Nearshore temperature dynamics can be highly heterogeneous both spatially and temporally, due to the complex interplay of air-sea heat fluxes with localized geometry, atmospheric forcings that influence water currents and mixing, and its interaction with shelf and deep ocean physical processes. In turn, these highly variable temperature regimes have potentially significant effects on valued ecosystem components and influence coastal ecosystem processes. Eelgrass (Zostera marina) ecosystems provide important ecosystem services such as shoreline protection, water filtration, carbon storage, and fisheries maintenance (Fourqurean et al., 2012; Nordlund et al., 2016). Light, temperature, and nutrients all influence eelgrass growth and production (Lee et al., 2007; Enríquez et al., 2019). In this study, we focus on nearshore temperature dynamics which is a direct and indirect structuring elements for seagrass ecosystems, and of particular interest in the context of climate change. Previous work has shown that temperature effects on eelgrass are multi-faceted (Krumhansl et al., 2021; Wong and Dowd, 2023). Eelgrass is also susceptible to marine heatwaves that originate offshore but propagate into, and are exacerbated by, nearshore conditions (Marbà and Duarte, 2010; Moore et al., 2014; Strydom et al., 2020; Wiberg, 2023). Understanding the relationship of coastal ecosystem processes with the physical environment requires high resolution physical data across large spatial scales. Unfortunately, in the nearshore zone it is often not feasible to obtain this information from in-situ measurements or satellite data due to limitations in data resolution and spatial scales. Hence, we must rely on properly calibrated and validated numerical ocean models. Despite advances in oceanographic model developments, predicting temperature accurately and capturing its variability on the small but important spatial and temporal scales characteristic of the nearshore is difficult. The reasons include accurately representing advective processes due to complex coastlines and bathymetry, adequately resolving air-sea heat exchange and absorption, and ensuring proper dynamical coupling with the adjacent shelf. Furthermore, model predictions are most useful if they are evaluated using ecologically meaningful temperature metrics that are linked to target ecosystems or species.

Our study region is the Atlantic coast of Nova Scotia, Canada. Water temperature over the Scotian Shelf has strong spatial and seasonal variability, with the main controlling mechanism being the air-sea heat flux (sum of the flux of solar heating, sensible heat, latent heat and longwave radiation) that accounts for about 85% of the observed temperature variability (Umoh and Thompson, 1994). Cold water upwelling yields an important temperature signal in summer, and horizontal advection and vertical mixing have relatively smaller contributions. Additionally, large scale variations in water temperature over the Scotian Shelf are related to the two dominant equatorward flows over the Scotian Shelf (Thompson et al., 1988; Petrie, 2007; Brickman et al., 2018). The first is the inner-shelf current along the Atlantic coast, fed by a branch of the outflow from Gulf of St. Lawrence; and the second is the current along the shelf break, that is an extension of the Labrador Current (Sutcliffe et al., 1976). The two seasonally varying currents are topographically steered by banks, basins and channels, leading to variations in water temperature (Petrie and Drinkwater, 1993; Drinkwater, 1996; Hannah et al., 1996; Wu et al., 2016).

Numerical models of the physical oceanography in this region have emphasized the Scotian Shelf, but largely ignored the nearshore due to the difficulty in adequately resolving it, despite its importance to many valued ecosystem components such as eelgrass. During the last four decades, numerical models have been developed for the Scotian Shelf based on various types of circulation models with different model resolutions. For example, using finite element models, Han et al. (1999) and Hannah et al. (2001) investigated the seasonal variation of the circulation over the shelf with model resolution that varied from 2 km over the coastal waters to 30 km in the deep ocean. Using a nested-grid modelling system, Sheng et al. (2006) studied the response of the upper ocean to storms using a model resolution over the shelf of about 7 km. Using an ice-ocean coupled model based on the Princeton Ocean Model, Wu et al. (2012) developed a circulation model with a horizontal resolution of about 10 km, while Katavouta et al. (2016) developed an ocean circulation model based on the Nucleus for European Modeling of the Ocean (NEMO) with a horizontal resolution of 2.8 km. These models represent well the key dynamics driving large-scale temperature variations over the Scotian Shelf, but cannot accurately represent nearshore processes due to the relatively coarse model resolution used. More recently, Feng et al. (2022) developed a model based on the Finite Volume Community Ocean Model (FVCOM) for the eastern shore island area of the Scotian Shelf, however the spatial variation of water temperatures in the target eelgrass areas in this study were still not adequately resolved. The challenge to accurately modelling the nearshore temperature is the complicated bathymetry and coastlines. This requires spatial resolution down to a few meters to achieve reasonable temperature predictions, and to accurately represent the nearshore dynamics (Lynge et al., 2010; Poje et al., 2010; McWilliams, 2016).

The eelgrass areas used for this study are characterized by irregular coastlines, deep bays with steep shorelines, shallow bays with elevated intertidal flats and tidal channels, and many islands and headlands with strong tidal flows. Consequently, these eelgrass beds inhabit a wide range of environmental conditions, from shallow, warm, protected waters to deep, cool, exposed waters (Wong, 2018; Krumhansl et al., 2021). Eelgrass beds also experience high temporal variability in water temperature from not only localized processes such as air-sea heat fluxes (i.e., local forcing), but also tidal and wind driven advective heat fluxes originating on the shelf (i.e., remote forcing) (Wong et al., 2013; Wong and Dowd, 2021). To understand the dynamics of water temperature in the eelgrass areas, in this study we develop a high resolution numerical oceanographic model which uses an unstructured mesh that allows for very high spatial resolutions at sites of interest. This allows us to represent the complex coastline and bathymetry and to provide accurate water temperature predictions where needed. Using the model results, we examine ecologically meaningful temperature metrics (i.e., mean temperature, heat accumulation, daily temperature range, thermal physiological threshold exceedances) that are known to influence seagrass growth and productivity (Krumhansl et al., 2021; Wong and Dowd, 2023). Finally, a simple heat budget is developed to identify the primary mechanisms underlying the temperature dynamics at select study sites.

We investigated temperature dynamics in nearshore regions of the Atlantic coast of Nova Scotia on time and space scales relevant for coastal ecosystem processes. Time series of water temperature from observations and FVCOM numerical model results for June 2018 to May 2021 at six different eelgrass sites were used for validation and characterization of temperature regime. We demonstrated that the numerical model can generally predict well the key attributes of temperature relevant to eelgrass ecosystems, and do so across large spatial scales, in this case the whole Atlantic coast of Nova Scotia. We therefore argue that validating models against observations using biologically ‘fit for purpose’ temperature metrics is important to ensure that relevant attributes are well predicted, such as thermal exceedances, heat accumulation, heat wave events and daily temperature variation. Moreover, a properly calibrated and validated numerical model allows for detailed site-specific characterization of the temperature regime, as was demonstrated for the various study site. It can also help identify at-risk areas for target ecosystems resulting from temperature stress, now and in the future (Krumhansl et al., 2020; Wong and Dowd, 2021).

An overall good agreement between the numerical model predictions and the observed temperature was found. The model simulated time series of the water temperature had a Willmott skill > 0.7, and we were able to assess this in a frequency dependent manner, a feature which extends the Willmott skill score concept (Equation 2). Summary statistics during the summer growing period from the model were within 2°C of the recorded data in most sites ( Table 4 ). The discrepancies between growing degree day from the model calculation and the observations (≤ 19%) were consistent with the systematic differences in the time series of water temperature. While the number and duration of warm events highly depended on the definition used (e.g., duration, length, and separation of events) due to the high temporal variation of temperature, the discrepancies in the mean duration of events between the model results and those from the observed data were generally within the error standard deviation of calculations.

The main discrepancies between the model and the observations were in the summer temperatures at the shallow sites (Port l’Hebert and Port Joli), which were consistently overestimated. It is important to identify such biases and consider their implications when using numerical models predictions to assess ecological implications of various temperature processes (e.g., growing degree days is quite sensitive to this discrepancy). Potential reasons are the following. Firstly, model bathymetries in the tidal channels may be too shallow, which would significantly limit exchange processes, in particular a decrease in the inward advection of cold offshore water and overestimated water temperature in the inner bay. Also, any local smoothing of bathymetry would lower the resolution of channels which ventilate the shallow intertidal flats and hence affect the temperature dynamics there. Secondly, the horizontal resolution of the air forcing (e.g., air-sea heat fluxes) from ERA5 is 31 km, which is relatively coarse compared to the model resolution used in this study. Hence, some coastal bays (e.g., Port l’Hebert and Port Joli) are represented as being on land in ERA5, which may cause artificially high water temperatures in summer (note the coarse ERA5 resolution can lead to underestimation of water temperature in deeper exposed sites by not underrepresenting the land). Thirdly, the overestimated water temperature in summer could also be caused by the uncertainty in solar radiation attenuation properties in the bottom layer, where the solar radiation not only heats the water, but also the bottom sediment due to the shallow water depth. Since the amount of the solar energy stored in the sediment is unknown, we considered all the solar energy as being distributed through the water column. To examine this process, we ran the model with different water column attenuation coefficients for solar radiation and found that the model performance can be improved by tuning them, however, development of a realistic attenuation parameterization in the bottom layer is beyond the scope of the present paper.

Previous studies have suggested that short-term, sub-seasonal temperature processes (i.e., warming events, wind events, upwelling) can play an important role in eelgrass growth and productivity (Wong et al., 2013, 2020, 2021; Krumhansl et al., 2021). However, it is important to note that the ecological function of seagrass beds is also influenced by the underwater light environment and its variations (Enríquez et al., 2019), which can act separately and together with temperature effects. The characterization of the temperature regime for our study sites indicated a similar seasonal variation in the bottom water temperature, but variability on sub-seasonal scales was markedly different across the sites. Higher water temperature in the high frequency band (<48 hr; Table 3 ; Figures 4D , 5D ; Supplementary Figures S2D - S5D ), larger dominance of solar heating and diurnal tides relative to the semi-diurnal tides ( Figure 2 ), and relatively higher daily temperature range ( Figures 8 ; Supplementary Figure S6 ) indicate a greater impact of processes on time scales< 48 hr at shallow sites (Port l’Hebert, Port Joli, Mason’s Island) compared to the other sites. Temperature variations in the meteorological band that were observed in all the sites ( Figures 4C - 5C ; Supplementary Figures S2C - S5C ) can be due to local wind events as well as coast-wide processes such as storms and wind-driven coastal upwellings. A coastal upwelling index based on Ekman transport and upwelling favorable winds (Petrie et al., 1987) has shown strong coherence with the meteorological temperature band in eelgrass sites at the eastern Scotian Shelf (Krumhansl et al., 2021), which can transfer cool nutrient rich water to the surface and support eelgrass growth and photosynthesis during periods of nutrient limitation (Sandoval-Gil et al., 2019).

Extended high temperature can have negative impacts on eelgrass health. Physiological impacts on eelgrass occur within 1-7 days when temperatures are 19-28°C (Evans et al., 1986; Gao et al., 2017), or as short as 15 min at > 30°C. Our results showed > 2 hr warm water events occurred only at Port Joli and Port l’Hebert based on 23°C (optimum temperature for photosynthesis; > 12 events year^-1) and 27°C thresholds, while other sites experienced these events based on a 20°C criteria. In shallow sites, these events are likely due to long periods of solar heating over the extensive shallow flats (Wong et al., 2013). Warm water events based on the 23°C threshold typically lasted > 10 hr. The high frequency of warm events in the eelgrass sites in this study suggest that these eelgrass frequently experienced physiologically unfavorable conditions. Alternatively, anomalous warming can result in persistent changes in eelgrass bed characteristics across multiple clonal generations and years (DuBois et al., 2020). Previous studies show that these events are typical on the Atlantic coast of Nova Scotia (Wong et al., 2013; Wong, 2018; Krumhansl et al., 2021), and that eelgrass can thermally adapt to varying temperature regimes, similar to other seagrass species (Marin-Guirao et al., 2016).

The simple heat balance analysis contrasting a shallow protected bay (Port l’Hebert) and a deeper exposed (Taylor Head) site showed that while the maximum annual changes in the heat content at the shallow site are greater than those at the deep site, the surface heat flux is the main contributor to the temperature variations during summer growing seasons at both sites ( Figure 10 ). Monthly mean contribution of the advective fluxes in both sites was negative and buffering the surface heating in the warm months. Higher values of surface heat flux (related to daily heating and cooling) and advective flux (related to tidal exchanges), which are of the high frequency temperature changes, during warm months can be an indicator of higher amplitudes of temperature variations within the high frequency during these periods. Monthly mean magnitude of the contributing terms in the heat budget showed small interannual variabilities at both sites. Heat budgets are an important way to determine to what degree a study site is dominated by local processes versus being driven by offshore ocean dynamics, with potential implications for understanding ecosystem functioning. They are readily computed from numerical model output, but difficult to obtain directly from temperature observations.

In summary, the FVCOM model developed for this study was able to reasonably predict water temperature variations and thermal metrics relevant to eelgrass condition. The representation of nearshore dynamics and its impact on coastal water temperature across large spatial scales can only be achieved by using targeted calibrated and validated high resolution numerical models. Furthermore, we argue that the validation process for these models should be targeted to the ecosystem process of interest, using specific variables (e.g. temperature) and biologically tailored metrics. In this way, we can improve our general understanding of the interaction between the physical and biological processes in the coastal environments, as well as by taking into account other key physical drivers such as light. It is hoped that results of this study, and the numerical model developed here, can contribute towards the understanding, conservation and protection of coastal ecosystems, and the maintenance of their functioning and resilience in this era of climate change.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

AJ: Conceptualization, Data curation, Formal Analysis, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing, Supervision. YW: Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Writing – original draft, Writing – review & editing. MW: Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. MD: Conceptualization, Formal Analysis, Methodology, Supervision, Writing – original draft, Writing – review & editing.