Persistent hot spots of CO2 and CH4 in coastal nearshore environments

Nearshore environments are typically supersaturated with the potent greenhouse gases methane and carbon dioxide, due to intense remineralization of the elevated supply of organic carbon in these systems. These environments are characterized by overlapping biogeochemical gradients and heterogeneous morphology, and the overall spatial variability in nearshore greenhouse gas concentrations remains unclear. We measured surface water partial pressures of carbon dioxide and methane synoptically with water quality parameters in the coastal Baltic Sea, covering two ice‐free seasons. The high‐frequency flow‐through data revealed sites with recurring very high partial pressures of carbon dioxide and methane (i.e., hot spots) scattered around the 50 km × 40 km study area, exceeding overall partial pressure averages by 455 μatm (CH4) and 2396 μatm (CO2). High partial pressures were linked with elevated inputs of allochthonous and autochthonous organic matter, underpinning the major role of organic enrichment of coastal environments in global carbon cycling.

. Carbon dioxide (CO 2 ) and methane (CH 4 ) emissions associated with mineralization of terrestrial (allochthonous) and autochthonous OC represent quantitatively important components of the global C cycle (Cotovicz et al. 2021;Downing et al. 2021;Rosentreter et al. 2021).Specifically, previous studies show that CO 2 and CH 4 fluxes at the water-atmosphere interface are strongly associated with phytoplankton biomass (Riebesell et al. 1993;Bizic 2021) and terrestrial organic matter loading (Amaral et al. 2021).As a result of elevated organic matter inputs to these systems, several recent studies have highlighted the importance of nearshore aquatic environments as hot spots of aquatic C cycling (Borges et al. 2016(Borges et al. , 2018;;Maier et al. 2021;Osterholz et al. 2021).The pronounced seasonality of high-latitude ecosystems adds further complexity to quantification of biogeochemical processes, leading to high variability of concentration and flux estimates of CO 2 and CH 4 from coastal environments.Accordingly, the partial pressures of CO 2 and CH 4 as well the associated fluxes at the water-atmosphere interface display pronounced diurnal (Honkanen et al. 2021;Roth et al. 2022) and seasonal variation (Borges et al. 2018).For example, methane emissions ranging between 1 and 5000 μmol m À2 d À1 were observed in a coastal bay adjacent to the Atlantic Ocean (Burgos et al. 2018).Such huge variability within relatively confined areas has resulted in nearshore environments, estuaries and inland waters being currently the most important source of uncertainty on the global methane budget (Saunois et al. 2016).Therefore, a key challenge of measuring and modeling coastal environments is determining the spatial scales needed to represent processes and systems with adequate accuracy and relative to the hot spots that characterize the system (Ward et al. 2020).
Different approaches have been used to measure greenhouse gas fluxes across the air-water interface.The most commonly used ones are the boundary layer method, eddy covariance flux measurements and the floating chamber method; all of which have their specific strengths and shortcomings (Erkkilä et al. 2018).A key constraint for all of them is the limited spatial resolution, often meaning measurements from a single, fixed location or from multiple locations during labor-intensive field sampling campaigns.Continuous in situ measurements, on the other hand, combine the sampling frequency of fixed monitoring stations and the mobility of shipborne sampling campaigns (Crawford et al. 2015;Canning et al. 2021;Jacobs et al. 2021).We used this high-resolution approach in this study primarily to characterize environmental heterogeneity at the system level.We determined CO 2 and CH 4 partial pressures (μatm) and fluxes (mmol m À2 d À1 ) and measured associated environmental variables in an extensive and highly heterogeneous coastal system in nine field campaigns spanning two consecutive growing seasons.To derive overall patterns, we quantified the role of key environmental drivers (allochthonous organic matter and phytoplankton biomass) as well as diurnal and seasonal variability for the partial pressures and fluxes and assessed the prevalence of concentrations hot spots in the area.

Study area
The study was carried out within a 2000 km 2 coastal area around the SW tip of Finland (Scheinin and Asmala 2020).The surface water (0.5 m depth) data were collected along an 830-km-long route (Fig. 1), where water depth varied between 0.7 and 61 m, with a mean depth of 9.9 m.The fixed route was surveyed monthly, on nine 4-d campaigns during the ice-free seasons in 2020-2021.The surveyed area can be considered as a model system for resolving coastal variability of greenhouse Fig. 1.Spatial variation in the surface water partial pressures of CH 4 and CO 2 on the SW coast of Finland in April 2020 (campaign 1 out of 9).Interpolation carried out using the diffusion kernel method, with 1 nautical mile buffer and applying standard deviation as stretch type in the symbology.Since the CH 4 data could not be corrected for time lag in the sensor response, they represent a running mean with a tendency for tailing in the direction of the survey vessel (red route).
gas (GHG) fluxes, for example, as the spatial heterogeneity is very high due to the complex bathymetry and geomorphology of the archipelago system.This area has also previously been observed to show pronounced patchiness in pelagic phytoplankton abundance, showcasing how biogeochemical processes are intrinsically linked with the physical features of the environment (Scheinin and Asmala 2020).The area is characterized by high freshwater inputs from multiple sources, which deliver relatively high quantities of allochthonous C (Asmala et al. 2016(Asmala et al. , 2019) ) and nutrients to the system, leading to high eutrophication status and water column OC concentration for the most parts of the area (Fleming-Lehtinen et al. 2015;HELCOM 2018).Salinity ranged from 0 in the river mouths to 6.8 in outer archipelago, and the overall mean salinity of the observations was 5.0.

Flow-through system
The spatially detailed and extensive in situ data were collected by a flow-through measurement system equipped with multiple optical and electrical sensors (Asmala and Scheinin 2022).A detailed description of the flow-through system is published by Scheinin and Asmala (2020).In short, water was led through a debubbler before entering a chamber enclosing a multiprobe (EXO2, Xylem Inc.) equipped with sensors for temperature, conductivity, dissolved oxygen, pH, humic-like fluorescent dissolved organic matter (FDOM), turbidity, chlorophyll a (chl a), and phycocyanin.Air pressure (at +1.0 m) was recorded by an interconnected EXO Handheld unit (Xylem Inc.).After the multiprobe, the water flow was divided in order to supply the gas sensors at the rate of 2.4 L min À1 .Partial pressures of dissolved CO 2 and CH 4 in water were recorded continuously with a CO2-Pro CV (Pro-Oceanus Systems Inc.; calibration range 0-20,000 ppm, accuracy AE 0.5% of measured value, resolution 0.01 ppm, equilibration time [t63] 50 s) and Mini CH 4 (calibration range 0-10,000 ppm, accuracy AE 3%, resolution 0.1% of max range, equilibration time [t63] 480 s), using infrared detection (Pro-Oceanus Systems Inc.).Time, GPS position and air pressure were logged continuously by the EXO Handheld unit, connected to the sonde with a data transmission cable.All data were recorded at 5-s intervals.The maximum distance between observations was kept below 60 m.All sensors were calibrated following manufacturers' guidelines.

Data processing and statistical analyses
Post-processing procedures of the sensor data are described in detail in Supplementary Information.In summary, after correcting for hydraulic lag (Scheinin and Asmala 2020), accounting for sensor drift (Fietzek et al. 2014) and converting the sensor output from ppm to μatm, the high-frequency data from the sensors were quality controlled using the procedures for biogeochemical ferrybox data (Linders et al. 2017).In addition, the CO 2 data were corrected for time-lag (Miloshevich et al. 2004; Supporting Information Fig. S2).Time lag corrections could not be done for CH 4 data mainly due to a comparably long sensor equilibration time.Observations were then binned to 60 Â 60 m spatial cells, which is roughly equal to the maximum distance between observations.Since the CH 4 data could not be corrected for time lag, they represent a running mean with a tendency for tailing in the direction of the survey vessel (Supporting Information Figs.S1, S6, S7), and thus, exceed the size of the spatial cells in terms of resolution.For removing obvious outliers in the sensor data, 99 th percentile from chl a observations was excluded from the analysis, and 1 st and 99 th percentiles from FDOM.After post-processing, the 9 campaigns resulted in a total of 117,715 observations (Supporting Information Fig. S1).
We used FDOM as a proxy for allochthonous OC loading and chl a concentration as a proxy for autochthonous OC loading (Pace et al. 2021).The temporal variables "time of day" and "day of year" were used to resolve diurnal and seasonal variability in the observed partial pressures.The FDOM data were corrected for variation in temperature and turbidity according to Snyder et al. (2018).The chl a data were calibrated following Scheinin and Asmala (2020), while the partial pressures of CO 2 and CH 4 were converted to concentrations according to Weiss (1974) and Wiesenburg and Guinasso (1979), respectively.Prior to further statistical analyses, transformations were employed to stabilize the variance, reduce the influence of observations in the tails of the empirical distributions and to produce a more even spread of the observations.CO 2 , CH 4 , chl a, and FDOM concentrations were strongly right-skewed suggesting that the observations should be log-transformed.
Generalized additive models (GAMs; Hastie and Tibshirani 1990) were used for illustrating how the partial pressures of CO 2 and CH 4 were driven by allochthonous and autochthonous organic loading (reflected by FDOM and chl a), along with diurnal (time of day) and seasonal (day of year) variation (Supporting Information Fig. S3).The analyses were done with the R package mgcv (Wood 2012).The degrees of freedom for the splines, a proxy measure of its curvature, were selected by general cross-validation, however, with a maximum of 6 of freedom imposed to reduce the potential deformation of the fitted curve.Smooth terms were extracted and visualized using the R package gratia (Simpson and Singmann 2018; Supporting Information Fig. S4).
To identify hot spots, we first calculated the annual mean for each of the four subareas (Supporting Information Fig. S5) within our sampling area for both CH 4 and CO 2 .Mean values of binned observations from 60 m Â 60 m spatial cells were used, and for every observation the deviation from the annual mean was calculated (analysis carried out separately for CO 2 and CH 4 ).We used two a priori criteria for hot spot classification: (1) number of observations (n OBS ) had to be at least 7 (maximum n = 9), and (2) number of observations exceeding the annual mean by 1 standard deviation unit had to be at least n OBS À 1.For a given grid cell, this equals to a threshold of approximately 85% of observations needing to exceed annual mean, when at least seven observations are collected.The mean deviation of the hot spot from the annual, area-specific mean value was also calculated.One-way ANOVA was used to compare differences among hot spot groups (CO 2 hot spot, CH 4 hot spot, no hot spot), and Tukey's range test was used to identify groups with significant difference between each other.
The Pearson correlation coefficient was computed to assess the linear relationship between CO 2 and CH 4 partial pressures.A simple linear regression model was used to determine relationship of CO 2 and CH 4 with depth.The depth data were obtained from the bathymetric charts by the VELMU program of the Finnish Environment Institute (SYKE).
Flux of CO 2 and CH 4 to the atmosphere was calculated individually for both gases with a two-layer model (Liss and Slater 1974), and detailed description of the flux calculations is presented in Supporting Information (Table S1).

Spatial variation and seasonality in CO 2 and CH 4 observations
The partial pressures of dissolved CH 4 ( pCH 4 ) and CO 2 ( pCO 2 ) ranged from non-detectable to 4670 μatm and from 4.52 to 13,100 μatm, respectively (Table 1).Both maxima exceed those previously observed in the surface waters of coastal Baltic Sea by an order of magnitude (Humborg et al. 2019;Myllykangas et al. 2020), while the averages for CH 4 (386 μatm) and CO 2 (442 μatm) correspond to findings from other coastal systems (Middelburg et al. 2002;Borges et al. 2006).As for the campaign-specific averages, the lowest CH 4 values were observed in May 2020 (237 μatm) and the lowest CO 2 values in April 2020 (323 μatm), while the highest values for both dissolved gases were observed in October 2020, 856 μatm for CH 4 and 756 μatm for CO 2 .The CO 2 undersaturation (compared to atmospheric equilibrium) in spring balanced during summer and became oversaturated in autumn.This temporal pattern reflects the annual dynamics of intense phytoplankton growth during spring, later shifting to senescence of aquatic primary producers and respiration of accumulated OC toward the autumn (Parard et al. 2016;Schneider and Müller 2018).However, the highest maximum and regional average CH 4 values were observed in late autumn, coinciding with the highest CO 2 values during autumn senescence and intense system respiration.During each sampling occasion, the surface water pCH 4 and pCO 2 displayed spatial variation of several orders of magnitude (Table 1; Fig. 1).There was a weak positive correlation between pCO 2 and pCH 4 , r (160,340) = 0.198, p < 0.001.This poorly constrained CH 4 : CO 2 relationship shows that respiration processes of OC are decoupled on the ecosystem level (Stanley et al. 2016).Furthermore, primary production by phytoplankton has opposite impacts on pCO 2 and pCH 4 , as indicated by the additive contributions of chl a (Fig. 2).Surface water partial pressures of CO 2 and CH 4 were not linked with total depth at the sampling locations, as it explained only 6% and 1% of the observed variability, respectively.

Drivers of pCO 2 and pCH 4 in coastal surface waters
The relationships between the four studied drivers and GHG partial pressures were very different for CO 2 and CH 4 (Fig. 2), underlining the profoundly different biogeochemical origin of these two gases.High CO 2 partial pressures were associated with high FDOM values (> 30 quinine sulfate unit [QSU]), which is indicative of the significance of mineralization of terrestrial organic matter in CO 2 emissions from nearshore environments.Most likely driven by inorganic carbon uptake by primary producers, phytoplankton biomass was observed to have a negative effect on pCO 2 at chl a concentrations higher than 8 μg L À1 .Although the pCO 2 followed a typical diurnally oscillating pattern (Honkanen et al. 2021), this effect was small compared to other drivers.We also observed strong seasonality in pCO 2 partial pressures, suggesting high CO 2 uptake during spring bloom and high mineralization rates in autumn, resulting from high primary productivity in summer.The partial pressures of methane were most strongly driven by seasonality, with very high values associated with increasing amounts of decaying biomass following autumn senescence (Elovaara et al. 2020).High chl a concentrations (> 20 μg L À1 ) were linked with elevated pCH 4 , potentially caused by direct methane production by pelagic phytoplankton during photosynthesis (Bizic 2021).Our results show clear diurnal variability in pCH 4 , potentially caused by wind-driven ventilation of CH 4 build-up (Staehr et al. 2018) after nighttime calmer conditions, typical for land-sea breeze systems (Miller et al. 2003), and by diurnal stratification, influencing convective mixing (Podgrajsek et al. 2014).Although the association between FDOM and pCH 4 was comparably weak, the elevated partial pressures at intermediate (15-25 QSU) and high (> 70 QSU) FDOM concentrations point toward two different origins for the observed methane, depositional locations further offshore and shallow nearshore environments receiving considerable runoff from land.

Occurrence of GHG hot spots in the coastal environment
We identified areas with disproportionately high pCH4 or pCO2 compared to the surrounding area, that is, hot spots (McClain et al. 2003).More specifically, these locations act as permanent control points (sensu Bernhardt et al. 2017), where continuous delivery of reactants and appropriate environmental conditions sustain high rates of OC mineralization into methane or carbon dioxide relative to the surrounding aquatic environment.Of all individual observation sites (n = 19,039), 1.1% were classified as CH 4 hot spots and 1.0% as CO 2 hot spots (Fig. 3).On average, partial pressures from hot spots exceeded overall partial pressure averages by 455 μatm (CH 4 ) and 2396 μatm (CO 2 ).CO 2 hot spots were observed almost exclusively at river mouths or very close to the shoreline, resulting from the lateral flux of reactive OC that transits through inland waters and across the riparian zone (Regnier et al. 2013).CH 4 hot spots were observed in two distinct environments, in comparably deep depositional locations and in shallower semi-enclosed embayments.Both environments are characterized by relatively high accumulation of organic material, sustaining high methanogenesis (Wallenius et al. 2021;Vizza et al. 2022).Locations without hot spots showed significantly lower chl a concentrations and humic-like organic matter compared to CH 4 and CO 2 hot spots (Fig. 4), underscoring the significance of OC inputs to occurrence of hot spots.

Implications for coastal carbon cycling
The scales of the spatial variability in coastal biogeochemical processes are influencing carbon budgets from local to  global scale (Ward et al. 2020).Accordingly, inadequate characterization of coastal spatial variability has been estimated to introduce errors up to 60-70% in regional emission models of CH 4 and CO 2 (Kodovska et al. 2016;Roth et al. 2022).Our results show that, as a result of patchy supply of reactants (autochthonous and allochthonous organic matter), the relevant spatial scale addressing hot spots of GHG concentrations in coastal nearshore environments is from tens of meters to kilometers, indicating the spatial resolution and extent needed for ecosystem models and process studies.
Our observations provide mechanistic underpinning for several other observations, including strong seasonal dependence of CO 2 and CH 4 emissions (Borges et al. 2006;Roth et al. 2022) and estuaries as hot spots of CO 2 emissions (Frankignoulle et al. 1998), CH 4 emissions associated to phytoplankton primary production (Bizic 2021) and rapid mineralization of terrestrial OC in coastal environments (Amaral et al. 2021).Our results underline that ongoing organic enrichment is an important general driver of elevated GHG concentrations from coastal ecosystems.Thus, reductions in direct allochthonous OC inputs and nutrient-driven autochthonous OC production should lead to decreases in GHG emissions from these systems.We show that large areas of coastal waters are frequently oversaturated with CH 4 throughout the annual cycle, indicating that coastal systems like these are a consistent source of methane, though the magnitude remains uncertain.We also show that a given coastal aquatic system can act both as a sink and as a source of CO 2 , depending on the spatial scale and time.Extrapolation of greenhouse gas fluxes from  partial pressure hot spots, CO 2 partial pressure hot spots or no hot spots.Lower and upper ends of boxes indicate the interquartile range, whiskers the lowest and the highest values within the range of 1.5 IQR, and thick black line indicates median value.Mean value for each group (indicated with black diamond) is also given.The hot spot groups were significantly different from no hot spot group for every variable (pairwise Tukey's range test; p < 0.001), except CH 4 did not differ significantly from no hot spots in regard to salinity.Number of spatial grid cells per group: CH 4 hot spot-220 (1.2%), CO 2 hot spot-189 (1.0%), and no hot spot-18,630 (98%).
offshore locations to nearshore environments is likely to lead to underestimation of their biogeochemical significance, unless the high spatial variability and the nonlinear relationship between gas emissions and OC loading is considered.

Fig. 2 .
Fig. 2. Smooth terms for the GAM for partial pressures of methane and carbon dioxide.The y-axis is the additive contribution of the spline to the fitted model over the range of the covariate.The dashed line indicates 95% confidence interval.Time of day refers to diurnal variation and Day of year refers to seasonal variation.

Fig. 3 .
Fig. 3. Observed CH 4 and CO 2 partial pressure hot spots in the study region 2020-2021.Colors represent different environment types.

Fig. 4 .
Fig. 4. Differences in (a) chl a concentration, (b) humic-like organic matter fluorescence, and (c) salinity among sampling locations classified as CH 4

Table 1 .
Summary of CH 4 and CO 2 observations for each sampling campaign in 2020 and 2021.Number of observations, and mean value AE 1 SD, range (min-max) of the partial pressures and fluxes is given.Positive flux value indicates emission from water to atmosphere.Total average values and total number of observations are presented with bold typeface on the bottom row.