The breathing seascape: Spatial and temporal up-scaling of an estuarine ecosystem function

Summary

Primary production is a key ecosystem function and health indicator. Here, we used high-resolution light modeling and in situ flux data to map seafloor primary production in a shallow estuary, incorporating spatial variability associated with bathymetry and seabed habitat type (mud, sand, and seagrass) and temporal dynamics associated with changes in suspended sediment concentration, seasonal light availability, water depth, and daily light-dark cycles. The estuary was net autotrophic (i.e., benthos contributing oxygen to overlying seawater) every day of the year, with an annual estuary wide average productivity of 51.3 ± 8.4 mmol m⁻² d⁻¹. Patches of seagrass were the most productive areas of the estuary, whereas deep channels and muddy seafloor habitats in the inner arms of the estuary (where turbidity was also the highest) were the least productive. The seafloor exhibited pulsed rhythms of primary productivity (daily, tidal, and seasonal differences in oxygen production and consumption), resembling a “breathing seascape”.

Graphical abstract

Highlights

• •Seafloor primary production dynamics were modeled in a shallow estuary
• •Seagrass habitats were the most productive, deep muddy areas the least
• •The estuarine seafloor was net autotrophic, on average, every day of the year
• •Productivity rises and falls daily throughout a year, creating a breathing seascape

Aquatic science; Biogeochemistry; Marine processes; Oceanography

Introduction

Marine primary production (use of sunlight energy by organisms that “fix” carbon) is an essential ecosystem function. It generates at least half of the world’s oxygen and acts to keep our atmosphere breathable, while also regulating the global carbon cycle.^1,2,3 It supports marine food webs: is used as an indicator of marine ecosystem health, resilience, and environmental change^4,5,6; and may be responding to the “darkening of the ocean” caused by changes in seawater properties that influence light penetration.⁷

Mapping of surface ocean productivity, using remotely sensed ocean color as a proxy for phytoplankton pigment concentration,^8,9,10,11,12 is yielding insights into global patterns and climate-related distributional shifts.¹³ Remote sensing of surface ocean productivity is also increasing our understanding of the extent, intensity, and duration of blooms,^14,15,16 including harmful algal blooms,¹⁷ and is being used as a predictor of biodiversity and fisheries hotspots.^18,19,20

In inshore waters, however, broad scale mapping of primary production remains challenging. Remote sensing algorithms for sea surface phytoplankton are confounded by absorption and reflectance by non-phytoplanktonic material (e.g., suspended sediments and colored dissolved organic matter).²¹ This material is common and often highly concentrated in inshore waters such as estuaries and river plumes. Moreover, the seafloor’s absorption/reflectance properties also confound the algorithms, with the strength of seafloor signals increasing as water depth decreases. Thus, assessing primary production at broad scales in shallow estuaries using satellites or remote sensing is problematic.^22,23,24,25

In shallow estuaries, rates of benthic primary production likely exceed those of water column primary production. The biomass of phytoplankton in the water column above a standard area of seafloor decreases as water depth decreases, whereas the biomass per area of seafloor of benthic primary producers (microphytobenthos, turfing algae, and seagrass) often exceeds that of phytoplankton and generally increases as water depth decreases. Thus, the shallower the water, the greater the relative contribution of the benthos.²⁶ In estuaries with very shallow depths (average ≤ 2 m), seabed primary production may account for a majority of the total in situ ecosystem primary productivity. Coastal darkening may shift the relative contributions of benthic and water column primary productivity down and up, respectively, resulting in changed ecosystem dynamics.²⁷ Furthermore, sea level rise and the deepening of shallow coastal ecosystems will exacerbate such effects.²⁸ Therefore, methods of measuring seafloor primary production dynamics in vulnerable shallow estuaries, where remote sensing performs poorly, need to be developed.

Published studies with maps of marine ecosystem functions and services are rare, which is a barrier to the protection of areas with high marine ecosystem functioning and ecosystem service delivery (e.g., through marine spatial planning). Protecting “areas of particular importance for ecosystem services” is a strategic goal of the Convention on Biological Diversity (The Kunming-Montreal Global Biodiversity Framework; https://www.cbd.int/gbf/targets). The few available examples of marine ecosystem function/service mapping (e.g., primary production, nitrogen removal, food provision, water quality regulation, and sediment stabilization)^{13,22,29,30,31,32} are static in time, rather than dynamic. A static map may be suitable to represent some marine ecosystem services. For example, food provision by shellfish beds can remain consistent for years if multiple generations of long-lived individuals form the bed. However, seafloor primary production—strongly driven by light availability—is highly temporally dynamic, varying according to daily, tidal, and seasonal cycles.³³ Moreover, sediment loading to estuaries and resuspension of bed-sediments by waves can reduce light transmission to the seabed irregularly (following storms), as can intermittent algal blooms (triggered by nutrient inputs and favorable temperatures), complicating the more predictable cyclical patterns of seafloor primary production.³³

In this study, we mapped seafloor primary production across the entire area of a shallow estuary in New Zealand (mean depth ∼2 m). We focused on incorporating spatial variability associated with bathymetry and seabed habitat type (i.e., mud, sand, and seagrass), and temporal dynamics associated with changes in suspended sediment concentration, seasonal light availability, tidal water depth variation, and daily light-dark cycles. We aimed to demonstrate a methodology that can be used to quantify and visualize the dynamics of seafloor primary productivity in shallow coastal systems in lieu of remote sensing tools. Our models were mainly informed by empirical data collected within and near the estuary (i.e., not satellite derived). Results are presented in a series of static figures in this article, but the dynamic “breathing seascape” models are available as movie files in the supplemental information.

Results

Photosynthesis-irradiance curves for seagrass, sand, and mud

The photosynthesis-irradiance (P-I) relationships derived using seafloor irradiance and aquatic eddy covariance (AEC) oxygen flux data showed the dependency of seafloor primary production rate on habitat type (Figure 1). For example, the maximum rate of photosynthesis, P_max, in seagrass was 183 mmol m⁻² d⁻¹, which was five and a half times higher than that of mud (33.1 mmol m⁻² d⁻¹) and more than double that of sand (88.1 mmol m⁻² d⁻¹). However, the seafloor’s rate of oxygen consumption during nighttime, y0, was also dependent on habitat type. Rates of nighttime seafloor oxygen consumption were highest in seagrass (72.3 mmol m⁻² d⁻¹), followed by sand (47.6 mmol m⁻² d⁻¹) and mud (28.6 mmol m⁻² d⁻¹). The three sets of curves indicate a switch from net oxygen consumption to net oxygen production at differing levels of seabed irradiance, known as the compensating irradiance point, I_C. This point occurred at the lowest level of irradiance for mud (I_C,mud = 23.8 μmol m⁻² s⁻¹), next for sand (I_C,sand = 34.6 μmol m⁻² s⁻¹), and at the highest level of irradiance for seagrass (I_C,seagrass = 69.3 μmol m⁻² s⁻¹). These differences, coupled with information on the distributions of each habitat type within the study domain, enabled us to characterize the spatial and temporal dynamics of oxygen production and consumption in the Ōhiwa Harbor at the daily scale over the course of a year.

Figure 1.

Photosynthesis-irradiance relationships developed from sixteen individual aquatic eddy covariance deployments in northern New Zealand, including in Ōhiwa Harbor

Oxygen flux (y, mmol O₂ produced per m² per d) was predicted from seafloor irradiance, PAR_S (x, μmol photons per m² per s), using a standard formula for photosynthesis-irradiance (P-I) relationships, y = y0+a(1-exp(-bx)). Individual curves were developed for seagrass, sand, and mud (four to seven multi-day aquatic eddy covariance [AEC] deployments in each habitat class), all with p < 0.0001 and 0.5276 ≤ r²_adj ≤ 0.8355. See Table 1.

Spatial variation in seafloor primary production

Spatial variation in seafloor primary production rate was readily apparent in all temporal snapshots (Figures 2 and 3; Figure S1; Videos S1 and S2). The highest rates of primary production occurred in seagrass, with patches of intertidal seagrass standing out as bright patches of color in the productivity snapshot maps (Figures 2 and 3; Figure S1; Videos S1 and S2). Dark areas on the maps indicated the influence of depth on seafloor primary productivity, with the lowest oxygen flux values in the subtidal channels even when the habitat class remained consistent as sand or mud.

Figure 2.

Yearly snapshot of spatial variation in seafloor primary production

Color of each 10 × 10 m grid cell is based on the yearly average of daily net oxygen flux values calculated for each 24 h period. The influence of habitat class (seagrass, bright green; sand, aqua, and mud, darker blue) can be observed, as can the effect of depth (i.e., subtidal channels in the northern part of the harbor have darker hues than shallower parts of the same sand habitat class).

Figure 3.

Seasonal snapshots (A, summer; B, autumn; C, spring; D, winter; chronologically clockwise) of spatial variation in seafloor primary production

Color of each 10 × 10 m grid cell is based on a seasonal average of daily net oxygen flux values calculated for each 24-h period. Map colors are darkest in winter (lowest rates of seabed primary production between June and August); numerical scale is the same in all images (i.e., comparable to each other and to the yearly snapshot in Figure 4).

Negative daily mean oxygen fluxes occurred in some pixels in the harbor throughout the year. However, the estuary as a whole was net autotrophic on average every day of the year, even during winter when turbidity was highest and sunlight irradiance was lowest. The minimum daily mean net productivity for the estuary as a whole occurred on June 25 2020 (16.7 mmol m⁻² d⁻¹) and the maximum was 65.1 mmol m⁻² d⁻¹ on September 9 2020. Productivity was significantly lower in winter (June–August, 44.7 ± 8.9 mmol m⁻² d⁻¹) than it was in the other three seasons (post hoc tests, p ≤ 0.0042). Productivity in spring (September–November, 55.9 ± 6.4 mmol m⁻² d⁻¹), autumn (March–May, 53.1 ± 8.2 mmol m⁻² d⁻¹), and summer (December–February, 51.5 ± 5.0 mmol m⁻² d⁻¹) did not differ significantly (p ≥ 0.1306, Figure 3).

Spatially averaged seafloor primary production rate across 12 months

The daily net productivity rate for the estuary was the highest during the months of September and March (means of 60.0 ± 5.2 and 59.9 ± 5.1, respectively), though not all monthly pairwise comparisons were significant at p ≤ 0.05 (Figure 4). The average productivity rate for July (38.6 ± 5.5) was significantly lower than all other months (p ≤ 0.0311 in all pairwise comparisons; Figure 4).

Figure 4.

Spatially-averaged daily net seafloor primary production rates for the whole Ōhiwa Harbor domain (mmol O₂ produced per m² per 24 h) averaged for each month of the year

Horizontal line in each box represents the monthly median (n = 28 for February, n = 30 or 31 for all other months). Size of box depicts interquartile range (IQR) of the daily values. Whiskers are the range between the lower quartile value minus 1.5∗IQR and the upper quartile value plus 1.5∗IQR. Outliers beyond the whiskers are denoted with dots. Primary production in months that share letters did not differ significantly (p ≥ 0.05 in post hoc tests).

Discussion

The idea that “estuaries are among the most productive habitats on Earth” appears frequently in published literature and on many governmental and non-governmental websites. Yet, there are key gaps in our understanding of the spatial and temporal dynamics of estuarine productivity at the scale of whole estuaries. This limits our ability to assess a key indicator of estuarine health and the regulating ecosystem service that primary productivity provides in supporting biodiversity and secondary and higher-order consumers in estuarine and coastal systems. The reproductive and recruitment cycles of estuarine organisms, including valued fish and shellfish, likely depend on and are often in sync with temporal cycles in productivity.^34,35 Seagrass may be essential habitat for fish and invertebrate recruitment not only for the complex three-dimensional structure it provides^36,37,38 but for its high rate of primary productivity.³⁹ The P-I curves indicated that seagrass was two to five times more productive than the other two habitat types under high light conditions, though it covered only three percent of the seascape in the Ōhiwa Harbor. Thus, even though the P-I curve for seagrass was developed from data collected outside of the Ōhiwa Harbor and in one season only (i.e., a key limitation; see STAR Methods), the harbor-wide productivity patterns such as those in Figure 4 will be relatively robust to seagrass P-I curve parameter inaccuracies.

Our study demonstrated the potentially key role of shallow estuaries in general, and areas of high seagrass cover in particular. Shallow estuaries should not be ignored just because remote assessment of primary production is challenging in these ecosystems. The very shallow average depth of the Ōhiwa Harbor (79% intertidal) and the high productivity of the shallow seafloor relative to the deep contributed to it being net productive throughout the entire year, with average monthly productivity never dropping below 38 mmol m⁻² d⁻¹. Our investigation focused on seafloor primary production estimates only. There would be additional productivity by phytoplankton in the water column, though the contribution of the water column was expected to be small relative to the benthos.

In addition to food web support, positive net primary production by seafloor habitats in shallow water estuaries will contribute dissolved oxygen to the water.^40,41 The reclamation of shallow intertidal habitats (i.e., replacement of them with land fill for various uses) has been substantial near many coastal cities and has likely resulted in the reduction of the oxygen production capacity of these systems, with the potential to contribute to recurring hypoxia. The accumulated number of coastal sites worldwide where hypoxia has been reported has increased exponentially since the 1950s⁴² Hypoxia can have a range of deleterious effects, including affecting sediment biogeochemistry, altering elemental cycling, sub-lethally stressing biota, and causing mortality (e.g., fish kills and shellfish die-offs). These can all affect amenity values and ecosystem services associated with estuaries.

The contributions of seafloor animals to oxygen dynamics were not explicitly explored during this study, though they likely play a key role. In New Zealand, sediments with more than 25% mud content are known to have fewer macrofauna than sand or seagrass,^43,44 with fewer macrofauna likely resulting in reduced rates of zoobenthic respiration. Furthermore, although muddy sediments tend to have high organic matter content (a driver of microbial remineralization), fewer macrofauna would lead to reduced bioturbation and bioirrigation, which would inhibit the transport of oxygen and thus dampen aerobic remineralization.

Our investigation also did not include the influence of a dense bed of seafloor mussels, Perna canaliculus, on oxygen dynamics in the harbor. The mussels were not present at the time of our study (June 2020 to May 2021) but have since become abundant in a patch near the harbor entrance. The expansion of mussels followed a successful restoration effort led by Ngāti Awa, the local Indigenous Māori guardians of the Ōhiwa Harbor.^45,46 This demonstrates why more detailed mapping of habitats in the Harbor is required, and why the development of P-I curves for additional habitat types beyond seagrass, sand, and mud is needed to update and improve our modeling. It also illustrates how a low rate of seabed primary production does not necessarily indicate an area of low value. The complex three-dimensional habitat created by the restored mussels is likely an oxygen sink due to its very high animal biomass, yet it is highly valued due to its provision of multiple ecosystem services (food, water purification, food web support, cultural services, and more).

Our study results are consistent with the way in which the Indigenous Māori conceptualize the natural world. Waterways are considered by the Māori to be living beings and, like human bodies, the health of all organs of the body supports overall wellness/life force/vital essence (mauri). The seafloor in the Ōhiwa Harbor exhibited pulsed rhythms of primary productivity (daily, tidal, and seasonal) and acted as a “breathing seascape” (see supplemental information). Maintenance of the “lungs” is central to the maintenance of overall ecosystem health.

Climate change is predicted to increase the loading of terrigenous sediments to estuaries (due to increased storm frequency and intensity^47,48) and sea level rise is gradually increasing estuarine water depths. The “darkening” of estuarine seabed habitats,⁴⁹ due to elevated turbidity and depth, will act to reduce seabed primary productivity. Our study, though it does not explicitly explore episodic events like storms, provides insights into the combined influences of both of these climate-related changes. Furthermore, we provide a tool that can be developed to help manage and predict estuary primary productivity—a key ecological indicator—under different scenarios of climate, sea level rise, sediment loading, and habitat loss.

Local actions in the Ōhiwa Harbor catchment will not stop sea level rise or alter the frequency of storms that deliver terrigenous sediment to estuaries. However, improvements to land-use practices (e.g., cessation of logging on steep hillslopes and reduced livestock herd sizes) and active interventions to limit sediment loading to estuaries (fencing that excludes stock from riparian edges, afforestation of catchments) will help to maintain the ecosystem service of coastal seafloor primary production moving forward. This will, in turn, support a broader range of productivity-dependent ecosystem services in the Ōhiwa Harbor, including food provision (shellfish and fish), biodiversity support, and nutrient and carbon removal. The local Regional Council and the indigenous guardians of the estuary are engaged in activities such as these to restore the mauri of the Ōhiwa Harbour.⁴⁶

Limitations of the study

The methodological caveats and assumptions of this study are specified in the STAR Methods. Our modeling was intended as heuristic (demonstrating patterns) rather than strictly numerically predictive. We acknowledge that we are using a seagrass distribution snapshot from a single point in time (based on Bulmer et al.⁵⁰), with seagrass morphology and shoot density considered invariant where present. We also treated sand and mud as homogeneous and temporally invariant habitat types. If the spatial and temporal dynamics of these habitat types were available as data layers (e.g., seagrass expansion and contraction on a seasonal and interannual basis), we would have incorporated them into our modeling. The primary purpose of our modeling was to demonstrate how habitat information, a robust seabed light model, and P-I curves can be used to predict seabed primary production at the scale of whole estuaries. With spatially and temporally up-to-date habitat data, results would be more realistic and accurate.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Andrew M. Lohrer (drew.lohrer@earthsciences.nz).

Materials availability

This study did not generate new unique reagents or plasmids or genetic lines.

Data and code availability

• •Data reported in this paper will be shared by the lead contact upon request.
• •All original code has been deposited at Figshare and is publicly available at https://figshare.com/s/91bc70a0a3ff2f612529 as of the date of publication.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We acknowledge Ngāti Awa, the local Indigenous Māori tribe and guardians of the Ōhiwa Estuary, and NIWA technicians and divers who contributed to field work: Peter Williams, David Bremner, Rod Budd, and others. This research was funded with support from Ministry for Business Innovation and Employment (MBIE) via the Sustainable Seas National Science Challenge C01X1901 (Awhi mai Awhi atu project: UOW21202), an Endeavor ‘Smart Idea’ (contract ID NIWA2501), and Strategic Science Investment Funding (SSIF) to the New Zealand Institute for Earth Science (formerly NIWA; projects FWCE2107, COME2106, CEME2501, and FPRS2605). We acknowledge additional support from Tidal Research (for R.H.B.) and Newcastle University (for F.S.).

Author contributions

A.M.L. contributed to conceptualization, investigation, formal analysis, and writing of the original draft. R.H.B. and F.S. contributed to conceptualization, investigation, formal analysis, and writing (review and editing). K.P.-B. contributed to conceptualization and investigation. J.M. contributed to investigation. I.T.M. contributed to formal analysis. O.L.-G. and G.L.P. contributed to data visualization and writing (review and editing). E.J.D. contributed to writing (original draft and review/editing). No generative AI or AI-assisted technologies were used during the preparation of this manuscript, including for data analysis, figure generation, or text editing. The exception is the graphical abstract, which was assisted by Co-pilot.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Figshare	https://info.figshare.com/user-guide/what-is-figshare/	https://figshare.com/s/91bc70a0a3ff2f612529
Software and algorithms
R software	R Core Team (2018). R: A language and environment for statistical computing. (R Foundation for Statistical Computing).	https://www.r-project.org/about.html

Method details

This study was conducted at Ōhiwa Harbour (Figure 5), a barrier-enclosed estuary located in the Bay of Plenty Region on the east coast of the North Island, New Zealand.⁵¹ It is an intertidally dominated (79%) estuary with an average depth of ∼2 m.

Figure 5.

Ōhiwa Harbor study domain (area inside black polygon) and locations of empirical data collection

Sediment grain size (for mud content) was assessed at brown and orange locations; PAR and turbidity were profiled from surface to depth at orange and blue locations. Monthly water quality monitoring sites (Bay of Plenty Regional Council turbidity data) are shown in yellow. Aquatic eddy covariance deployment locations (O₂ fluxes, PAR_I, PAR_S, and sediment characteristics) are indicated with white stars. See inset legend.

Capturing spatial and temporal variation in seafloor primary productivity in Ōhiwa Harbour involved three principal steps: (1) development of a harbour-specific seabed light model with outputs driven by variation in incident sunlight intensity, water turbidity, and water depth; (2) compilation of information on the distribution of the main three seabed types (i.e., mud, sand, seagrass) present in the harbour; and (3) generation of habitat specific seafloor irradiance-photosynthesis relationships to link light model outputs to rates of seafloor primary production.

Harbour-specific seabed light model

Incident light

In a shallow estuary such as Ōhiwa, seafloor light levels are driven primarily by the amount of light arriving at the sea surface (incident sunlight irradiance). No light is present on the seafloor at night; incident irradiance increases as the sun rises, peaks around noon, and decreases towards sunset. Incident sunlight irradiance also varies seasonally (longer daylengths in summer) and daily (e.g., depending on cloud cover and smoke/particulate matter in the air). We gathered incident photosynthetically active sunlight radiation (PAR_I) data from the closest weather station to Ōhiwa Harbour (Whakatane, 37.94815° S, 176.96773° E; ∼8 km away) on a two-hourly basis for an entire year (01-June-2020 to 31-May-2021). The data were extracted from CliFLo (New Zealand’s National Climate Database - https://cliflo.niwa.co.nz/).

Water column light attenuation

A proportion of the sunlight arriving at the sea surface attenuates as it travels downward through the water column. The attenuation of light is due to absorption, reflectance, and scattering by material in the water and by the water itself. The further the light must travel in the water, the greater the proportion that is attenuated. Therefore, water depth is an important contributor to seafloor light levels. We compiled 10 m by 10 m resolution water depth information in Ōhiwa Harbour by combining vessel-based University of Waikato bathymetric surveys and remote Bay of Plenty Regional Council-held Light Detection and Ranging (LIDAR) data. Water depth relative to chart datum (lowest astronomical tide) was available for all grid cells in Ōhiwa Harbour below the mean high water spring tide mark. Water depth also varies temporally due to the tides (12.4 h high tide low tide oscillations, 29.53 d spring tide neap tide cycles, etc.), therefore, tidal heights every two hours for the entire year were collated from local Ōhiwa Harbour tide model predictions.⁵²

Because water turbidity increases the rate at which light is attenuated as it travels through the water column, we profiled the water column (water surface to sea floor) at >40 locations throughout Ōhiwa Harbour, quantifying turbidity and light levels at a range of depths. Our dataset consisted of 1243 spatially and temporally paired measurements of the percent incident light (%PAR_I) reaching a given depth (DEP), given the water’s turbidity level at that point in space and time (measured in standard Nephelometric Turbidity Units, NTU). These data were supplemented with monthly Bay of Plenty Region Council estuarine water quality monitoring surveys conducted at two stations in Ōhiwa Harbour in 2020, for which turbidity data were available (Figure 5). This information was used to temporally adjust (using a monthly multiplication factor) interpolated maps of NTU created using empirical data from the 40 sampling sites across the estuary (Figure 5). That is, all interpolated points were adjusted upward or downward monthly, based on how many times higher or lower each month’s NTU was relative to annual average NTU during the period.

Spatio-temporal prediction of seabed light

Boosted Regression Trees (BRTs) were used to predict %PAR_I across the entire Ōhiwa Harbour spatial domain (10 m x 10 m grid cells), at a two-hourly frequency, for an entire year (1-June 2020 to 30-May-2021). BRT modelling combines many individual regression trees (models that relate a response to their predictors by recursive binary splits) and boosting (an adaptive method for combining many simple models to give improved predictive performance) to form a single ensemble model (Elith et al., 2008). Detailed descriptions of BRT methods are available in⁵³ and.⁵⁴ All statistical analyses were undertaken in R⁵⁵ using the Dismo⁵⁶ and gbm⁵⁷ packages. Relationships between %PAR_I and the three predictor variables (PAR_I, DEP, NTU) were investigated using BRT models fitted with a Gaussian error distribution, a tree complexity of 2, a learning rate of 0.01 (2,000 and 3,000 trees per model), a bag fraction of 0.60, and a 10-fold cross-validation (following best practice outlined in Elith et al., 2008). The learning rate was set to 0.01 so that the number of trees selected by cross-validation exceeded the recommended minimum of ∼1,000 trees; this yielded optimal ensembles of 2,000 to 3,000 trees across models.

Cross-validation revealed BRT model fit metrics to be high, with a mean deviance explained of 0.94 and a mean correlation of 0.97, giving us confidence to use the model for spatial predictions. At each two-hourly interval over the course of the entire year, spatial predictions of %PAR_I were made using BRT results and the spatially and temporally specific environmental data layers (PAR_I, seafloor water depth, NTU) that we had developed for Ōhiwa Harbour. The final step was converting all %PAR_I values to PAR_S with units of μmol photons m^-2 s^-1.

Habitat map

There is one species of seagrass present in New Zealand, Zostera muelleri. This short-bladed seagrass species generally occupies middle-to-lower intertidal zone soft-sediments, occasionally extending into the shallow subtidal zone. Bulmer et al.⁵⁰ recently used a combination of remote sensing and machine learning to map intertidal fringing vegetation (saltmarsh, mangroves, seagrass; mapping classification accuracy K value of 0.86) in New Zealand estuaries. The area of Ōhiwa Harbour that was classified as seagrass in the Bulmer et al.⁵⁰ study, the most recent distributional information on seagrass available for this harbour, was used to delineate “Seagrass” habitat again here (Figure 6). Stands of mangroves and saltmarsh, whose photosynthetic structures remain above the waterline at all stages of the tide, were excluded (clipped) from the spatial domain of the current study.

Figure 6.

Distribution of sand (yellow), mud (brown), and seagrass (green) in the Ōhiwa Harbor

Mud and sand refer to > or ≤25% of fine particles (≤63 μm), respectively, and distributions were estimated from inverse distance weighted interpolation of empirical data (n = 43). Seagrass distribution was satellite-derived and taken from Bulmer et al.⁵⁰

All parts of the Ōhiwa Harbour domain not defined as Seagrass were classified (simplistically) as either “Mud” or as “Sand”, with empirical sampling used to differentiate Mud and Sand (Figure 6). Seafloor sediments were collected by divers from 43 stations (brown dots in Figure 5) using 2 cm internal diameter by 2 cm deep cores. Analysis of samples for particle size distribution was performed using standard methods.^58,59 Sites with sediments comprised of >25% silt and clay particles (sized <63 μm) were classified as Mud, and sites with lower amounts of silt and clay (<25% by weight) were classified as Sand. Areas of Mud and Sand in the harbour were then mapped (interpolated between sampling points) using an inverse weighted distance algorithm available in ArcGIS PRO v.3.3.

Habitat-specific P-I curves

Photosynthesis by seagrass and MPB involves the capture of sunlight energy to fuel the conversion of carbon dioxide and water into glucose, with one oxygen molecule released for every carbon dioxide molecule converted. Photosynthesis can therefore be estimated by measuring rates of oxygen production.⁶⁰ Oxygen production generally rises as light availability increases, although the relationship is usually non-linear and asymptotic.^61,62

To link seabed light levels to rates of primary production by seagrass and MPB, we used data from sixteen Aquatic Eddy Covariance (AEC) instrument deployments to seafloor habitats in northern New Zealand, including in Ōhiwa Harbour (Table 1). AEC is a well-described and robust method of measuring rates of oxygen exchange between seafloor and overlying water.⁶³ Oxygen fluxes can be measured continuously at high frequencies (e.g., every 15 min) for up to a week (i.e., multiple day-night cycles) and paired with high frequency PAR_S measurements throughout deployments.^63,64 We used data from two published studies^40,65 and two additional deployments (Ōhiwa Harbour, Whangarei Harbour) to generate statistically significant photosynthesis-irradiance relationships for Seagrass, Mud, and Sand (Table 1). The number of data points (paired records of O₂ flux and PAR_S) used to define the photosynthesis-irradiance relationship in each habitat class was 769 (Seagrass), 1256 (Mud), and 1974 (Sand). The relationship between oxygen flux and PAR_S in each habitat was described using a standard formula for photosynthesis-irradiance (P-I) relationships (Figure 1).

Table 1.

Metadata from aquatic eddy covariance instrument deployments used to develop photosynthesis-irradiance relationships (P-I curves) for the present study

New Zealand coastal water body	AEC deployment location (Latitude, Longitude)	Date sampled (season)	Habitat classa	Maximum depthb
Whangārei Harbor	−35.823000, 174.420117	25/11/2020–27/11/2020 (spring)	seagrass	1.5c
Whangārei Harbor	−35.821823, 174.423613	25/11/2020–27/11/2020 (spring)	seagrass	1.9c
Whangārei Harbor	−35.818628, 174.427395	25/11/2020–27/11/2020 (spring)	seagrass	3.0
Whangārei Harbor	−35.814736, 174.431029	25/11/2020–27/11/2020 (spring)	seagrass	4.0
Tamaki Strait	−36.918303, 175.106033	30/11/2021–03/12/2021 (spring)	mud	5.8
Tamaki Strait	−36.90203, 175.16617	30/11/2021–03/12/2021 (spring)	sand	4.8
Tamaki Strait	−36.84496, 175.172069	30/11/2021–03/12/2021 (spring)	mud	5.2
Tamaki Strait	−36.833542, 175.193806	30/11/2021–03/12/2021 (spring)	sand	4.7
Whangārei Harbor	−35.841408, 174.530143	28/02/2022–04/03/2022 (summer)	sand	5.0
Whangārei Harbor	−35.804919, 174.483912	28/02/2022–04/03/2022 (summer)	sand	5.0
Whangārei Harbor	−35.784795, 174.446974	28/02/2022–04/03/2022 (summer)	mud	4.8
Whangārei Harbor	−35.785656, 174.390575	28/02/2022–04/03/2022 (summer)	sand	4.5
Ōhiwa Harbor	−37.98787, 177.08678	02/06/2022–08/06/2022 (winter)	mud	4.6
Ōhiwa Harbor	−37.98545, 177.10048	02/06/2022–08/06/2022 (winter)	mud	4.6
Ōhiwa Harbor	−37.99122, 177.13390	02/06/2022–08/06/2022 (winter)	sand	4.1
Ōhiwa Harbor	−37.99005, 177.11183	02/06/2022–08/06/2022 (winter)	sand	4.9

Methodological details and results from 2020 to 2021 are available in Lohrer et al. (2024) and (2025), respectively. The 2022 results are used for the first time here. Locations of the Ōhiwa Harbor deployment are shown on Figure 5.

^a“Seagrass” denotes 25%–85% cover of Zostera muelleri; “mud” (>25%) and “sand” (≤25%) refer to % by weight of particles ≤63 μm present in site sediments.

^bMaximum depth of seabed recorded at site during dates of AEC deployment.

^cIntertidal site (site exposed to air at low tide).

Methodological caveats and assumptions

Only 25% of the AEC deployments (4 of 16 sites) were conducted within Ōhiwa Harbour. AEC instruments were deployed to Ōhiwa Harbour sites in one season only (Winter) and Seagrass habitat was not sampled in this harbour. Thus, we are assuming that AEC data gathered from Seagrass, Mud, and Sand habitats in other New Zealand harbours and at other times of year are broadly similar (would generate P-I curves of identical shape) to those of Ōhiwa Harbour. We are also aware that PAR_I data gathered from a weather station ∼8 km away from Ōhiwa Harbour may not be identical to levels of PAR_I arriving in specific grid cells in Ōhiwa Harbour at the same instant in time, and we did not attempt to validate or correct the weather station data using in situ PAR measurements as part of this study. We assume a uniformly vertically mixed (non-stratified) water column, with equal levels of turbidity from surface to seabed, which is likely approximately true in Ōhiwa Harbour due to its shallow average depth and almost complete flushing during every flood-ebb tidal cycle.⁵¹ Although our seabed light modelling is spatially and temporally dynamic, we do not account for seasonal or interannual variation in the spatial distribution of the three habitat types, or event-scale variation in water turbidity. Finally, we assume that production on intertidal flats exposed at low tide is the same as production at maximum sunlight irradiance when submerged. Prior studies^28,66 suggest that intertidal seagrass and mud habitats in New Zealand, despite higher light, do not have higher rates of production when exposed to air at low tide, relative to when immersed, due to seagrass blades lying flat on top of one another in air (self-shading) and photo-inhibition (shutting down of photosystems in very high light for protective purposes).

Variation in seafloor primary production

Output from the light model (PAR_S data available for every grid cell in Ōhiwa Harbour for all two-hourly time periods between 1-June-2020 and 30-May-2021) was converted into O₂ fluxes using the appropriate P-I curves for Seagrass, Mud, and Sand. The two-hourly output for each grid cell was averaged for each 24-hour period (a full day-night cycle), yielding daily Net Primary Production (NPP) estimates. NPP is thus the integrated amount of oxygen produced by seafloor primary producers during daylight hours minus oxygen consumption by the benthic system as a whole over 24 hours (sometimes alternatively referred to as Net Ecosystem Metabolism). Rates of benthic oxygen consumption are readily apparent at night (i.e., average oxygen flux is negative when PAR_S = 0; Figure 1). Daily maps of NPP were averaged monthly and seasonally (monthly and seasonal ‘snapshots’, respectively) and compiled chronologically into a movie file (Videos S1 and S2).

Quantification and statistical analysis

Statistical analysis

Differences in daily mean seafloor NPP among months and seasons, found in Results section spatial variation in seafloor primary production and spatially averaged seafloor primary production rate across 12 months and in Figure 4, were assessed using generalized least-squares (GLS) models using the “nlme” package.⁶⁷ Daily mean NPP was used as the response variable, with either Month or Season included as a categorical fixed effect. Because observations were collected as a continuous daily time series, temporal autocorrelation was accounted for by incorporating a first-order autoregressive correlation structure (AR1) into each model. This approach allows residuals from consecutive days to be correlated and therefore avoids violating the assumption of independent observations. Monthly differences were evaluated using the model: Daily Mean NPP ∼“Month”, and seasonal differences using: Daily Mean NPP ∼“Season” with an AR(1) correlation structure fitted using date as the temporal covariate. Following model fitting, estimated marginal means were calculated using the “emmeans” package,⁶⁸ and pairwise comparisons among months and seasons were performed using Tukey-adjusted tests to control for multiple comparisons. Model parameters were estimated using restricted maximum likelihood (REML).