen-backed occurrence records for all species in: (i) the weasel family Mustelidae (GBIF 2022b) and (ii) the order Rodentia (GBIF 2022c) to create an occurrence ‘target-group background’ dataset (Fig. 1; Anderson (2003)). This background dataset was used to test the significance of observed least weasel absence and for sampling bias correction in ENMs. Records lacking a collection year or those georeferenced to bodies of water were removed from analysis.

Range shifts are both an anticipated and documented outcome of contemporary climate change, corroborated by a strong correlation between rising temperatures and the poleward migration of species’ ranges (Lenoir and Svenning 2014). Some endotherms defy that general trend, however, suggesting potential resilience to changing temperatures and further investigation of those exceptions can refine our understanding of the factors affecting species’ distributions (VanDerWal et al. 2012). By leveraging historical, georeferenced museum specimen records as a proxy for historical sampling effort, we were able to rigorously test for a pattern of range expansion in least weasels in the Great Plains. We found significant support for post-1960s southward expansion of least weasels in the central Great Plains and, specifically, in Kansas (Fig. 2). Our results support expert, albeit anecdotal, accounts of least weasel range expansion into the central Great Plains over the last 65 years. By testing historical observations using modern methods, we validate the assertion that least weasels defy the poleward migratory trend observed in other mammals and are instead expanding southward. The observed increase in climatic suitability, in areas that tested significant for range expansion, adds an additional layer of confidence (Fig. 3) and allows us to better identify the specific environmental variables correlated with that trend (i.e. lower maximum VPD and higher precipitation). Trajectories of environmental change for total precipitation and VPD could facilitate further expansion into regions currently identified as ‘unsuitable’, such as the mountainous portions of the US (Ning et al. 2015).

Our cross-time tests of the significance of absence compliment historical observations of least weasels appearing in Kansas after the 1960s. While our method of evaluating absence is more rigorous than concluding from non-detection, it is performed under a few key assumptions related to detectability and observed prevalence. Here, we refer to detectability as the ability of an individual species to be sampled, assuming they are at the location being sampled (i.e. probability of detection given presence). We refer to observed prevalence as a function of a species’ detectability and local abundance and true prevalence only as a function of abundance; both given that the area being sampled is suitable (i.e. probability of sampling an individual at a location within their suitable range) (Anderson 2023). In hexagons with confirmed post-1960s least weasel presences, we assume that least weasels were also present pre-1960s (i.e. the hexagon was suitable), with a probability of sampling determined by the ratio of post-1960s least weasels to post-1960s target-group background samples (i.e. observed prevalence). That is, the suitability of the hexagon, prevalence and detectability are consistent between time periods. Statistically significant results could, therefore, allude to a change in any combination of these variables. Substantial unforeseen differences in detectability between time periods (assuming constant presence) could lead to a type I error, where least weasels were present in Kansas pre-1960s, but exceptionally less detectable than post-1960s. Given the consistently inconspicuous nature of least weasels, further investigation into the differences between habitat-level variables (e.g. dominant vegetation, predator/prey density etc.) in each time period could provide valuable insight into the importance of detectability in our case study.

If true prevalence were changing over time (i.e. detectability removed from the equation), it remains possible that hexagons statistically significant for range expansion in Kansas were suitable for least weasels pre-1960s, but, due to exceptionally low abundance (but not absence), still went unsampled, despite intense local sampling effort. Indeed, our binarised models (Fig. 2) show a third of Kansas being climatically suitable during both time periods, not just post-1960s. Beyond the general increase in suitability described below, why exactly the 1960s would facilitate such rapid population growth as to go from statistically absent to noticeably abundant within a decade is unclear. Tertiary, anecdotal accounts by fur trappers have described previous explosive population events of least weasels in the Dakotas during the late 1800s (Hall 1951); further investigations might aim to determine the ecological merit behind this emerging pattern.

If local suitability were changing between time periods, then ENMs would show increases in suitability values and the number of suitable pixels between time periods. Indeed, our niche models show significant improvement in least weasel suitability across the majority of the accessible US, particularly at the southern limit of their range where expansion was detected. While temperature is often an influential abiotic factor contributing to range shifts, other environmental variables like precipitation and VPD also affect suitability (Crimmins et al. 2011; Keith et al. 2011; Pinsky et al. 2013). For example, Tingley et al. (2009) demonstrated that of the 53 bird species native to the Sierra Nevada Mountains in California, 40% had shifted their ranges in response to change in precipitation more so than in response to temperature and an additional 26% tracked a combination of temperature and precipitation. Our least weasel ENMs highlight an association between high suitability, decreasing maximum VPD and increasing precipitation. That is, suitability is linked with increased water availability in the environment. Whether VPD and precipitation directly affect least weasel fitness or physiology (e.g. transpiration rate, water availability etc.) or secondary aspects of their life history (e.g. prey prefers wetter conditions) remains unclear. We expect that the diminutive size of least weasels allows them to take advantage of rodent burrows and similar microhabitats, providing thermally stable spaces that may behaviourally buffer them from warming ambient temperatures (Hall 1951; Hayward 1964; Riddell et al. 2021).

While our results are consistent with expert historical observations, they contradict recent findings by Cheeseman et al. (2024). Cheeseman and colleagues also built ENMs using climatic variables to predict change in suitable area during the 1900s for North American weasels. Their models incorporated land-use variables and did not restrict simulations to accessible areas. Land use contributed significantly to the fit of their models, which resulted in a substantial 54% decrease in suitable area for least weasels in North America. The history of land use change in the Great Plains is complex and particularly so through the 1900s. The Homestead Act of 1862 led to widespread, rapid conversion of native prairie to agricultural land (Williams Shanks 2005), while the Conservation Reserve Program (Food Security Act, US Congress 1985) later reconverted > 8% of agricultural land back into grass and shrublands (and oftentimes back-and-forth again; Drummond et al. (2012)). Additionally, the vegetation forming agricultural land fluctuates (e.g. crop to crop or crop to feed grass) as the economy dictates which use of the land is most valuable (Sohl et al. 2016). As such, we expect that the inclusion of land use would contribute more noise than signal to our models. However, there may be additional variables contributing to least weasel range dynamics than climate alone. Future methodological investigations could elucidate the discrepancy between the two models by combining land-use variables with our historical framework. Better yet would be thorough, modern mammalian community surveys concentrated in regions identified as newly unsuitable under one set of variables and suitable under the other.

Key to our methods are the georeferenced historical and contemporary museum specimen records that comprise our background data. Concentrated sampling in Kansas led by in-state natural history museums over much of the 20^th century made our sampling effort analysis statistically sound, as evidenced by the concentration of hexagons with adequate sampling in the region (Fig. 2). Modern resurveys of historically sampled sites then gave us a baseline to conduct cross-time tests within these areas. To measure ongoing change in the least weasel’s geographic range, including outside of Kansas, more hands-on physical specimen records are needed, as the species is not readily identified via photos. Our results underscore the importance of natural history collections (i.e. physical specimens and associated data) for scientific exploration of the biosphere and, specifically, for measuring change through time. Expanded resurveys of historically sampled sites, in addition to sampling novel areas, will be critical to documenting range shifts empirically over time (e.g. Moritz et al. (2008); Rubidge et al. (2011); Smith et al. (2013); Tingley and Beissinger (2013)).

Occupancy models are another powerful tool for studying changes in distributions over time or space, as they estimate the probability of species presence while accounting for imperfect detection and can link these changes to climatic factors (Bailey et al. 2014; Devarajan et al. 2020). Applying occupancy models at large scales, however, can be challenging due to the need for extensive spatial and temporal data derived from repeated surveys for robust detection estimates (Bailey et al. 2014; Altwegg and Nichols 2019). Species accumulation estimates and sampling completeness analysis can help address these challenges by identifying gaps in survey efforts and quantifying how well the observed data represent the true species pool (Chao and Jost 2012; Chao et al. 2020). Similar to occupancy models, however, those analyses also rely heavily on good sampling design and ad hoc or opportunistic data can undermine their validity (Chao et al. 2009; Costa et al. 2016; Henriksen et al. 2019). The weighted locality approach that we employed, using target-group data, is a viable alternative that can provide valuable insights into species range dynamics. By leveraging background sampling intensity and implementing thresholds for inclusion, this approach accommodates areas with limited or sporadic sampling effort and reduces bias from incomplete data. This benefit is particularly useful when high-resolution survey data are unavailable, as it allows researchers to detect significant geographic patterns and temporal changes in species distributions, even with sparse datasets.

Least weasels have been in North America since the Wisconsin glaciation (Kurtén and Anderson 1980), but our results indicate they have only recently expanded further south and west into the central US. Demonstrating that this species was likely absent in Kansas prior to the mid-20^th century allowed us to better identify environmental factors driving this range expansion. The methods we employed are ideal for inferring perceived absence in rare or difficult to sample taxa. This includes species that are at-risk or endangered, particularly those distributed in otherwise well sampled geographic areas such as Franklin’s ground squirrel (Poliocitellus franklinii, Martin et al. (2003)), Henslow’s Sparrow (Centronyx henslowii, Young et al. (2019)), or white-tailed jackrabbits (Lepus townsendii, Brown et al. (2019)), amongst others. While we can never know the true past distribution of a species with certainty, by providing statistical support to perceived absences we can reduce assumptions and better inform conservation and management efforts.

We thank 83 natural history collections for making their specimen data digital and openly available through GBIF. A complete list of contributing collections is included in Supplementary Acknowledgements. We thank Bob Timm, Caroline Kisielinski, Morgan S. Lowry and three anonymous reviewers for helpful comments and feedback on this manuscript. This study was partially supported by an NSF grant awarded to JPC (DBI-2100955), a Chickadee Checkoff Grant from the Kansas Department of Wildlife and Parks (KDWP), University of Kansas Research GO grant and an NSF award to ATP (OIA-1920946).

PDC, BJW and JPC conceptualised this study and created the visuals. PDC, BJW and MEC performed the formal analyses. All authors contributed to the methodology, writing and revisions of the manuscript.

All code used to perform the analyses can be found at: https://github.com/PeterCampbell8/SouthwardNivalis or https://doi.org/10.5281/zenodo.15041889.