The phenology of environmental DNA detection for coastal aquatic invasive species

Environmental DNA (eDNA) has transformed aquatic invasive species (AIS) surveillance, yet the temporal ecology of eDNA signals remains poorly understood. Here, we introduce the concept of “eDNA phenology” to describe predictable seasonal variation in eDNA detectability arising from interactions among species life history, environmental conditions, and the ecological processes governing DNA production, transport, and persistence. We investigated weekly eDNA dynamics (2, 663 samples) of five coastal AIS (Botrylloides violaceus, Didemnum vexillum, Ciona intestinalis, Membranipora membranacea, and Carcinus maenas) across five locations spanning a broad latitudinal gradient in the Northwest Atlantic using year-round, species-specific qPCR monitoring. Despite substantial ecological differences among taxa, species exhibited coherent regional phenological patterns, with predictable latitudinal shifts in seasonal eDNA abundance. Seasonal and regional effects explained most variation in eDNA dynamics, whereas instantaneous environmental variables provided comparatively modest additional explanatory power. Comparisons with conventional settlement monitoring further showed that eDNA signals frequently preceded or extended beyond observable settlement periods, suggesting that eDNA captures ecological dynamics not fully resolved by organismal observations alone. Together, these findings highlight that environmental and temporal variability should be considered when interpreting relationships between eDNA concentration and species abundance. Quantifying eDNA phenology provides a predictive framework for interpreting species ecology beyond simple detection and for improving the design, sensitivity, and forecasting capacity of biodiversity surveillance.

Aquatic invasive species (AIS) are non-native species introduced outside their natural distribution range, either intentionally or accidentally through human-mediated activities, that can cause harmful impacts to ecosystems, the economy, and society. AIS are among the leading drivers of global biodiversity loss and can cause severe, often irreversible ecological and economic impacts (WWF, 2024). In the absence of natural predators, AIS can rapidly outcompete and prey on native species, resulting in the disruption of coastal ecosystem structure and function (Rilov et al., 2024). Beyond ecological harm, the natural and human-meditated spread of AIS poses substantial economic costs comparable to natural disasters (Turbelin et al., 2023). For example, the expansion and establishment of benthic marine AIS, such as the European green crab (Carcinus maenas) and biofouling species (e.g., solitary tunicate Ciona intestinalis, colonial tunicate Botrylloides violaceus, bryozoan Membranipora membranacea), have negatively affected aquaculture operations and productivity, caused infrastructure damage, and increased management expenditures (Daigle and Herbinger, 2009; Fletcher et al., 2013; Sievers et al., 2013; Sephton et al., 2015). These widespread impacts highlight the urgent need for effective preventative measures against future anthropogenic introductions. Preventing introductions remains the most effective strategy for limiting AIS impacts; however, once establishment occurs, rapid detection and response become critical for constraining spread and minimizing ecological and economic damage (Beric and MacIsaac, 2015; Reaser et al., 2020).

Environmental DNA (eDNA) refers to genetic material shed by organisms into their surroundings (Rourke et al., 2022). As a monitoring tool, eDNA enables species detection without direct observation or capture, offering a powerful complement to traditional AIS biomonitoring approaches. This method is advantageous as it overcomes many field logistical constraints, enables broad spatial coverage, provides high detection sensitivity, and is non-invasive (LeBlanc et al., 2020). Like any monitoring tool, however, eDNA has limitations. Currently, it cannot reliably differentiate all species, distinguish between living and deceased organisms, resolve life stages (e.g., planktonic larvae vs. benthic adults), and it has limited capacity to estimate abundance (Iacaruso et al., 2025). Nonetheless, eDNA surveys have proven highly effective for AIS surveillance, particularly for benthic species that are difficult to detect using conventional methods such as scuba surveys or settlement plates (Lacoursière-Roussel et al., 2012; Rees et al., 2014; Branstrator et al., 2017; Simard et al., 2017; Zaiko et al., 2018). Unlike settlement plates, which rely on organisms actively settling on surfaces, eDNA offers a powerful approach for monitoring benthic organisms that may be in low abundance or fail to settle on plates (e.g., crabs, pancake batter tunicate [ Didemnum vexillum; Sephton and Vercaemer, 2015], Cangelosi et al., 2024). eDNA techniques have now been implemented in routine AIS monitoring programs to complement and enhance AIS detection efforts, improving sampling frequency and spatial coverage to support AIS regulations (Abbott et al., 2021; Chevrinais et al., 2023b; DFO, 2020; Laschever et al., 2023; LeBlanc et al., 2020; Lee et al., 2024).

Seasonal variation in AIS eDNA detection has been increasingly documented (Aylagas et al., 2024; Melliti et al., 2025; Sevellec et al., 2025; Morrison et al., 2026), highlighting the influence of environmental and biological factors on detection rates. Consequently, seasonal dynamics must be considered when integrating eDNA into long-term biosecurity monitoring frameworks to maximize detection probability and improve management outcomes. Despite growing evidence that eDNA detectability varies seasonally, no general framework currently exists to interpret or predict these temporal dynamics across species and regions, limiting the transferability and optimization of eDNA-based monitoring programs.

Species life-history processes, including reproduction, larval dispersal, recruitment, and molting, have been associated with seasonal variation in eDNA detectability; however, the extent to which these patterns generalize across species and regions remains unclear. For instance, LeBlanc et al. (2020) observed greater eDNA detection for tunicates and bryozoans in the northwest Atlantic in late summer, likely due to peak reproductive output. Similarly, zebra mussels (Dreissena polymorpha) in Minnesota lakes also demonstrated peak eDNA concentrations in mid-summer, potentially related to larval release (Lacoursiere-Roussel and Deiner, 2021; Rounds et al., 2024). Spatiotemporal variability in eDNA detection rates is also attributable to biophysical factors, such as hydrodynamics and environmental conditions, which can influence species-specific life histories as well as rates of eDNA shedding (e.g., gamete or larval release), dispersal, and persistence (Rourke et al., 2022). In coastal regions, for example, environmental variables including tides, salinity, and temperature stratification interact to shape eDNA-derived community composition (Sevellec et al., 2024, 2025).

Building on the concept of “eDNA ecology” proposed by Barnes and Turner (2016), we introduce the concept of “eDNA phenology”. Analogous to classical phenology, which examines the timing of seasonal biological events such as reproduction, recruitment, and migration (Park and Post, 2022), eDNA phenology refers to predictable seasonal variation in species-specific eDNA detectability driven by interactions among organismal biology, environmental conditions, and ecological processes governing DNA production, transport, persistence, and degradation. These patterns may reflect periods of increased biomass, metabolism, reproductive activity, larval release, molting, or other processes associated with elevated eDNA production and persistence. Consequently, seasonal eDNA dynamics may represent emergent ecological signals that integrate organismal phenology with the environmental processes governing DNA fate in aquatic ecosystems. Under this framework, eDNA signals are interpreted not solely as indicators of species presence, but as temporally structured ecological signals reflecting both organismal activity and environmental context. Understanding eDNA phenology could provide an adaptable framework for predicting optimal sampling windows and improving the sensitivity and consistency of AIS surveillance programs across regions and taxa. Optimizing eDNA sampling strategies is therefore critical, particularly given the risks of false-negative and false-positive detections. False negatives may delay management responses to newly established populations, whereas false positives can trigger unnecessary interventions and misallocation of resources. Identifying periods of elevated eDNA concentration may therefore improve detection reliability, sensitivity, and the cost-effectiveness of AIS monitoring programs.

This study investigates eDNA phenology to improve eDNA-based surveillance of coastal AIS across regions and taxa. Using year-round, species-specific qPCR monitoring along a broad latitudinal gradient in the Northwest Atlantic, we characterize seasonal eDNA dynamics and test whether AIS exhibit coherent regional phenological patterns despite differences in taxonomy and life history. The selected AIS represent diverse benthic taxa, including colonial tunicates (B. violaceus, D. vexillum), a solitary tunicate (C. intestinalis), a bryozoan (M. membranacea), and a benthic arthropod (C. maenas). We further examine how environmental conditions influence seasonal eDNA dynamics and evaluate whether eDNA signals correspond with conventional settlement plate observations for B. violaceus and C. intestinalis. By integrating seasonal timing, environmental variability, and conventional monitoring observations, we demonstrate how eDNA phenology can be established as a framework for interpreting temporal eDNA dynamics and improving the design, sensitivity, and predictive capacity of AIS surveillance programs.

Study site

Seawater samples were collected periodically in five coastal sites in three ecoregions along the Northwest Atlantic with known populations of biofouling AIS: the Gulf of St. Lawrence (Prince Edward Island, Canada [PEI] and Îles-de-la-Madeleine, Québec, Canada [Magdalen Islands; MAG]), Scotian Shelf (Halifax Harbour, Nova Scotia, Canada [HAL]), and the Bay of Fundy (BOF) and Gulf of Maine (GOM; Figure 1).

The Gulf of St. Lawrence is a semi-enclosed embayment roughly 250, 000 km 2 that connects the Atlantic Ocean to the St. Lawrence River, serving as a transition zone between freshwater and saltwater. It is influenced by a mix of river inflows and tidal currents, creating a dynamic system with strong seasonal variations in temperature, salinity, and circulation. In winter, cold, dense waters dominate, while in summer, fresh water from the St. Lawrence River forms a relatively warm surface layer over a cold intermediate layer (CIL), which covers the deeper, saltier ocean water below. Ice cover is also common in winter, affecting navigation and marine ecosystems (Galbraith et al., 2024). In recent years, however, warmer winters have led to weaker CILs and less sea ice cover across the Gulf of St. Lawrence (Galbraith et al., 2023). Sampling was conducted in PEI in Souris Harbour, on the northeastern coast of the island, and the MAG sampling site was within the Cap-aux-Meules Harbour, located near the centre of the archipelago and on the eastern shore. A main ferry terminal connects the Souris and Cap-aux-Meules Harbours (Figure 1).

For the Scotian Shelf, sampling occurred in a small boat marina in Halifax (HAL) within the Bedford Basin, a 17 km 2 semi-enclosed embayment at the innermost end of the Halifax inlet system on the southeastern coast of Nova Scotia (NS). The basin is estuarine, with a maximum depth of 71 m and freshwater inflow that creates distinct seasonal patterns in temperature and salinity (Li and Dickie, 2001).

For the BOF, sampling occurred at one site within Passamaquoddy Bay, a semi-enclosed sub-basin of the BOF, which is straddled by New Brunswick (Canada) and Maine (United States). Although the inner basin of Passamaquoddy Bay is partially protected from the BOF by a series of islands, extreme tides (~ 8 m) and freshwater input in the spring generate strong currents and significant upwelling of deep, cold water from the BOF (English et al., 2024).

The GOM is a semi-enclosed embayment spanning 93, 240 km 2 with freshwater river inputs (Brickman et al., 2021) that experience powerful tides (along with the BOF), which mix the influx of North Atlantic waters. For the GOM, the two stations sampled in southern Deer Isle (located in the midcoast Penobscot Bay) included a recreational area and a ferry dock in a busy lobster port.

Five benthic AIS were selected as they are high-priority invasive species in the Northwest Atlantic. The vase tunicate (C. intestinalis), violet tunicate (B. violaceus), and coffin-box bryozoan (M. membranacea) are established in all sampling regions (DFO, 2023a). Pancake batter tunicate (D. vexillum) is established south of GOM (Massachusetts, USA), although it has been first observed in the BOF in 2013 (see Moore et al., 2014; Vercaemer et al., 2015), and has not been detected in HAL. Finally, green crab (C. maenas) is established along the GOM, in the BOF, along the southeast coast of NS, and throughout the south-eastern Gulf of Saint Lawrence including PEI, MAG and the southern coast of Newfoundland (DFO, 2023b).

eDNA sampling

Sampling was conducted at 14 stations distributed across the five study regions (i.e., sites; Figure 1). At each site in all regions except GOM, three water samples (1.6–2 L) were collected weekly from 1.5 m below the surface using an eDNA Citizen Scientist Sampler (Smith-Root) and clean tubing attached to a Smith-Root Filter Pack containing a Whatman 1.2 μm glass fibre filter (BOF: May 2023 - May 2024; HAL and PEI: April 2023 - March 2024; MAG: May 2023 - February 2024). At GOM sites (GOM: January-December 2023), a sterile magnetic cup with vacuum filtration was used for filtration instead of the Smith-Root sampler, but otherwise collection methods were identical to other sites. A field negative control (1.6–2 L distilled water) was filtered weekly alongside the 3 field samples. Sampling stations were within 50 m for each site except for GOM, where the two stations were 2.5 km apart. Filters were stored in 2 mL tubes and preserved with 200 μL non-denatured 95% ethanol, except for MAG samples which were preserved in silica beads. Preserved filters were transported from the field on ice, except for MAG which were stored at room temperature (max. 1 month), and stored at -20 °C in the lab until DNA extraction.

Cross-contamination risk was minimized by soaking sampling materials (filter packs, tweezers, tubing, and weights) in a 1% bleach solution (1:9 parts 10% bleach to tap water) for a minimum of one hour, rinsed three times in tap water, and air dried before each site visit. Gloves were worn between each station and replaced between samples and controls.

DNA extraction and qPCR

Lab processing of the PEI, HAL, BOF and GOM samples was conducted at the Gulf Fisheries Centre - Aquatic Animal Health Lab (GFC-AAHL, Moncton, NB), while the MAG samples were processed at l’Institut Maurice-Lamontagne (IML, Mont-Joli, Qc). Half of each filter was extracted using the Macherey-Nagel ® NucleoSpin Tissue kit following Harris et al. (2023), except for the MAG samples which were extracted using the QIAGEN DNeasy ® Blood and Tissue Kit following Chevrinais et al. (2023a). Primer and probe sequences for the C. intestinalis, B. violaceus, M. membranacea and D. vexillum assays were from LeBlanc et al. (2020), while those for the C. maenas assay were from Chevrinais et al. (2023b). DNA was amplified in triplicate reactions using QuantStudio™ 5 real-time PCR (qPCR) instruments (Thermo Fisher Scientific, or for MLI a QuantStudio™ 3). Each 25 µL (20 µL for MAG) qPCR reaction consisted of 3 µL of template DNA, 0.48 µM of target-specific forward and reverse primer, 0.2 µM of target-specific FAM-labelled TaqMan™ fluorescent probe, 0.04% BSA, and 1 X of TaqPath™ ProAmp™ Master Mix (Thermo Fisher Scientific). The reaction conditions for all species were as follows: initial hold for 10 min at 95 °C, followed by 50 cycles of denaturation at 95 °C for 30 s (15 s for MAG) and annealing/extension at 60 °C for 60 s. Minor differences in qPCR reactions between GFC-AAHL and IML (i.e., total reaction volumes of 25 µL compared to 20 µL, and denaturation cycling steps of 30 s versus 15 s) are attributable to laboratory-specific preferences during assay validation. These minor discrepancies do not affect assay performance as demonstrated below with similar performance characteristics (i.e, LOD and amplification efficiencies) observed between laboratories. Possible qPCR inhibition was assessed by spiking samples with an exogenous internal positive control (IPC) at the DNA extraction lysis step (T4 bacteriophage) or at the PCR step (for MAG, plasmid DNA according to Dumoulin et al., 2024). Inhibition was considered present if the quantification cycle (Cq) value was more than two cycles higher than the no template control Cq value. The IPC qPCR assay was performed using the same reagents and amplification parameters as for targeted assays.

Sample eDNA concentration (copies of DNA per litre of filtered water; copies/L) were calculated from qPCR quantification cycle (Cq) values using the equation: X 0

where X 0 is the estimated eDNA concentration (copies/L), m (slope) and b (intercept) were derived from standard curves generated from serial five-fold dilutions of synthetic gBlock gene fragments (Integrated DNA Technologies) or plasmid DNA quantified by digital PCR (3901 bp, GENEWIZ, Azenta Life Sciences; quantified by digital PCR, Absolute Q™, Thermo Fisher Scientific), and FV is the volume of water filtered (L). Each dilution from 93, 750 to 0.048 copies/rx were analyzed using 9 technical replicates for dilutions ≥150 copies/reaction and 18 technical replicates for dilutions ≤30 copies/reaction, along with 18 no-template controls (NTCs) (for IML: from 6250 to 0.08 copies/rx, 6 technical replicates for dilutions ≥10 copies/rx and 16 technical replicates for dilutions ≤2 copies/rx, along with 16 NTCs). A linear regression of Cq against log 10 copy number was used to estimate assay slope, intercept, and amplification efficiency. The factor of 3 represents the volume of DNA (µL) per qPCR reaction, 100 µL is the extraction elution volume per filter half, and the factor of 2 accounts for both filter halves. eDNA concentrations were calculated separately for each qPCR technical replicate, using the corresponding instrument-specific standard curve formula, and replicate values were averaged to obtain a mean concentration for each sample.

Assay performance was determined for the five assays according to Lesperance et al. (2021), with limits of detection (LOD) ranging from 0.5 to 1.4 copies per reaction (0.6 to 1.0 copies per reaction at IML) and amplification efficiencies ranging from 95.6 to 97.7% (95.9 to 104.8% at IML).

Conventional monitoring with settlement plates

Settlement plates, also known as recruitment or settlement plates, were deployed in nearshore, sheltered locations with reduced wave exposure to promote natural recruitment, as ascidian survival is generally higher in protected environments such as embayments and harbors (Osman and Whitlatch, 1999; Davis and Davis, 2007; Howes et al., 2007). PVC plates, sanded on the downward-facing side due to the light sensitivity of tunicate larvae, were used as the settlement substrate because they are a preferred artificial surface for invasive tunicates in the study region (Darbyson et al., 2009; Ma, 2012 MSc thesis).

Species density was evaluated by deploying settlement plates in each of the three stations except for GOM (i.e., BOF: June 2023 - May 2024; HAL: June 2023 - March 2024; PEI: May 2023 - March 2024; MAG: June-October 2023; Supplementary Tables). Settlement plates were attached to floating docks (effectively subtidal) at a depth of 1 m using a weighted rope, and were preferentially placed in marinas to allow natural settlement of biofouling AIS. Settlement plates consisted of three 10 x 10 cm PVC recruitment plates spaced 10 cm from each other. BOF and HAL settlement plates consisted of 10 lines deployed at a depth of 1 m while PEI and MAG had three lines. At PEI (Souris Marina), floating docks are only installed in late May and taken out in mid-September, thus placing more plates at that site would not have widened the geographic range. The number of plates at PEI and MAG was reduced and restricted to the commercial dock to ensure the lines would not be disturbed or go missing during the duration of this study.

Species abundance was estimated using the standardized DFO AIS biofouling monitoring methodology, in which AIS presence on settlement plates is assigned to semi-quantitative percent cover categories (0 = absent; 1 = > 0 to< 25%; 2 = > 25 to< 50%; 3 = > 50 to< 75%; 4 = > 75 to 100%). During plate retrieval, photographs were taken of the downward-facing surface of each settlement plate. Percent cover estimates considered only the outermost settlement layer when underlying organisms were completely obscured (e.g., colonial tunicates fully covering barnacles). However, partially covered organisms were also estimated when visible, as underlying growth could remain extensive (e.g., tunicates partially covered by mussels). All non-indigenous species potentially present on plates were identified independently by two experts (C. DiBacco and B. Lowen, DFO). Species assessments included taxonomic identification, approximate age class (particularly recent recruits), reproductive condition when visually detectable, and evidence of senescence or mortality. Recruitment timing and reproductive windows were further inferred through successive weekly photo assessments. When additional identification was required, either entire plates or individual specimens removed from plates were retained in seawater-filled containers for further examination.

Environmental data collection

Environmental conditions were obtained to test the effect of the biophysical factors on AIS eDNA concentration. Water temperature and salinity were monitored at BOF, PEI, and MAG during sample collection using a handheld YSI ® ProDSS Multiparameter Digital Water Quality Meter (Xylem). Conductivity, temperature, depth profiles were conducted with a Seabird SBE 49 Fastcat profiler (Sea-Bird Scientific) at the HAL site during each weekly sampling event to assess temperature, salinity, and pH. Environmental variables were not collected for the GOM. On December 18, 2023 all sites experienced hurricane-force winds of approximately 100 km/h, and all but PEI had heavy rainfalls averaging 60 mm (Supplementary Material).

Statistical analyses

Statistical analyses were designed to characterize seasonal eDNA phenology, quantify regional temporal structure, identify optimal detection periods, and evaluate environmental and biological drivers of eDNA concentration. Regional eDNA phenological timing refers to the relative seasonal position of eDNA concentration dynamics within a region, estimated from temporal patterns in weekly eDNA time series. Individual biological samples were retained for descriptive summaries of the dataset (e.g., tables and text describing sample-level variation). For inferential statistical analyses, the three biological water samples collected during each weekly sampling event, together with their corresponding technical qPCR replicates, were combined to obtain a single weekly concentration for each species and site. These weekly concentrations were log10-transformed and used in all inferential statistical analyses, including time-series reconstruction, cross-correlation, generalized additive models, and settlement-plate analyses. For the optimal detection window (ODW) analysis, monthly mean concentrations were subsequently calculated from these weekly values. All statistical analyses and graphics were conducted in R v. 4.5.1 (R Core Team, 2023) using the tidyverse suite of packages (Wickham et al., 2019), with graphics produced using ggplot2 v. 4.0.2 (Wickham, 2016). The data and code for analyses can be found at https://github.com/Struckhour/ais_weekly.

Time-series reconstruction and normalization

qPCR concentration was used in place of detection rate because detection was consistently high across many species and regions, limiting the ability of detection rates to differentiate seasonal variation in eDNA (Appendix A: Figure A1). Across regions and species, many months exhibited 100% detection, including GOM (C. maenas), and PEI (C. intestinalis, B. violaceus) which showed 100% detection throughout the entire year. Consequently, qPCR concentration was used in place of detection rate to provide sufficient discriminatory power for differentiating and comparing seasonal variation in eDNA, as it is strongly correlated with detection rate and captures variation in signal intensity when detection remains consistently high (Appendix A; Figures 2, 3).

To assess general seasonal variation in eDNA abundance, separate one-way analyses of variance (ANOVAs) were conducted for each species × region combination, with log-transformed qPCR concentration (log10(concentration + 1)) as the response variable and month as a categorical predictor. Statistical significance was assessed at α = 0.05.

Weekly eDNA concentration time series were reconstructed for each species and region using four analytical steps: (1) averaging log 10 -transformed concentration, (2) weekly aggregation, (3) interpolation of missing observations, and (4) normalization of reconstructed time series. These reconstructed time series were used only for the circular cross-correlation and optimal detection window (ODW) analyses.

First, eDNA concentrations were log 10 -transformed to reduce right-skew and stabilize variance across the wide range of observed concentrations. Weekly time series were then constructed by averaging log-transformed concentrations within each ISO week (weeks 1–52). To preserve seasonal continuity, the observed annual series was duplicated on either side of the original sequence prior to interpolation, allowing missing values near the year boundary to be interpolated circularly. Single missing weeks were estimated as the mean of adjacent observations, whereas larger gaps were filled by linear interpolation between the nearest available values. After interpolation, only values corresponding to the original 52-week annual cycle were retained. This approach ensured continuity between weeks 52 and 1 while minimizing edge artifacts associated with cyclic seasonal data. Weeks of the year without data were limited (MAG = 16, PEI = 15, HAL = 5, BOF = 5, GOM = 5)).

To facilitate comparison of the seasonal trends across regions, each regional reconstructed weekly time series was normalized within species and region using min–max scaling (0–1). This approach emphasizes the relative shape and timing of seasonal variation by scaling each series to its own range, allowing patterns to be compared even when absolute variability differs among regions.

Cross-regional eDNA phenological alignment

Regional differences in seasonal eDNA timing were quantified using pairwise circular cross-correlation. Because seasonal dynamics are cyclical, weekly time series were treated as circular such that weeks 52 and 1 were considered adjacent. For each species, all regional pairs were compared by calculating Pearson correlation coefficients across all possible circular temporal shifts (lags in weeks) between normalized time series.

For a given lag, one regional time series was circularly shifted relative to the other, with values extended beyond week 52 wrapped to the beginning of the annual cycle. Correlations were evaluated across all possible lags (0–51 weeks), and the lag that maximized the correlation between each pair of regions was retained as the estimated temporal offset in seasonal timing. Positive or negative values therefore represented earlier or later seasonal timing. Lags were subsequently converted to signed values (−26 to +26 weeks) to represent both the direction and magnitude of regional phenological offsets.

To account for regional geographic gradients in species phenology, all regional pairs were compared separately for each species to identify the lag (in weeks) that maximized correlation. For each species, circular cross-correlation produced an observed temporal lag (L i, j, in weeks) for every pair of regions (i and j), together with the corresponding maximum Pearson correlation coefficient (r i, j). We then estimated a single relative seasonal timing position (T i) for each region by minimizing the weighted sum of squared differences between the observed pairwise lags and those predicted from the estimated regional timing positions:

where T j -T i ​ represents the lag predicted from the estimated seasonal timing positions of regions j and i. Pairwise correlation coefficients (r ij) were used as weights so that regional comparisons exhibiting stronger seasonal similarity contributed proportionally more to the optimization. The resulting regional timing positions therefore represent the configuration that best explains the observed pairwise lag relationships across all regions.

The consistency of regional eDNA phenological structure was evaluated by comparing observed pairwise lag estimates with those predicted from the estimated regional seasonal positions.The predicted lag for each pair of regions was calculated as the difference between their estimated seasonal positions (T i −T j ​). Agreement between observed and predicted lags was quantified using the weighted root mean square error (RMSE), with the same pairwise correlation coefficients (r i, j ​) used as weights. Lower RMSE values indicate that the observed pairwise lags are well explained by a single regional seasonal timing structure, whereas larger RMSE values indicate greater inconsistency among pairwise comparisons. To assess whether regional timing in eDNA phenology was more consistent across species than expected by chance, a permutation test was performed in which region labels were randomly re-assigned within each species and the full alignment procedure was repeated. The null hypothesis was that the association between region identity and seasonal timing was random. For each permutation, region labels were randomly reassigned within each species while preserving the observed pairwise lag estimates. The weighted least-squares optimization was then repeated to estimate the regional seasonal positions, and the mean seasonal timing position of each region across species was calculated. The test statistic was the variance among these regional mean timing positions. This procedure was repeated 1, 000 times to generate a null distribution, and the observed variance was compared with this distribution to calculate the permutation p-value. Larger observed variance indicates a more consistent regional ordering across species than expected under random assignment.

Optimal detection windows and seasonal overlap

To identify periods of peak seasonal eDNA availability, concentration profiles were first normalized by dividing each weekly concentration by the maximum concentration observed for that species-region combination. The Optimal Detection Window (ODW) was defined as the contiguous period surrounding the annual peak during which normalized concentrations remained at or above 90% of the annual maximum. Consequently, concentrations within the ODW represent periods of relatively elevated eDNA concentration compared with the remainder of the annual cycle. The 90% threshold was selected based on the sensitivity analysis described in Appendix B, which demonstrated that it produced finite and biologically interpretable detection windows while remaining robust across species and regions.

ODWs were derived from monthly mean qPCR concentration profiles to provide a consistent and comparable framework for evaluating temporal patterns across species and regions (Figure 4). Concentrations were log-transformed and then normalized within each species–region combination to emphasize relative seasonal patterns rather than absolute differences in magnitude. For each profile, the peak month (maximum mean concentration) was identified, and all monthly mean log concentrations were scaled relative to this peak (i.e., expressed as a proportion of the maximum log concentration). An ODW was then defined as the contiguous set of months surrounding the peak during which normalized concentrations remained at or above 90% of the maximum value.

The 90% threshold was selected to balance temporal specificity and window continuity while ensuring finite detection windows across all species–region combinations. Threshold evaluation and sensitivity analyses are presented in Appendix A. Briefly, this threshold (1) ensured that all species–region combinations yielded finite windows (window duration< 12 months); (2) preserved sufficient variation in window duration to support comparative analyses of temporal overlap (mean window duration = 2.5 months); (3) yielded window strength estimates that were robust to threshold selection (hereafter referred to as probability of superiority: PS) that were largely robust to threshold choice; and (4) minimized the number of threshold-exceeding months excluded from the optimal window. Window strength was quantified using probability of superiority (PS), representing the probability that concentrations within the ODW exceeded concentrations outside the window. To determine whether the regional ordering inferred from the optimal detection windows exceeded that expected under random organization, we performed a separate permutation test using the ODW centre months. The null hypothesis was that the association between region identity and the timing of ODW centres was random across species. For each permutation, region labels were randomly reassigned within species while preserving the observed ODW centre months. The mean ODW timing for each region across species was then recalculated, and the variance among regional mean timings was used as the test statistic. This procedure was repeated 1, 000 times to generate a null distribution against which the observed variance was compared.

Seasonal clustering among regions was quantified using circular statistics based on ODW center months using the ‘circular’ v. 0.5.2 package in R (Agostinelli and Lund, 2025). Because calendar months are cyclical, months were treated as circular variables. Months were treated as points on a circle (to account for the cyclic nature of the calendar year), and clustering was measured using the mean resultant length (R), which ranges from 0 (timing distributed across the year) to 1 (all regions aligned in timing).

Seasonal overlap among regions was quantified for each species using the Jaccard index of similarity (Jaccard, 1912), calculated from the set of months inside the optimal window for each species-region combination.

ODW centers were defined as the midpoint of each optimal detection window, accounting for circular month structure. These center points were used to estimate relative regional seasonal positions while minimizing sensitivity to isolated peak months. For each species, regional timings were projected onto a common seasonal axis and centered to estimate relative timing differences among regions.

To assess whether regional timing structure exceeded expectations under random organization, permutation tests were conducted by randomly re-assigning region labels within species and repeating the full workflow. For each permutation, the variance among mean regional timing positions was recalculated to generate a null distribution.

Local environmental drivers of eDNA concentration

Generalized additive models (GAMs) were used to evaluate relationships between qPCR concentration and local environmental conditions (i.e., temperature and salinity) for each species. Models were fitted separately by species using the ‘mgcv’ v. 1.9.3 R package (Wood, 2017) with restricted maximum likelihood (REML) estimation.

To account for regional structure, eDNA concentration was modeled as a smooth function of week of year using a cyclic cubic regression spline. Cyclic smoothers were used to preserve continuity between the beginning and end of the annual cycle. Temperature and salinity were included as additional smooth terms to allow for nonlinear relationship with qPCR concentration. Only observations with finite values for all predictors were included in analyses.

Model performance was evaluated using deviance explained and Akaike Information Criterion (AIC). The significance and complexity of smooth terms were assessed using estimated degrees of freedom (EDF) and approximate F-tests provided in model summaries.

Plate abundance and life stages on eDNA concentration

Relationships among eDNA concentration, plate abundance, life-stage composition, and seasonal timing were evaluated using species-specific regression tree models. Regression trees were selected to assess potentially nonlinear and interacting relationships among biological and seasonal predictors.

Seasonal timing was represented using sine and cosine transformations of week of year (i.e., sin [2π · week/52], cos [2π · week/52]) to account for the circular structure of seasonal timing. This transformation allowed early and late weeks in the annual cycle to be treated as temporally adjacent rather than opposite endpoints of a linear scale.

For each species, we compared five predictor sets: (1) plate abundance alone; (2) plate abundance plus life-stage variables; (3) plate abundance, life stage, and seasonal timing; (4) seasonal timing alone; and (5) plate abundance plus seasonal timing. This approach allowed evaluation of the relative contributions of settlement observations and seasonal timing to variation in eDNA concentration.

To account for differences in seasonal timing among regions, weekly observations were temporally aligned using the regional lags estimated from the circular cross-correlation analysis before fitting the regression tree models. Each regional series was shifted relative to the reference region by its estimated lag, allowing observations representing comparable positions in the seasonal cycle to be evaluated together. Regression trees were then fitted to the aligned observations to assess environmental and biological predictors of eDNA concentration.

Regression trees were fitted using the ‘rpart’ v. 4.1.24 R package (Thernau and Atkinson, 2025) with an analysis of variance splitting criterion for continuous response data. Model performance was summarized using in-sample RMSE and coefficient of determination (R 2). Comparisons among predictor sets were used to assess whether plate abundance and life-stage composition improved prediction of eDNA concentration beyond seasonal timing alone.

Summary of qPCR results by region and species

The dataset comprised 2, 663 samples and 8, 004 qPCR replicates spanning five regions and five species (Table 1). All targeted AIS were detected in their expected distribution using eDNA (Figures 2, 3). At the PEI site, although C. intestinalis was never observed on the underside of settlement plates, individuals were detected between August and December 2023 beneath floating pontoons near the marina adjacent to the commercial dock where settlement plates were deployed.

Since qPCR detection rates approached saturation across samples and replicates (>90% detection), binary detection metrics provided limited discriminatory power for comparing species, regions, and temporal dynamics (Appendix A: Table A1). Concentration summary statistics differed across groups, with some species and regions exhibiting greater temporal and within-sample variability than others (Table 1). Notably, the GOM region had less variability than the other regions (SD = 1.78), and C. maenas had less variability than other species (SD = 1.74). As expected, D. vexillum was absent from HAL and MAG, and was not tested in PEI; these cases were thus excluded in some subsequent analyses.

Seasonal patterns

A one-way ANOVA for each species and region combination, with month as a categorical predictor, showed significant difference between months (all p< 0.001; Appendix A: Table A1). The strength of the seasonal effect varied among taxa and regions, with the largest F-statistics observed for Botrylloides violaceus in MAG (F 8, 306 = 128.4), Membranipora membranacea in MAG (F 8, 306 = 120.0), and Botrylloides violaceus in PEI (F 10, 322 = 91.5). Although seasonal effects were weaker for Carcinus maenas, they remained highly significant across all regions, with F-statistics ranging from 4.24 to 21.79, indicating that eDNA concentrations consistently varied across months. This seasonal variation across species and regions provided the foundation for evaluating whether seasonal eDNA dynamics were consistently ordered among regions (Figure 3, Table A1). Regional timing estimates derived from pairwise cross-correlation lags revealed coherent shifts in seasonal timing among regions, and application of these estimated shifts substantially improved alignment of weekly log-concentration curves (Figure 4). Across two species (i.e., M. membranacea and B. violaceus), temporal alignment greatly increased overlap among regional time series, with Pearson’s r increasing 0.26 and 0.34 respectively, consistent with a shared seasonal structure and temporal offsets among regions. In contrast, C. maenas and C. intestinalis showed weaker and less consistent cross-correlations, and alignment resulted in limited improvements (Δr = 0.08 and 0.13, respectively), suggesting weaker regional synchrony and greater local variability in seasonal eDNA phenology (Figure 4).

Pairwise circular cross-correlations revealed clear temporal offsets among regions for most species, with moderate to high correlation values observed across many region pairs (Figure 5a). Estimated pairwise lags varied in both magnitude and direction, reflecting differences in the timing of seasonal peaks among regions (Figure 5b). When these pairwise lag estimates were estimated into a common regional seasonal timeline using a least-squares approach, the resulting lag estimates were generally similar to the observed pairwise lag estimates (Figure 5c). Differences between pairwise and estimated lags were typically small (often< 5 weeks), indicating that pairwise lag relationships were largely consistent with a single underlying ordering of regions in relative seasonal timing. Although C. intestinalis had a lower mean pairwise correlation than other species (mean r = 0.54, Figure 5b), indicating greater regional differences in concentration curve shape, it still exhibited a relatively low RMSE (~3 weeks), consistent with a single coherent timing between regions. The high RMSE value for C. maenas is largely due to the MAG region (Figure 5a) and is thus examined separately below.

When region labels were randomly re-assigned within each species and the full alignment procedure was repeated 1000 times, the observed variance in mean regional timing positions exceeded that expected under random region assignment 991 times (p = 0.009), indicating that the inferred regional ordering was unlikely to arise by chance. Inferred timing positions revealed a consistent regional ordering across species, with HAL exhibiting the earliest seasonal timing, followed by GOM and BOF, whereas MAG and PEI showed later timing respectively (Figure 4, Table 2). Mean adjusted positions ranged from 6.69 weeks in HAL to −5.58 weeks in PEI, corresponding to an overall spread of approximately 12.5 weeks in seasonal timing across regions.

Variability in timing estimates differed among regions. HAL exhibited the highest variability (SD = 7.50 weeks), suggesting greater species-specific divergence despite its relatively early seasonal timing. GOM, BOF, and PEI showed moderate variability (SD = 4.01–5.70 weeks). In contrast, MAG exhibited the lowest among-species variability (SD = 3.26 weeks), indicating relatively consistent seasonal timing across species.

Differences between pairwise optimal lags and least-squares estimated lag estimates were substantially larger for C. maenas comparisons involving MAG than for other regions (Figure 5c). eDNA concentrations for C. maenas in MAG were low and observations were sporadic relative to other species–region combinations (Figure 3), suggesting that different ecological dynamics may be occurring in this region. To evaluate the influence of this apparent outlier, analyses were repeated with C. maenas MAG data excluded. This reduced the lag model RMSE for C. maenas to 0.89 weeks, comparable to values observed for other species (Figure 5a). Exclusion of these data did not alter the inferred regional ordering of seasonal timing (Table 1), but strengthened support for non-random regional structure in the permutation test (p = 0.001; all 1000 permutations produced lower variance than the observed variance in regional timing positions). Overall, the regional timing structure became more consistent across species following removal of the C. maenas -MAG outlier. Regardless of whether these data were included, results consistently supported a coherent regional structure in seasonal qPCR timing across species, with earlier seasonal peaks in HAL and progressively later timing toward MAG and PEI (Figure 6).

Timing of optimal detection windows

ODWs differed among regions, with earlier windows occurring in GOM and BOF and progressively later windows in HAL, PEI, and MAG (Figure 6). Consistent with the cross-correlation analysis, M. membranacea, B. violaceus, and D. vexillum exhibited stronger detection windows (mean probability of Superiority (PS) = 0.89, 0.95, and 0.90, respectively) than C. intestinalis (0.85) and C. maenas (0.74), indicating that seasonal eDNA concentrations were more tightly concentrated around the annual peak for these species. Clustering of window centers across regions was moderate to high for most species (M. membranacea: R = 0.65, p = 0.12; B. violaceus: R = 0.89, p = 0.01; D. vexillum: R = 0.71, p = NA due to two regions only; C. intestinalis: R = 0.71, p = 0.07), indicating that window timing was broadly similar across regions (Appendix C: Table C1). In contrast, C. maenas showed weak clustering (R = 0.34, p = 0.58). This was also the species with weakest window strengths (mean PS = 0.74).

The ODW analysis recovered the same regional seasonal pattern observed in analyses of the full concentration curves (Appendix C: Figure C1). HAL exhibited the earliest mean relative timing (−1.60 ± 1.48 months), followed by GOM (−0.68 ± 1.71 months) and BOF (−0.68 ± 1.55 months), whereas MAG (1.53 ± 1.71 months) and PEI (1.78 ± 2.63 months) showed progressively later seasonal timing across species (Appendix C: Table C2; Figure C1). Overlap in seasonal signal among regions was generally low when considering months exceeding the abundance threshold of 90%, with mean Jaccard similarity ranging from 0.10 to 0.34 across species (M. membranacea = 0.10, D. vexillum = 0.14, B. violaceus = 0.22, C. maenas = 0.23, and C. intestinalis = 0.34). Consistent with this reduced overlap, permutation tests based on randomly re-assigned region labels were not significant (p = 0.10), suggesting that some seasonal structure was lost when weekly concentration data were reduced to monthly optimal detection windows.

Environmental drivers

Temperatures increased from spring to late summer, peaking between approximately 15–22 °C across regions before declining through winter and early spring (Figure 7a). HAL and MAG exhibited the greatest seasonal temperature variability, including winter temperatures approaching or below 0 °C, whereas GOM and BOF maintained comparatively moderate conditions year-round. Salinity was generally stable in GOM, MAG, and PEI (approximately 28–32 ppt), while HAL exhibited substantially greater variability and episodic low-salinity events, with values occasionally dropping below 20 ppt (Figure 7b).

GAM modelling revealed that region-specific seasonal structure accounted for most of the variation in qPCR concentration (Figure 8; Appendix D: Table D1). Deviance explained increased from 5–51% in models with a single seasonal smooth to 53–81% when seasonal patterns were allowed to vary by region, indicating substantial regional differences in seasonal dynamics. Seasonal effects were significant in most regions for most species, with most p values clustered below 0.01, although the strength and shape of these patterns differed by species and region, with C. maenas eDNA concentration showing notably less seasonal effect in the BOF region (p = 0.85) (Appendix D: Table D2).

Temperature was a consistent predictor of qPCR concentration for some species, exhibiting significant nonlinear relationships (Figure 9; Appendix D: Table D2) and improving model performance up to 5% beyond region-specific seasonal models (Figures 8, 9; Appendix D: Table D1). Species differed in the consistency of their seasonal structure and overall model performance: M. membranacea, B. violaceus, and D. vexillum exhibited stronger seasonal patterns and high deviance explained (0.81–0.84), whereas C. intestinalis and C. maenas showed less (0.61 and 0.53, respectively). Salinity effects were weak or non-significant and contributed little additional explanatory power (Appendix A: Table D2).

Cross-correlation analyses revealed strong temperature-qPCR relationships (r = 0.5–0.9), (Table 3). Changes in temperature generally preceded the same direction of change in eDNA concentration, although lag magnitude and correlation strength varied among species. The strongest and most consistent patterns were observed for M. membranacea and B. violaceus, with temperature preceding concentration in all regions except HAL for B. violaceus (-8 weeks). In contrast, salinity relationships were weaker, less consistent, and exhibited highly variable lag directions (Figure 10).

HAL exhibited three pronounced salinity drop events during the study period, with two occurring in July 2023 (Event 1, 2) and one in November 2023 (Event 3; Figure 10a). These storm-associated declines were abrupt and short-lived, likely resulting from freshwater input from the river into the HAL estuary. Salinity drop events were generally associated with decreases in eDNA concentrations, although the magnitude and duration of responses varied among species and events (Figures 10b, c). During Event 1, eDNA concentrations decreased for B. violaceus, C. intestinalis, and C. maenas. Event 2 produced the strongest response, with declines observed across all species and highly significant decreases detected for the two tunicates, B. violaceus and C. intestinalis. This event also coincided with complete die-off on the settlement plates for these two species, followed by re-establishment in September 2023, especially for C. intestinalis (Figure 11). Event 3 also showed decreases in eDNA concentrations, although the effects were comparatively weaker.Overall, abrupt reductions in salinity were associated with short-term declines in eDNA concentrations; however, the magnitude and significance of these responses differed among species and events.

Plates and life stages

Settlement plate observations were limited to B. violaceus and C. intestinalis and were uneven across regions and time, with relatively sparse life-stage observations compared to the qPCR time series (Figure 11). At the PEI site, although C. intestinalis was never observed on the underside of settlement plates, individuals were detected between August and December 2023 beneath floating pontoons near the marina adjacent to the commercial dock where settlement plates were deployed. Relationships between settlement plate metrics and eDNA concentration were inconsistent and varied by species, with seasonal dynamics explaining most variation. Notably, when temporal relationships were detected, changes in eDNA concentration consistently preceded changes in settlement plate abundance, suggesting that eDNA may provide an earlier indicator of seasonal population dynamics than settlement observations (Figure 11).

Relationships between settlement plate abundance and eDNA concentration were inconsistent across species and regions. B. violaceus exhibited stronger and more consistent associations than C. intestinalis, with significant positive correlations in MAG (Pearson correlations: n weeks = 18, r = 0.49, p = 0.039), PEI (n = 26, r = 0.58, p = 0.002), and HAL (n = 22, r = 0.54, p = 0.010), whereas BOF showed a significant negative relationship (n = 41, r = −0.48, p = 0.002). In contrast, C. intestinalis exhibited weak and non-significant correlations in MAG (n = 18, r = 0.25, p = 0.326), HAL (n = 31, r = −0.12, p = 0.535), and BOF (n = 41, r = −0.07, p = 0.651); no analysis was possible for PEI due to insufficient observations on the settlement plates.

In several cases, increases in eDNA concentration preceded increases in plate abundance. This pattern was most consistent for C. intestinalis, which exhibited negative optimal lags in MAG (−7 weeks, r = 0.87), HAL (−20 weeks, r = 0.60), and BOF (−11 weeks, r = 0.48). B. violaceus showed more variable lag structure, with negative lags in MAG (−6 weeks, r = 0.74) and BOF (−18 weeks, r = 0.70), but weak positive lags in PEI (1 week, r = 0.54) and HAL (1 week, r = 0.73), indicating regional differences in the temporal relationship between eDNA signal and plate abundance.

Regression tree models indicated that cyclic seasonal timing (i.e., sine and cosine transformations of sampling week [week sine/week cosine]) generally provided the best predictive performance for eDNA concentration, with models including these predictors consistently outperforming those without them (B. violaceus: RMSE = 0.152–0.178, R 2 = 0.454–0.599 versus RMSE = 0.217, R 2 = 0.191; C. intestinalis: RMSE = 0.162–0.183, R 2 = 0.312–0.456 versus RMSE = 0.184, R 2 = 0.306). Plate abundance provided modest additional explanatory power, particularly for C. intestinalis where inclusion of plate abundance improved model performance from RMSE = 0.183 and R 2 = 0.312 to RMSE = 0.170 and R 2 = 0.407. Inclusion of life-stage variables resulted in only minor additional improvements (B. violaceus: no improvement beyond RMSE = 0.152 and R 2 = 0.599; C. intestinalis: RMSE improved from 0.170 to 0.162 and R 2 from 0.407 to 0.456), and these effects appeared to be species-dependent. Regression tree splits further showed that cyclic seasonal timing and plate abundance were the primary drivers of eDNA concentration, whereas life-stage effects contributed more locally within subsets of the data. Because seasonal timing covaries with environmental variables such as temperature, particularly over a single annual cycle, these results should be interpreted as demonstrating that seasonal predictors improved model performance rather than establishing that seasonal timing was more influential than temperature or other correlated environmental variables. Likewise, because life-stage composition covaried with seasonal timing, the models provided limited evidence for an independent effect of life stage on eDNA concentration.

Incorporating temporal lags into the regression tree models provided little evidence that plate metrics predicted future eDNA concentrations (Appendix E: Table E1). Instead, model performance was slightly improved when predicting prior eDNA values, indicating that eDNA signals preceded observable changes in plate abundance by several weeks. For B. violaceus, model performance was highest at lag3 (R 2 = 0.601, RMSE = 0.159, n = 65) compared with lag0 (R 2 = 0.599, RMSE = 0.152, n = 77). Similarly, for C. intestinalis, model performance increased progressively from lag0 (R 2 = 0.456, RMSE = 0.162, n = 83) to lag3 (R 2 = 0.539, RMSE = 0.153, n = 71). However, these improvements were modest and should be interpreted cautiously.

This study provides empirical evidence for eDNA phenology in marine ecosystems, showing that eDNA signals from coastal invertebrate AIS exhibit predictable seasonal dynamics and regionally coherent patterns in eDNA concentration. Ecoregional eDNA phenology exhibited an apparent latitudinal pattern, with earlier seasonal periods of elevated eDNA concentration observed in the southern GOM and BOF and later timing in more northern regions (MAG and PEI). Despite substantial differences among species in life history, invasion stage, and ecology, temporal offsets among regions were largely reproducible across taxa, suggesting that seasonal variation in eDNA concentration reflects structured ecological dynamics operating at regional scales rather than stochastic temporal variation alone. Although our study did not directly evaluate eDNA detectability, characterizing these temporal patterns provides information that can help inform eDNA-based detection by identifying periods when eDNA concentrations are expected to be higher, thereby supporting the optimization of sampling strategies and potentially reducing false-negative detections. Importantly, continuous qPCR concentration data provided greater insight into seasonal ecological patterns than simple presence–absence metrics, supporting the use of quantitative eDNA monitoring to improve the timing and effectiveness of invasive species surveillance.

The eDNA phenology described here extends beyond organismal seasonal dynamics and aligns with the broader framework of eDNA ecology proposed by Barnes and Turner (2016), encompassing seasonal variation in DNA production, transport, persistence and degradation, all of which influence observed eDNA concentration. Seasonal changes in organism abundance, reproduction, biomass turnover, larval release, and shedding rates likely contribute to temporal variation in eDNA production, while hydrodynamics, temperature, and degradation processes further shape the persistence and concentration of eDNA signals through time (Sevellec et al., 2021, 2024, 2025; Rourke et al., 2022; Morrison et al., 2026). The results presented here provide insight into the environmental factors associated with eDNA dynamics in marine ecosystems. Although eDNA concentrations may reflect abundance, particularly when seasonal variation in DNA production associated with reproduction, biomass turnover, larval release, and shedding rates is accounted for, they also integrate environmental processes governing DNA transport, persistence, and degradation. Framing eDNA dynamics within a phenological context establishes a quantitative framework linking species ecology with environmental signal dynamics and provide a basis for developing predictive models of eDNA concentration. Emerging advances in autonomous field technologies, robotic laboratory workflows, high-throughput molecular analysis, and AI-driven modeling are likely to further accelerate the development of quantitative, scalable, and predictive eDNA-based ecosystem assessment frameworks (Lacoursière-Roussel et al. Under review).

The consistent regional ordering observed across species suggests that eDNA phenology is structured by broad-scale environmental gradients operating across coastal ecosystems. We observed earlier seasonal peaks in eDNA concentrations in the GOM and BOF than in HAL, PEI, and the MAG. This pattern is consistent with previously reported differences in regional thermal regimes and seasonal productivity (Brickman et al., 2021; Galbraith et al., 2024). Similar latitudinal organization has been documented across marine plankton communities, food webs, and seasonal productivity cycles, where phenological transitions shift predictably along climatic and hydrographic gradients (Klassen and Locke, 2007; Pepin et al., 2011; Locke, 2013; Harris et al., 2017; Pollet-Calderini et al., 2026). Coastal eDNA metabarcoding studies have similarly shown that marine eDNA community composition is strongly structured by seasonal and hydrodynamic forcing (Sevellec et al., 2024, 2025; Morrison et al., 2026). The comparatively delayed and stable seasonal timing observed in the MAG may reflect more constrained environmental seasonality or reduced hydrodynamic variability within the southern Gulf system (Galbraith et al., 2024; Lavoie et al., 2021). The persistence of regional temporal offsets across unrelated invasive taxa suggests that eDNA phenology may represent a broader ecosystem-level property emerging from coupled environmental and biological seasonality.

Temperature emerged as a consistent predictor of eDNA concentration across species, whereas salinity was not a primary driver of seasonal eDNA dynamics (Figure 10). The nonlinear thermal responses observed across taxa suggest that temperature influences eDNA signals through multiple pathways, including metabolism, reproductive activity, biomass turnover, shedding, and DNA degradation (Lacoursière-Roussel et al., 2016; Wood et al., 2021; Morrison et al., 2023). Similar seasonal increases in eDNA concentration associated with reproductive activity and larval release have been reported for tunicates, bryozoans, and zebra mussels (Rounds et al., 2024). Recent temporally resolved eDNA metabarcoding studies similarly demonstrate that seasonal environmental forces strongly structure marine eDNA dynamics across coastal ecosystems (Morrison et al., 2026). In our study, positive temporal lags between temperature and DNA concentration of species further suggest that environmental warming precedes the increases in eDNA concentration through delayed biological responses rather than immediate environmental forces alone. These lagged relationships, together with the stronger predictive power of seasonal timing relative to environmental measurements, indicate that eDNA phenology integrates ecological processes occurring across multiple temporal scales.

In contrast to temperature, salinity exhibited comparatively weak and inconsistent relationships with eDNA dynamics across species and regions. Notably, Halifax displayed distinct temporal dynamics relative to the other study regions, coinciding with greater salinity variability and suggesting comparatively disrupted and less predictable eDNA phenological patterns. Halifax Harbour is a hydrologically dynamic estuarine system subject to episodic storm-driven freshwater input, tidal mixing, and rapid flushing that often results in abrupt but transient salinity declines (Shan and Sheng, 2012). In the present study, these storm events coincided with short-term decreases in eDNA concentrations across all four AIS examined. Similarly, salinity during several different storm events around the world approached or fell below the lower tolerance thresholds reported for multiple taxa, including C. intestinalis (19 PSU; Carver et al., 2006; Vercaemer et al., 2011), B. violaceus (20 PSU; Carver et al., 2006; Dijkstra et al., 2008), M. membranacea (20 PSU; Radhakrishnan Nair et al., 1992), and C. maenas (5 PSU; Young and Elliott, 2019). These conditions may have induced physiological stress and localized mortality, consistent with the complete die-off we observed on settlement plates following the second Halifax rainstorm event. Although such stress responses could transiently elevate eDNA release through biomass turnover or organismal die-off, the observed temporal patterns were more consistent with rapid dilution and hydrodynamic export of eDNA. These stochastic hydrodynamic disturbances likely altered eDNA transport, retention, and release dynamics within the estuary. Resolving these transient processes would require higher-frequency sampling at daily or sub-daily resolution, particularly for studies aiming to use eDNA concentrations as proxies for species abundance. Recurrent disturbance events may also have contributed to broader seasonal dynamics, as Halifax exhibited an earlier seasonal peak in eDNA concentrations relative to other regions, followed by declines after major rainfall events while concentrations elsewhere continued increasing later into the season. Together, these findings suggest that episodic coastal storm events can disrupt otherwise consistent eDNA phenology and reduce the predictability of eDNA concentration at local scales.

Consistent with our study objectives, we found that AIS exhibited coherent regional phenological patterns across the latitudinal gradient despite substantial differences in taxonomy and life history. These recurring regional patterns suggest that seasonal variation in eDNA concentrations reflects broad-scale ecological processes operating across coastal ecosystems rather than local environmental conditions alone. The coherent seasonal patterns observed across regions, together with the comparatively weaker and less consistent relationships with the instantaneous environmental variables measured in this study, suggest that seasonal eDNA dynamics are shaped by a combination of broad-scale environmental forces, including hydrodynamic connectivity, productivity cycles, delayed biological responses, and temporal variation in DNA transport and persistence (Buxton et al., 2017; Rourke et al., 2022; Gu et al., 2026). This interpretation is further supported by the lagged relationships observed both in cross-correlation analyses (Table 3) and between eDNA concentration and settlement plate abundance, where increases in eDNA concentration frequently preceded observable recruitment by several weeks. Together, these findings suggest that eDNA phenology reflects temporally integrated ecological dynamics rather than instantaneous environmental conditions alone.

Variation among species further indicates that eDNA phenology is influenced by species-specific ecological traits. Differences in life history, reproductive timing, dispersal capacity, biomass dynamics, invasion stage, and mobility likely contributed to variation in the strength and coherence of seasonal eDNA patterns among taxa (Chrismas et al., 2023; Hervé et al., 2022; Jensen et al., 2022; Lin et al., 2024). Colonial sessile organisms such as B. violaceus and M. membranacea exhibited stronger and more coherent regional synchrony, potentially reflecting localized biomass accumulation and relatively direct coupling between benthic populations and water-column eDNA signals. Consistent with previous observations of elevated late-summer eDNA detectability for tunicates and bryozoans in the Northwest Atlantic (Morrison et al., 2026), these taxa displayed comparatively stable and pronounced seasonal trajectories across regions.

The weaker synchrony observed for C. maenas may reflect greater mobility, transient population dynamics, weaker benthic coupling of eDNA signals, or habitat-specific occupancy resulting in sporadic marina presence. Species-specific variability in eDNA detectability has similarly been observed for invasive crustaceans using targeted qPCR approaches (Chevrinais et al., 2023b). The anomalous dynamics observed in MAG further suggest that invasion stage, population structure, and regional larval transport may influence eDNA phenology, with low-density or spatially patchy populations generating weaker and less coherent temporal signals. In the MAG sampling site, C. maenas detections likely reflect intermittent presence and low local abundance rather than a large established population (Figure 2), as suitable habitat is limited within the harbour itself and persistent populations on the MAG are primarily associated with lagoon systems. Repeated eDNA detections and recent diver observations of adult crabs nevertheless support ongoing local presence. Temporal detections in MAG may also be influenced by larval transport from external source populations, although due to oceanographic conditions this is unlikely to be significant. Natural larval dispersal from Nova Scotia is considered highly improbable due to prevailing southward currents flowing out of the Gulf of St. Lawrence (Johnson et al., 2024; Krumhansl et al., 2023; Ryan et al., 2019). Dispersal from Newfoundland is also considered unlikely given the large geographic distance and prevailing surface circulation patterns, particularly the Laurentian Channel outflow. In contrast, PEI and coastal New Brunswick likely represent the most plausible regional sources for natural dispersal. Interestingly, C. maenas populations in PEI and MAG belong to a distinct cold-tolerant genetic lineage relative to the warm-tolerant southern populations found in the GOM and BOF (Jeffery et al., 2017; Lehnert et al., 2018). Lineage-specific environmental response curves, including differences in thermal optima associated with key life-history processes such as reproduction and growth, may therefore contribute to regional differences in invasion dynamics, phenology, and population density, with downstream consequences for eDNA detectability (Cosham et al., 2016; Lehnert et al., 2018). These effects may be further amplified by the MAG representing the northern range limit of C. maenas distribution, where populations are likely more sensitive to interannual environmental variability.

Discrepancies between settlement plate observations and eDNA concentrations further suggest that eDNA captures ecological processes extending beyond local substrate recruitment. Positive eDNA detections in the absence of visible settlement may reflect propagule pressure, nearby source populations, larval transport, cryptic biomass, habitat mismatch, or partial decoupling between benthic populations and water-column DNA signals. Although some tunicate species preferentially settle on shaded undersurfaces of artificial structures (Gulliksen, 1972; Svane and Young, 1989; Howes et al., 2007), settlement can also occur on nearby submerged structures rather than directly on settlement plates. This pattern was observed for C. intestinalis in Souris, PEI, where eDNA concentrations remained consistently positive despite the absence of settlement on settlement plates. Similar spatial decoupling has been observed for D. vexillum, which can proliferate extensively on nearby artificial and natural substrates, including wharves, vessel hulls, rocky benthic habitats, and suspended structures, while remaining unevenly distributed across adjacent settlement surfaces (Chase et al., 2016; Teed et al., 2024). Regional boating-mediated dispersal pathways, hydrodynamic connectivity, and local substrate competition could also contribute to these spatially decoupled eDNA signals, although the relative importance of these mechanisms remains poorly resolved (Simard et al., 2017). The temporal lags observed between eDNA concentration and settlement plate abundance further support this interpretation, with eDNA signals often preceding observable settlement by several weeks. These offsets likely reflect the detection of larvae, propagules, reproductive activity, or nearby source populations prior to successful settlement and visible colonization on monitoring substrates. Such lag structure suggests that eDNA may provide earlier indication of invasion activity than settlement-based monitoring alone.

The coherent regional differences in eDNA phenology observed during the study period have important implications for AIS surveillance and the design of monitoring programs. Rather than assuming uniform sampling schedules across the Northwest Atlantic, surveillance programs may benefit from accounting for regional differences in seasonal eDNA dynamics. However, because our study was conducted over a single annual cycle, the temporal stability of these regional patterns remains to be established. Multi-year studies will be needed to determine whether these phenological differences are consistently maintained among years or vary in response to interannual environmental variability.

The limited overlap in periods of elevated eDNA concentration among regions indicates that temporally static sampling strategies may systematically fail to capture periods of peak detectability across broad geographic scales. Within the framework developed here, seasonal peaks in eDNA concentration represent discrete phenological features of eDNA signals characterized not only by periods of maximum concentration, but also by their timing and duration. Because eDNA concentrations remained highly seasonal even when detection rates approached saturation, sampling timing strongly influenced both detection sensitivity and ecological interpretability. These findings suggest that temporal optimization of sampling schedules may improve surveillance efficiency more effectively than increasing sampling intensity alone. For management-oriented AIS surveillance, region-specific monitoring schedules that account for local eDNA phenology may improve eDNA-based surveillance by increasing the likelihood of sampling during periods of elevated eDNA concentration, thereby reducing false-negative detections, particularly during the early stages of invasion when eDNA concentrations are expected to be low. However, low eDNA concentrations and non-detections also represent important ecological information and should be interpreted within the context of seasonal eDNA dynamics rather than solely as limitations of surveillance. Furthermore, because phenological patterns may vary among years, region-specific sampling schedules should be considered adaptive and refined as additional multi-year observations become available. These findings support growing recommendations for integrating temporally optimized eDNA surveillance into operational AIS management frameworks (DFO, 2020; Zaiko et al., 2018).

Several methodological and ecological constraints should be considered when interpreting the mechanisms underlying the observed patterns of eDNA phenology. Seasonal trajectories were reconstructed using interpolation to address incomplete temporal coverage, which may have influenced the fine-scale timing and shape of some seasonal patterns despite efforts to preserve cyclic continuity. Similarly, temporal smoothing approaches may reduce short-term variability while emphasizing broader seasonal trends. eDNA concentration also represents an indirect proxy for organism abundance and cannot distinguish among variation in biomass, shedding rates, DNA production, transport, or persistence. Consequently, the relative contributions of biological activity versus environmental variables on eDNA dynamics could not be disentangled. Environmental analyses were limited to temperature and salinity and therefore did not capture additional potentially important drivers such as hydrodynamics, nutrient availability, productivity, or water-column mixing. In addition, the reduced regional representation of D. vexillum limited the ability to assess spatial consistency for this species relative to the other taxa. Although the mechanisms underlying eDNA phenology likely differ among taxa and ecosystems, the consistency of the observed regional structure suggests that temporally organized eDNA dynamics represent an emergent ecological property of coastal systems.

Beyond optimizing sampling timing, the eDNA phenology framework could also complement spatial decision-support tools used in aquatic invasive species surveillance. For example, integrating seasonal detection windows with geographic information systems (GIS), habitat suitability models, or species distribution models could help prioritize where and when surveillance efforts are most likely to detect target species, thereby improving the efficiency of monitoring programs and supporting early detection and rapid response strategies. Similar integration of eDNA and GIS has been proposed as a promising approach for invasive species management and biosecurity planning (Mohd Dali et al., 2023).

Together, these findings extend the ecological framework proposed by Barnes and Turner (2016) by demonstrating that eDNA phenology represents a reproducible and predictable ecological property of coastal ecosystems. Consistent with the emerging GOTeDNA framework (Lacoursière-Roussel et al., 2026) and recent eDNA metabarcoding studies (Morrison et al., 2026; Sevellec et al., 2024, 2025), our results suggest that recurring seasonal organization in eDNA concentration contains ecologically meaningful information operating across ecosystem scales. Because these temporal patterns were conserved across taxa and regions, they may provide a foundation for predictive ecological modeling of seasonal eDNA dynamics. Such models could ultimately help identify periods when eDNA concentrations are expected to be highest, thereby informing the timing of eDNA sampling. By linking organismal phenology with the seasonal ecology of eDNA signals, temporally resolved eDNA datasets provide opportunities for advanced time-series modeling, ecological forecasting, anomaly detection, and adaptive monitoring frameworks. More broadly, these findings highlight the potential of integrating ecological theory with AI, machine learning, and spatiotemporal analytical frameworks to develop predictive surveillance systems capable of capturing dynamic seasonal processes across coastal ecosystems.