Integrating eDNA and citizen science observations to model distribution of a temperate freshwater turtle near its northern range limit

A primary goal of ecology and conservation biology is to determine the distributions of organisms and understand the factors that shape them. Reasons for why a species persists in some regions but is absent from others are complex as species distributions are influenced by multiple factors, including abiotic factors (e.g., precipitation, temperature), biotic interactions, and dispersal ability (Soberón & Peterson, 2005). Critical to understanding species distributions are comprehensive occurrence data, which can comprise presences and absences, or presences-only (MacKenzie et al., 2018). The traditional sources for presence data include museum records, dedicated surveys, and increasingly, online databases that contain citizen science observations (Newbold, 2010; Tiago, Pereira & Capinha, 2017). However, such online databases can be taxonomically biased, and skewed to data from regions with higher human populations or recreational activities (Titley, Snaddon & Turner, 2017), potentially leading to inaccurate descriptions of species ranges and habitat preferences (Araújo & Guisan, 2006).

Various methods have been used for assembling occurrence data, with their effectiveness depending particularly on species life histories (e.g., sociality, diurnality). Visual and auditory surveys are common sources for occurrence records (Gibb et al., 2019; Winship et al., 2020), as well as seining and trawling for fish (Porter, Rosenfeld & Parkinson, 2000), mist netting for birds and bats (Chandler et al., 2018; Scherrer, Christe & Guisan, 2019), and live trapping for mammals (Sofaer et al., 2019). New tools, including trail cameras, geolocators, satellite tracking, and environmental DNA (eDNA), are proving fruitful for deriving occurrence records (Coxen et al., 2017).

eDNA is increasingly featured in ecology, conservation biology (Beng & Corlett, 2020), and epidemiology (Ogden, 2021). eDNA surveys are now widely used to assess species occurrences, especially in aquatic systems (Rees et al., 2014; Seymour, 2019). Single species eDNA detection techniques such as quantitative PCR (qPCR) and digital PCR can supplement traditional survey data (Wineland et al., 2019). Such eDNA datasets have been used in invasive species monitoring (Sepulveda et al., 2020), species conservation (Takahara et al., 2020), quantification of biotic assemblages (West et al., 2020), and establishing range extensions (Gorički et al., 2017; Nardi et al., 2020). eDNA surveys may be more cost-effective than traditional methods, easier to replicate, and more sensitive for detection of rare or cryptic species (Muha et al., 2017; Olson, Briggler & Williams, 2012).

To infer range limits and habitat suitability for species, describe their ecological niches, and project their distribution over space and time, species distribution models (SDMs) are often used, including those generated using a Maximum Entropy algorithm (MaxEnt. Phillips, Anderson & Schapire, 2006). With the promise of overcoming some of the deficiencies of traditional survey methods, eDNA data have been recently brought into such SDMs (Carraro et al., 2018; Neto et al., 2020; Schmelzle & Kinziger, 2016), though such applications have yet to be robustly tested. Here, for a small freshwater North American turtle we ask whether targeted eDNA sampling changes our understanding of the environmental factors that underlie its distribution or extends its known geographical range.

The common musk turtle (Sternotherus odoratus) is an excellent candidate to test the value of eDNA surveys for SDMs. Ranging from Florida to Southern Ontario and Quebec, and west to Wisconsin and central Texas, it occupies shallow and slow-moving waterbodies with soft substrates (Dreslik & Phillips, 2005). While widespread, S. odoratus is difficult to observe because it is often crepuscular (Ernst, 1986). Sternotherus odoratus mostly basks just below the water surface often within patches of aquatic vegetation making visual detection from above the surface challenging. Overland movement of the turtle is uncommon, and its home ranges are probably confined to one waterbody (Ernst & Lovich, 2009), making incidental observations less likely. The known range of S. odoratus in Canada includes scattered locales across southern and central Ontario, and a small portion of western Quebec, with the current suspected northern range limit along the southern edge of the Canadian Shield (Fig. 1A; COSEWIC, 2012). Additional challenges near the northern range limit include a shorter active season (April to September) than at more southerly latitudes (e.g., year-around in Florida), and human presence that diminishes rapidly northward from the Canada-USA border leading to fewer overall citizen science observations. We thus suspect that the northern range limit of S. odoratus is underestimated and our understanding of its habitat usage is incomplete.

To address this deficit, we developed eDNA protocols to survey S. odoratus near and beyond its northern range limits and modelled its distribution with citizen science observations alone and with citizen science combined with eDNA occurrence data. We asked: (1) Can eDNA surveys reliably detect musk turtles? (2) How do additional eDNA occurrences change the outcomes of niche modelling? (3) Will eDNA data expand its range, especially towards its currently diagnosed range limit? (4) What environmental factors limit S. odoratus at its northern range boundary? To address these questions, we first examined the robustness and sensitivity of a new eDNA qPCR assay to detect S. odoratus in natural environments. We then modelled the distribution of S. odoratus using MaxEnt, with and without eDNA occurrences at two geographical scales, and compared the niche models built from different datasets. Finally, we evaluated all models to determine the underlying factors that may contribute to the northern range limit of S. odoratus.