Geographic biases in invasion science are common (Pyšek et al. 2008, Hulme et al. 2013, Lowry et al. 2013, Bellard and Jeschke 2016) and have the potential to skew our understanding of invasive plant impacts, efficacy of management, and native community susceptibility if these studies focus on a portion of the invaded range. Our results suggest that there is often a significant geographic bias in one or more study type, though it is more common for scientific studies to match manager records. The distribution of scientific studies is not as extensive as the distribution of manager records, but the geographies are similar (Fig. 1). Overall, though biases exist for individual species and study types, we do not find strong evidence of consistent or systemic geographic or climatic biases in scientific studies (Fig. 1, Fig. 3), suggesting that, as a whole, scientific studies tend to encompass the invaded range.
Distribution of records
With an order of magnitude more occurrence records, manager records typically described a larger invaded range than scientific studies (Table 1, Fig. 1). The comparison of literature and manager records reveal three types of patterns between these species which can be related to their total number of articles. The first pattern can be observed for A. altissima, F. japonica, and T. ramosissima, which have few articles in the lower 48; there are significant gaps in the literature records compared to manager records. The second pattern concerns A. petiolata, L. maackii, L. salicaria, M. vimineum, and P. arundinacea; literature records essentially encompass the same range as manager records and no large gaps exist between the two. Finally, for B. tectorum and P. australis, literature records have a larger range than manager records. This broader geography is due to genetic studies seeking to understand the introduction, spread, or hybridization of these species (Meyerson et al. 2016; Arnesen et al. 2017). However, genetic studies likely include locations where the species are naturalized, but not necessarily invasive (spreading or having impact). For example, B. tectorum is recorded throughout the lower 48, but is most problematic in western states (Knapp 1996; Bradley et al. 2018), which aligns with manager records. Because EDDMapS and iMap records tend to focus on areas of high priority for monitoring and management, it is likely that manager records provide an effective description of the invaded range.
Although the invaded range is encompassed by scientific studies in general, different study types are not as evenly distributed or pursued. Between each species, the relative proportion of each study type is variable, indicating that the research priorities for these species are not the same. These uneven prioritizations of study types affect the total number of study locations for each species, as certain study types, such as genetic studies (recorded under invasive trait), retrieve data from more sites than others, such as impact studies (Fig. 2B). As a whole, management and mapping studies, which relate to ways in which stakeholders can track and respond to plant invasions, consistently represent a low proportion of the total number of studies (Fig. 2A). Considering their more applied nature, questions related to management and mapping might be more extensively addressed in grey literature instead of scientific literature. However, even if management information is available in the grey literature, it will be harder to locate and therefore less useful for supporting effective management between counties or states. The large differences in research priorities makes it harder to identify comparable information for multiple species that would enable broader insights into prominent processes of invasion.
On average, literature records tend to occur in warmer environments than manager records, although, for most species, the temperature differences between datasets are small (< 1°C). With climate change, temperatures are likely to increase in the near term (Allen et al. 2018). When ecological studies varied climatically from manager records, they tended to occur towards the warm range margin (Fig. 3, Table S1). This focus on the warm range margin suggests that ecological studies could provide an effective illustration of the future ecology of invasions as temperatures increase. However, a focus on the warm range margin could also suggest that invasions are of greatest concern or highest impact in these areas. In this case, climate warming could make plant invasions worse than anticipated (Hellmann et al. 2008; Bradley et al. 2010; Mainka and Howard 2010) as more of the invaded range becomes climatically similar to the more problematic warm range margin. These changes could also produce greater management challenges because treatment strategies, notably herbicides, are not only affected by climate, but can be less efficient in warmer regions (Bailey 2004; Matzrafi et al. 2016; Ziska 2016).
Mean annual precipitation had up to 180 mm difference between literature and manager records but showed inconsistent directionality towards wetter or drier climates. Similarly, for all but one species, at least one study type occurred on a climate range margin, compared to manager records, but there is no consistent trend towards warmer/colder or wetter/drier climates across species (Fig. 3). This suggests that the focus on a more limited climate range is driven by other factors, such as impact, land use, or access, which varies between species. Two study types that never differed significantly in climatic space for any species were impact and management studies. The similarity between studies on management and studies on impact suggests that they focus on areas with the largest impacts that are also the most important to control. Impact studies were also most likely to be found at a climatic range margin (Fig. 3), which suggests that reported invasive plant impacts may not apply to the entire invaded range. The varied biases with respect to precipitation and inconsistent biases by study type show that invasion science often does not encompass the climate of the invaded range. This could lead to higher uncertainty in ecological forecasting of risk from future invasions because we lack information across climatic gradients encompassing the invaded range.
Disturbances and other biases
Invasive species are known to preferentially colonize disturbed areas, especially roadsides, which are both a habitat and a dispersal corridor for invasive plants (Christen and Matlack 2009; Menuz and Kettenring 2013). However, a major question in invasion ecology is whether invasive species are drivers of ecological impacts or passengers taking advantage of disturbance, but not the main drivers of impact (MacDougall and Turkington 2005). Our results show that both scientific studies and manager records are skewed to sites near roads, compared to random points within each plant’s distribution, though managers are more strongly skewed than scientific studies across all species (Fig. 4, Figure S1). While managers report invasions near roads, scientists focus on less disturbed areas, further from roads. Because invasion science tends to target sites that are farther from roads than many reported infestations, it is likely that studies are more representative of natural areas rather than highly disturbed areas. However, reports on one species, T. ramosissima, did not skew towards roadsides, this is likely linked to the species’ lifecycle: T. ramosissima spreads via water and is thus less likely to be found near roads than the other species (Zouhar 2003). This outlier may indicate that the reason managers and researchers focus on roadsides is because that is where these plants are mostly located, rather than a bias towards roadsides per se.
Scientific studies are, however, biased towards proximity to colleges and universities. This finding is not surprising given a desire for easy access to field sites and is consistent with past research showing high proportions of herbarium specimens collected near herbaria (Daru et al. 2018). Nonetheless, proximity to universities can lead to larger-scale biases because higher learning institutions are not evenly distributed throughout the country. This may be particularly problematic for species located in the western U.S., such as B. tectorum and T. ramosissima, where universities are less common. While our results do not suggest that this bias affects the geography of ecological research on invasive plants, the bias towards a more populated and university-dense eastern U.S. may contribute to the overrepresentation of these ten plants in invasion literature.
The large proportion of manager records on public lands in comparison to researchers is likely due to their focus on public land management. Nonetheless, public land is overrepresented in both datasets, representing only 7.8% of the total land area in the lower 48 (Jenkins et al. 2015) but 67% manager records and 50% literature records. Public lands tend to have more natural areas and accessibility to federal researchers, so this finding is consistent with larger trends in ecology that focus on natural areas (Martin et al. 2012). However, cities and suburban landscapes are experiencing faster warming with climate change and have different plant succession than surrounding environments (George et al. 2009). These landscapes are also more disturbed and thus prone to invasions, they could therefore provide insight into the behavior of invasive plants under future climates. Furthermore, if research is mostly focused on natural environments, we may not have an accurate portrait of invasion processes that affect most of the land area in the country. This bias could also affect predictive models, which often use manager records for calibration, by overrepresenting phenomena at natural sites.
Finally, the inclusion of a plant on a noxious weed list did not seem to impact the reporting of that plant. These lists are designed to regulate the sale and distribution of these plants and do not always include plants that are found in unmanaged areas (Quinn et al. 2013). It is therefore unsurprising that they do not affect where these plants are reported or studied.
The results from these two datasets highlight differences between where studies occur and where invasive plants are found, however, they are both imperfect records. Considering the volume of literature records analyzed, it was not possible to confirm site locations with more precision than was given in an article. This means that some sites could be up to 11 km (~ 0.1 decimal degrees) away from their recorded locations, this uncertainty is even greater for studies that present locations as broad, regional-scale maps. Nonetheless, 69% of study locations were recorded with greater than 0.1 decimal degree precision, so these errors are a small portion of the overall dataset (Table S3). We also find consistent trends with regard to distance to roads and to colleges and universities, the two most sensitive variables to these imprecisions, which suggests that our data reasonably capture where studies occur. Not all states record the presence or distribution of invasive species in accessible databases, which can create gaps in manager datasets. It is also possible that certain counties and states have priority species for monitoring and management that could lead to overreporting in certain areas. For example, A. petiolata has been reported throughout Wisconsin in EDDMapS, but is practically unreported in neighboring states. Ultimately this work reflects where researchers and managers report the presence of these plants rather than their potential or even actual range.