Integrating In Situ Measurements and Satellite Imagery for Coastal Physical and Biological Analysis in the Cape Fear Coastal Region

Highlights

What are the main findings?

• Transects from shipboard tow-yo profiles and satellite imagery are used to characterize a small plume at the mouth of a coastal inlet.
• The plume and shipboard sampling were in shallow (<10 m) and relatively clear waters, such that satellite imagery may be contaminated by benthic reflectance.

What are the implications of the main findings?

• Cloud cover during in situ sampling prevented contemporaneous matchup analysis, but a comparison of multiple satellite and in situ datasets elucidates discoveries enabled by various sensors.
• Matchups across multiple satellite and in situ datasets are poor in this region, suggesting the need for continued in situ sampling close to shore where ocean color remote sensing is known to suffer from challenges including land adjacency, bottom reflectance, and optical complexity.

Abstract

Monitoring coastal and estuarine dynamics is crucial for understanding coupled physical, biogeochemical, and human impacts on coastal waters. Motivated by the availability of high spatial resolution ocean color data from the proof-of-concept SeaHawk-HawkEye ocean color CubeSat, this study assesses the capabilities and limitations of satellite remote sensing in capturing shallow water (<10 m) coastal dynamics by integrating in situ measurements with satellite imagery. A Sea Sciences Acrobat collected detailed transects at the mouth of Masonboro Inlet (Wilmington, NC, USA), with “tow-yo” style profiles from the surface to 10 m. It measured conductivity, temperature, and depth (CTD), chlorophyll a (Chl a), turbidity, and dissolved oxygen. Satellite data from SeaHawk-HawkEye, Aqua-MODIS, and Sentinel 3A/3B-OLCI provided extensive spatial coverage, revealing surface-level physical/biological interactions, but were only available 48 h after in situ sampling due to cloud cover during field sampling. Tow-yo profiles elucidated a three-dimensional phytoplankton plume, the spatial extent of which we further characterize with satellite imagery, demonstrating the value of integrating in situ and satellite data. A spatial matchup comparison between data from each satellite and the in situ sensor package revealed significant discrepancies across all satellite sensors analyzed, attributed to differences in sensor resolution, atmospheric correction approaches, and proximity to land/benthos. This study emphasizes key challenges with study design and data interpretation in dynamic nearshore environments. In particular, results suggest that meaningful comparisons of satellite vs. in situ observations in such systems require near-synchronous sampling, careful consideration of spatial scale, and improved characterization of optical complexity.

1. Introduction

Coastal regions, comprising only 7% of the ocean’s surface area, disproportionately influence marine biogeochemical cycles [1]. These zones channel various forms of allochthonous carbon from land to ocean and are important sites for autochthonous carbon production by marine organisms [2]. Coastal zones often experience upwelling, bringing nutrient-rich deep waters to the surface, enhancing primary productivity, and supporting diverse marine ecosystems [3]. However, nutrient influx from rivers, agricultural runoff, and urban wastewater can exacerbate eutrophication, leading to oxygen-depleted waters and reduced biodiversity [2]. In particular, the Cape Fear River drains a substantial portion of North Carolina’s nutrient- and organic matter-rich surface water into the coastal ocean. Concentrated Animal Feeding Operations sprinkled across the Cape Fear River watershed introduce substantial nutrient loads to the river system [4] and the highly developed Wilmington region contributes additional surface water runoff with high nutrient concentrations. Furthermore, the Cape Fear River Estuary (CFRE), North Carolina, USA, supports diverse species such as the flounder, eastern oyster, blue crab, and migratory birds, serving as essential nursery grounds and habitats [5,6,7,8]. The combination of dense human settlement, coastal resources, and intensive livestock farming [4] highlights the complex relationship between human activity and environmental health in this region.

Of particular interest in this work is the ability to observe biogeochemical processes using a combination of in situ and remote sensing tools and to further evaluate the proof-of-concept SeaHawk-HawkEye mission—the first operational ocean color CubeSat mission [9]. The HawkEye sensor offers a best-in-class (among dedicated ocean color missions) spatial resolution of 130 m. While only a single SeaHawk satellite has been launched to date, CubeSats may offer distinct advantages in dynamic coastal settings due to the fact that their lower cost of design, build, and launch could facilitate their use in constellations. However, importantly, SeaHawk had a lengthy nominal 18-day revisit period at launch (growing even longer throughout its operational lifetime) and suffered “ghosting” issues in its NIR bands, resulting in its reliance on a single aerosol model for atmospheric correction [9] and likely complicating its use in optically complex waters.

The foundational principle of ocean color (OC) remote sensing lies in the relationship between the intensity and spectral distribution of visible light reflected from the sea surface and the underlying biogeochemical properties of the water column [10]. Chlorophyll a (Chl a), the primary photosynthetic pigment in phytoplankton, serves as a key proxy for marine productivity and as an indicator of ecosystem health. Suspended particulate matter (SPM) in coastal waters plays a critical role in modulating light penetration, directly influencing phytoplankton productivity [11]. Estimating particulate organic carbon (POC) concentrations is essential for understanding the oceanic carbon cycle [12].

This study also examines the light attenuation coefficient at the 490 nm wavelength, K_d,490, as an indicator of water turbidity [13]. Lower K_d,490 values indicate clearer water, while higher values indicate more turbid water; the parameter also varies with factors such as sun angle, sky conditions, sea surface roughness, and light absorption and scattering [14]. The inverse of K_d,490 provides a measure of the first optical depth, corresponding to the 37% light level [15,16]. If the estimated optical depth is deeper than the seafloor in that location, bottom reflectance may significantly contaminate remote sensing reflectances at the corresponding wavelength [15]. Conversely, if the first optical depth is shallower than the depth of in situ measurement, satellite data may not fully capture the water column’s vertical variability [16].

Although ocean color remote sensing excels in open ocean (i.e., “Case 1”) waters, accuracy declines in coastal and estuarine (“Case 2”) waters due to the increased presence of optically active substances like colored dissolved organic matter and suspended sediments that may not covary with Chl a concentration [17,18]. This variability complicates the application of standard bio-optical algorithms designed for open ocean conditions [19]. Integrating remote sensing with frequent in situ validation is crucial for advancing biogeochemical monitoring and addressing the unique challenges of coastal environments [19,20,21]. This study aims to evaluate the challenges of integrating vertically resolved in situ measurements with satellite observations in a dynamic coastal environment, with a particular focus on identifying limitations in study design, temporal alignment, and optical complexity that influence satellite–in situ comparisons. This work represents an initial observational effort to better understand the constraints of coastal ocean color validation and to inform the design of future studies in similarly dynamic nearshore environments.

2. Materials and Methods

The study was conducted in the CFRE in southeastern North Carolina, USA, which connects inland water bodies to the Atlantic Ocean through the Cape Fear coastal region and Onslow Bay (Figure 1A,B). We focused on the discharge offshore of the Masonboro Inlet, which is ≈300 m wide, connects the Intracoastal Waterway to the Atlantic Ocean, and separates Masonboro Island from Wrightsville Beach, North Carolina (Figure 1C). Masonboro Inlet serves as an important federally maintained navigation channel for the region and is bounded by a jetty on both its north and south side [22]. The inlet’s tidal prism has been computed to be ~23 × 10⁶ m³, representing substantial water flow and mixing potential on tidal timescales [23].

On 5 May 2023, we collected in situ measurements on board the R/V Cape Fear using the Acrobat platform (Sea Sciences Inc., Arlington, MA, USA; Figure 2A). The Acrobat is a versatile marine research tool, accommodating commercial sensors and real-time data acquisition. Most notably, the Acrobat is capable of conducting “tow-yo” profiles: that is, as the research vessel transits, the Acrobat zig-zags between user-defined depths, thus enabling multi-dimensional data collection. We integrated a comprehensive suite of sensors, including the Sea-Bird (Bellevue, WA, USA) SBE 25plus Sealogger CTD, SBE 4 Conductivity sensor, SBE 43 Oxygen sensor, Seapoint (Exeter, NH, USA) Chlorophyll Fluorometer, and Seapoint Turbidity Meter. All sensors were calibrated according to manufacturers’ protocols prior to deployment, with calibration dates as follows: CTD pressure sensor (6 September 2022), conductivity sensor (2 September 2022), oxygen sensor (7 September 2022), chlorophyll fluorometer (1 December 2022), and turbidity meter (26 September 2022) [24,25,26,27,28,29,30].

On 5 May 2023, we collected measurements along seven transects, each spanning ≈5 km alongshore and extending 2.5–4.5 km offshore (Figure 1C). The placement and orientation of the transects were selected to capture the primary physical and biogeochemical gradients associated with flow through the Masonboro Inlet and adjacent coastal waters. This region is characterized by strong tidal exchange and localized outflow from the estuary, which can generate variable plume structures and tracer transport patterns. Transects were oriented approximately parallel to the coastline and spanned both the deeper channelized regions associated with inlet outflow and adjacent shallower areas to assess how bathymetry and hydrodynamic forcing influence vertical structure and surface expressions detectable by satellite sensors. This design was intended to sample across expected gradients in turbidity, Chl a concentration, and water mass structure driven by tidal forcing and estuarine discharge.

CTD, chlorophyll a, turbidity, and dissolved oxygen data were continuously recorded from the surface to depths of ≈10 m. Following data collection, the datasets were downloaded using Sea-Bird Scientific Seasoft software (V2) and processed for initial quality control, including checks for timing consistency and value range validation according to NOAA’s Information Quality Guidelines [31]. A Seapoint Ultraviolet Fluorometer was also integrated for fluorescent dissolved organic matter data collection, but the sensor malfunctioned for unknown reasons and, as a result, these data are not reported here.

Data were visualized with contour plots that linearly interpolated each variable in two dimensions—across depth and along transect distance—using a custom Python script and the griddata function from Python’s scipy.interpolate. These contour plots represent variations in measured parameters both vertically and horizontally along each transect. To assess variability over depth, we also generated box plots for several depth intervals (0–4 m, 4–7 m, 7–10 m, and 0–10 m). Bathymetry data from GEBCO’s global gridded bathymetric datasets were incorporated to contextualize our observations [32].

We collected OC satellite data from Aqua-MODIS, Sentinel 3A and 3B-OLCI, and SeaHawk-HawkEye (V2018 data release, following [33]) for 5–7 May 2023, targeting the Lower CFRE and the vicinity of Masonboro Inlet. We downloaded all satellite datasets via NASA’s OB.DAAC Direct Data Access (DDA) interface (https://oceandata.sci.gsfc.nasa.gov/directdataaccess/, accessed on 10 August 2025). Landsat 8/9-OLI and Sentinel 2A/B-MSI were excluded because they did not acquire imagery over the study region during our sampling period.

Launched in 2018, the SeaHawk-HawkEye mission employs the HawkEye OC sensor on board a 3U CubeSat [9,33,34,35]. Despite its compact size, HawkEye’s performance has been shown to rival that of other OC radiometers [33] while achieving a spatial resolution of 130 m, significantly finer than the ~1000 m resolution of SeaWiFS or MODIS-Aqua and the 300 m resolution of Sentinel 3A/B-OLCI. HawkEye bridges the gap between high-spectral-quality sensors like SeaWiFS, MODIS, and OLCI, and high-spatial-resolution imagers like OLI and MSI. This balance was intended to enhance the detection of small-scale bio-optical gradients that are critical to nearshore OC research.

Custom processing began with downloading NASA’s Level-2 (L2) ocean color data, which include remote sensing reflectance (R_rs(λ)) and other geophysical products at the native spatial resolution of each sensor. Using SeaDAS v9.2.0, we applied quality control flags to ensure data reliability, removing pixels affected by failed atmospheric correction (ATMFAIL), land pixels (LAND), high or saturated radiance (HILT), excessive satellite sensor zenith angles (HISATZEN), potential stray light (STRAYLIGHT), cloud or ice contamination (CLDICE), presence of coccolithophores (COCCOLITH), low water-leaving radiance (LOWLW), Chl a algorithm failures (CHLFAIL), and navigation errors (NAVWARN). We used L2 products for quantitative analysis and Level-3 (L3) products—which consist of spatially binned and temporally composited data mapped onto a uniform grid—for visualizing broader spatial patterns in Chl a and K_d,490.

Unfortunately, cloud coverage on the day of in situ sampling, 5 May 2023, rendered that day’s satellite data unusable (Figure 2B). To enable an in situ vs. remotely sensed data comparison, we relaxed our ideal temporal constraint from same-hour matchups (e.g., [36] and references therein) to include imagery captured within 24–48 h of in situ sampling; satellite vs. in situ sampling times are described in

Table 1. Satellite imagery details captured by the satellite sensors.

Satellite-Sensor	In Situ Sampling Timeframe (EDT)	Satellite Image Capture Time (EDT)	Post-In Situ Sampling Time (Hours)
SeaHawk-HawkEye	5 May, 10:00–15:00	7 May, 11:09:55	44
Sentinel 3A-OLCI	5 May, 10:00–15:00	7 May, 11:34:21	44.5
Sentinel 3B-OLCI	5 May, 10:00–15:00	7 May, 10:55:11	44
Aqua-MODIS	5 May, 10:00–15:00	7 May, 14:45:01	47.5

. The especially dynamic nature of inlets—strongly influenced by tidal, river/creek outflow, and wind changes—necessitates testing temporal stability assumptions in our data. While past studies suggest satisfactory matchup analyses with modest temporal mismatch between satellite and in situ data collection [37], in dynamic tidal ecosystems, the window is likely shorter [20]. We benchmarked temporal variability in the vicinity of our mission by analyzing the 24 h Chl a change in Sentinel 3A-OLCI data from 6 to 7 May.

We contrasted in situ Chl a concentrations measured in the upper 10 m of the water column with satellite-derived Chl a estimates (Figure 3A). Ocean color radiometry inherently detects an integrated signal over primarily the first optical depth of the water column; that is, most photons that reach a satellite-borne radiometer are reflected from depths up to the point where light availability has markedly decreased. This weighted averaging process is influenced by the water’s vertically variable inherent optical properties (IOPs), which dictate how light is absorbed and scattered by substances within the water [16,38].

We performed satellite–in situ matchups using exact single-pixel comparisons by identifying the nearest Level-2 satellite pixel to each in situ sampling location based on geographic coordinates. We applied no spatial averaging (e.g., multi-pixel windows) and treated each sensor independently to preserve its native spatial resolution. Within this framework, we averaged in situ Chl a concentrations over depth intervals of 0–4 m, 4–7 m, 7–10 m, and the full 0–10 m range (Figure 3B) to assess depth dependence of the matchup quality.

To evaluate the accuracy of satellite sensors in measuring Chl a concentration, we applied five standard metrics: the coefficient of determination (R²), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Bias, and Mean Absolute Error (MAE) [39]. These metrics were calculated based on the differences between satellite-estimated (Q_est,i) and in situ-measured (Q_mea,i) Chl a values, according to the following formulas, where ‘N’ represents the total number of matched data points:

$${\mathrm{R}}^2=1-{\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{Q}}_{\mathrm{e}\mathrm{s}\mathrm{t},\mathrm{i}}-{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}\right)}^2}{{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}-{\overline{\mathrm{Q}}}_{\mathrm{m}\mathrm{e}\mathrm{a}}\right)}^2}},$$

(1)

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{N}}}{\left({\mathrm{Q}}_{\mathrm{e}\mathrm{s}\mathrm{t},\mathrm{i}}-{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}\right)}^2}{\mathrm{N}}}},$$

(2)

$$\mathrm{M}\mathrm{A}\mathrm{P}\mathrm{E}={\displaystyle \frac{1}{\mathrm{N}}}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{N}}}\left|{\displaystyle \frac{{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}-{\mathrm{Q}}_{\mathrm{e}\mathrm{s}\mathrm{t},\mathrm{i}}}{{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}}}\right|\times 100\%,$$

(3)

$$\mathrm{B}\mathrm{i}\mathrm{a}\mathrm{s}={\displaystyle \frac{1}{\mathrm{N}}}\sum _{\mathrm{i}=1}^{\mathrm{N}}\left({\mathrm{Q}}_{\mathrm{e}\mathrm{s}\mathrm{t},\mathrm{i}}-{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}\right),$$

(4)

$$\mathrm{M}\mathrm{A}\mathrm{E}={\displaystyle \frac{1}{\mathrm{N}}}\sum _{\mathrm{i}=1}^{\mathrm{N}}\left|{\mathrm{Q}}_{\mathrm{m}\mathrm{e}\mathrm{a},\mathrm{i}}-{\mathrm{Q}}_{\mathrm{e}\mathrm{s}\mathrm{t},\mathrm{i}}\right|.$$

(5)

We then compared these metrics across satellite sensors and depth ranges. Although statistical comparisons of Chl a are often performed using log₁₀-transformed concentrations due to the typical log-normal distribution of oceanic chlorophyll [40,41], all metrics here were computed in linear space, considering the range of variation in these coastal waters is much smaller than the orders of magnitude observed in global ocean validations [41], to preserve absolute concentration differences and facilitate interpretation in this optically complex, shallow coastal environment. The routines used for processing and analyzing the satellite and in situ datasets are publicly available at https://github.com/COAST-Lab/HawkEye_Evaluation (accessed 31 January 2026) and are summarized in the workflow diagram presented in Figure S1. All analyses were conducted in Python 3.13.5, incorporating cmocean v4.0.3 for perceptually uniform oceanographic colormaps [42], and SciPy v1.16.0 [43] for interpolation and statistical operations. Satellite ocean color products were processed using SeaDAS v9.2.0 and the associated OCSSW (Ocean Color Scientific Software, V2024.5) processing system, maintained by NASA’s Ocean Biology Distributed Active Archive Center [44].

3. Results

The study site demonstrated temporal variability influenced by several environmental factors. Over the observation period from 5 to 7 May, wind speeds increased notably, ranging from 0.03 m/s to 9.46 m/s, indicating potential surface water mixing and variable distribution patterns (Figure 4C). Concurrently, there was a subtle decrease in Chl a concentration, with an overall average 24 h change within the study region from 6 to 7 May of −0.15 µg/L, but some pixels experienced more extreme swings up to ±3.21 µg/L (Figure 4A), reflecting advection of water masses and biological responses to changing environmental conditions. The tidal range also increased throughout the week. The recorded tidal ranges were 1.563 m on 5 May, 1.632 m on 6 May, and 1.729 m on 7 May (Figure 4B), which may contribute to the temporal and spatial variability of water quality due to the exchange of water masses.

During Acrobat tow-yo profiling, Chl a concentration ranged from 0.0 to 2.02 µg/L, turbidity from 0.0 NTU to 10.99 NTU, salinity from 34.92 PSU to 35.05 PSU, density from 1024.7 kg/m³ to 1024.9 kg/m³, temperature from 19.55 °C to 20.22 °C, and dissolved oxygen from 7.37 mg/L to 7.46 mg/L. Contour plots provide visual representations of the spatial variations with depth for the oceanic parameters (see Supplementary Materials, Figures S2–S5, for contours of temperature, salinity, dissolved oxygen, and turbidity). Density (Figure 5) showed a subtle decline from transect 1 to transect 7, decreasing marginally from 1024.85 kg/m³ to 1024.81 kg/m³. A modest vertical gradation (i.e., slight stratification) was observed, with density rising very slightly from 1024.76 kg/m³ at the surface to 1024.87 kg/m³ in deeper waters. There are no significant differences in salinity as a function of depth (Figure S2), and the modest density gradients are primarily driven by slight surface water heating during this midday field study (temperature contours in Figure S3).

A pattern of increasing average Chl a concentration was noted from transect 1 (0.535 µg/L) to transect 7 (0.726 µg/L) as we moved offshore (Figure 6A). Additionally, Chl a concentration consistently increased with depth, averaging 0.577 µg/L between 0 and 4 m and 0.837 µg/L at a depth of 7–10 m (Figure 6B). In other words, a moderate phytoplankton patch is visible in the bottom depths of the Chl a transects, especially as distance from shore increases.

The analysis of satellite imagery data acquired on 7 May presents a wide range of detail across the suite of sensors. We generated a mosaic of the satellite imagery spanning the study site, covering coordinates 34.10–34.25°N and 77.85–77.70°W (Figure 7), illustrating a plume of elevated Chl a concentration near the coastline and extending offshore.

Within the spatially coincident region (Figure 7), Aqua-MODIS recorded an average Chl a concentration of 2.517 µg/L (range: 2.148–3.668 µg/L). SeaHawk-HawkEye reported an average concentration of 0.0983 µg/L (range: 0.028–0.458 µg/L). S3A-OLCI produced an average of 2.740 µg/L (range: 1.912–3.874 µg/L), while S3B-OLCI returned the highest values, averaging 5.926 µg/L (range: 3.176–7.941 µg/L) (

Table 2. Satellite-derived Chl a (µg/L) and K_d,490 measurements that spatially coincide with in situ measurements.

Satellite-Sensor	Average Chl a (µg/L)	Chl a Range (µg/L)	Average Kd,490 (m−1)	Optical Depth at 490 nm (m)
Aqua-MODIS	2.517	2.148–3.67	0.187	5.35
SeaHawk-HawkEye	0.0983	0.028–0.458	0.0252	39.7
S3A-OLCI	2.740	1.91–3.87	0.163	6.13
S3B-OLCI	5.926	3.18–7.94	0.324	3.09

).

Average K_d,490 values were 0.187 m⁻¹ for MODIS, 0.0252 m⁻¹ for HawkEye, 0.163 m⁻¹ for S3A-OLCI, and 0.324 m⁻¹ for S3B-OLCI (Figure 8; Table 2). These correspond to the first optical depths of approximately 5.35 m (MODIS), 39.68 m (HawkEye), 6.13 m (S3A), and 3.09 m (S3B), indicating notable spatial differences in light attenuation. Critically, considering that in situ sampling occurred where water depth ranged from 10 to 15 m, these optical depth estimates suggest the likelihood of remotely sensed signal contamination by benthic reflectance (i.e., enough light likely penetrated through the full water column and reflected off the bottom to skew the signal).

Finally, we compared the satellite-derived Chl a dataset to the in situ measurements collected by the Acrobat instrument across the full and binned depth ranges. Quantitative comparisons incorporated RMSE, MAPE, Bias, and R-squared to assess agreement between in situ and satellite observations for the full depth profile (0–10 m) as well as segmented intervals (0–4 m, 4–7 m, and 7–10 m). Across all depth ranges and statistical measures, no sensor showed strong agreement with in situ data (

Table 3. Performance metrics summary of satellite sensors for Chl a estimation across different depth ranges.

Sensor Name	Depth Range (m)	RMSE	MAE	MAPE (%)	Bias	R-Squared
HawkEye	0–10	0.55	0.48	78.9	−0.48	0.007
MODIS	0–10	1.95	1.96	430	1.90	0.002
S3A	0–10	2.21	2.13	491	2.16	0.012
S3B	0–10	5.48	5.33	1270	5.39	0.005
HawkEye	0–4	0.30	0.26	69.8	−0.26	0.017
MODIS	0–4	2.12	2.14	560	2.10	0.36
S3A	0–4	2.41	2.35	680	2.37	0.050
S3B	0–4	5.65	5.56	1580	5.57	0.017
HawkEye	4–7	0.52	0.47	80.0	−0.47	0.021
MODIS	4–7	1.95	1.99	381	1.92	0.023
S3A	4–7	2.22	2.13	453	2.17	0.046
S3B	4–7	5.47	5.29	1160	5.38	0.019
HawkEye	7–10	0.73	0.69	86.2	−0.69	0.003
MODIS	7–10	1.82	1.87	288	1.78	0.025
S3A	7–10	2.12	2.02	335	2.07	0.019
S3B	7–10	5.31	5.17	841	5.22	0.004

, Figure 9), further indicating that benthic reflectance, land adjacency effects, or atmospheric correction may have contaminated the remotely sensed aquatic data, that the standard ocean color algorithms were insufficient in optically complex waters, and/or that the time difference between in situ and remote sensing was simply too large for a meaningful comparison.

4. Discussion

The integration of in situ measurements with satellite-derived data in dynamic coastal environments presents significant challenges that are clearly illustrated in this study. While combining these data sources is often expected to provide a more comprehensive understanding of coastal bio-physical and biogeochemical dynamics and support robust monitoring of ecological dynamics across heterogenous coastal landscapes [45], our results demonstrate that meaningful agreement between datasets can be difficult to achieve in shallow, optically complex, and highly variable nearshore systems such as the CFRE.

The in situ measurements collected using the Acrobat instrument captured fine-scale vertical and horizontal variations in Chl a, temperature, salinity, and other parameters within the water column, essential for validating surface-level satellite-derived measurements, which often average (or depth-weighted-average) these dynamic properties [38]. The satellite data offered extensive spatial coverage, revealing broader patterns and trends in Chl a and K_d,490, but, due to cloud cover, were unavailable until approximately two days after in situ sampling. Our findings highlight both the potential and the challenges of using satellite sensors to monitor Chl a concentrations and other related parameters in complex coastal waters (Case 2), which are typically heterogenous and dynamic, and where the presence of optically active substances and reflectance from the shallow seafloor dramatically alter light absorption and scattering patterns [2,17,46,47].

The depth-dependent increase in Chl a supports the hypothesis that deeper light penetration promotes phytoplankton growth in offshore waters [48]. However, this trend is complex and can be influenced by several factors. Non-photochemical quenching (NPQ) effects, where excess light energy is dissipated as heat in phytoplankton to prevent damage, could explain some vertical variations in Chl a concentrations [49]. Additionally, the observed lateral increases in Chl a with distance from shore may seem counterintuitive, as nutrient-rich runoff would typically results in higher concentrations closer to shore [50]. This apparent contradiction suggests that other processes may play a meaningful role in determining the observed patterns.

Temperature variations, with warmer surface waters compared to deeper layers, indicated a slight temperature decrease with depth potentially influencing vertical nutrient mixing and biological productivity [51]. However, it is important to note the relatively small overall change of <1 °C across the entire Acrobat dataset. Additionally, the sampling was conducted across different spatial locations (i.e., increasing distance from shore) and at different times of the day. This temporal aspect is critical, as the gradual increase in temperature from morning to midafternoon demonstrates diurnal variations that affect ecological dynamics [52]. The observed decrease in salinity from the first to the last transect could reflect the impact of tidal cycles. The first half of the in situ sampling occurred during a rising tide, while the latter half took place during a falling tide. During a rising tide, seawater influx increases salinity near the shore, while a falling tide enhances freshwater influence, potentially lowering salinity further offshore [53].

The observed density trends, coupled with an analysis of buoyancy frequency squared (N²) values across the depth profile reveals an equilibrated water column, indicating dynamic mixing processes within the water column. These findings align with the observed vertical gradients in temperature and salinity, further emphasizing the complex interplay of physical processes affecting water column stability and mixing during the sampling period.

Lastly, the increase in in situ turbidity measurements from the shore towards the ocean may indicate heightened planktonic concentrations (noting turbidity and Chl a patterns are correlated), both of which can alter optical properties and influence heat retention and biological productivity [54]. Turbidity exhibited more modest spatial variability across most transects, with a pronounced increase observed primarily along transect 6 (Figure S5). Higher turbidity at greater depths suggests elevated sediment concentration near the benthos, potentially associated with particle settling and resuspension processes [55]. We also note that Chl a fluorescence and turbidity are not independent in complex coastal waters, as suspended particles and bubbles can bias fluorescence-based Chl a estimates through optical scattering and absorption [56].

One of the key questions addressed in this study was the identification of discrepancies between in situ and satellite-derived Chl a measurements and the factors contributing to these differences. The imagery collected on 7 May 2023 revealed noteworthy variability in Chl a concentrations across the CFRE, illustrating the capabilities and limitations of different sensors. MODIS-Aqua provided broader spatial coverage with moderate resolution, which is valuable for broad environmental assessments of Case 1 waters, but can be challenging to use in coastal regions [15]. Sentinel 3A and 3B-OLCI offered a middle ground in terms of resolution and coverage, with Sentinel 3B-OLCI showing higher Chl a values, indicating potential issues with using the standard, global ocean color algorithms and/or environmental variability during the full data collection period.

The matchup analysis between in situ and satellite-derived Chl a concentrations showed varying degrees of agreement across different sensors and depth ranges (acknowledging again the significant separation in time), consistently indicating lack of agreement between satellite sensors and in situ measurements from the Acrobat. Interestingly, we observe that HawkEye-derived Chl a exhibits the closest estimates to in situ Chl a (Table 3). The range of estimated K_d,490 values illustrates how variations in IOPs influenced by factors such as water turbidity, salinity, and temperature—which varied considerably within our study region—can significantly influence sensor performance [16,57,58,59]. Notably, the low estimated K_d,490 values suggest a potential contribution from bottom reflectance, illustrating another likely challenge with data accuracy and comparability amongst datasets, especially in dynamic, shallow coastal regions [38]. To confirm the hypothesis that benthic reflectance contaminated R_rs(λ), ideally, K_d would have been measured in situ with a photosynthetically active radiation (PAR) sensor or K_d,λ with spectral irradiance measurements, but these were not available during the study.

The analysis revealed significant variability among the sensors, with MODIS and Sentinel-3B OLCI showing higher Chl a values compared to in situ data, while SeaHawk-HawkEye generally underestimated Chl a concentrations. The SeaHawk-HawkEye sensor demonstrated the most favorable performance in terms of RMSE and MAPE, particularly at shallower depths (0–4 m), suggesting its higher spatial resolution may be beneficial for surface-level measurements, especially in optically complex coastal waters, and that its atmospheric correction and Chl a algorithm perform relatively well. However, its accuracy decreased with depth, highlighting the need for complementary in situ data for subsurface analysis, and a need for further calibration and validation efforts to enhance its reliability in varying depth conditions in regions where the first optical depth exceeds the total water depth. Recent related work [33] also provides a thorough characterization of HawkEye across the R_rs spectrum and a wide range of geographic and optical settings.

Additionally, the timing of data collection proved crucial, with in situ samples collected 48 h prior to satellite acquisition (noting again that fieldwork had been planned to coincide with multiple satellite overpasses, but cloud cover resulted in the loss of those contemporaneous datasets). The resulting temporal disparities between satellite overpasses and in situ sampling likely played a leading role in the noted discrepancies. Changes in wind speed, tidal currents and phase, and cloud cover during the study period influenced water mixing and light penetration, impacting Chl a distribution and detection, highlighting the need for tightly coordinated sampling efforts to minimize such mismatches [37]. In highly dynamic inlet systems such as the Masonboro Inlet [23], water mass properties can change substantially over tidal cycles, suggesting that a 24–48 h mismatch may result in comparisons between fundamentally different water masses. As a result, the satellite–in situ matchups presented here should be interpreted with caution, as they may not represent physically consistent observations of the same water body. That said, field expeditions are logistically difficult and financially costly, underscoring the value of satellite monitoring of coastal waters and the critical need for high-resolution, high-quality and low-cost missions that freely distribute their data, such as the SeaHawk-HawkEye CubeSat mission.

5. Conclusions

This study highlights the challenges of integrating in situ and satellite-derived observations in dynamic, shallow coastal environments. Our results emphasize key limitations that must be considered when designing and interpreting ocean color validation studies in nearshore systems. In particular, the results indicate that (1) temporal mismatches greater than a couple hours can significantly limit the physical validity of satellite–in situ comparisons in tidal ecosystems, (2) shallow water depths relative to optical penetration increase potential influence of benthic reflectance on satellite retrievals, and (3) spatial variability associated with inlet dynamics and plume structure can complicate comparisons across sensors with differing resolutions.

These findings suggest that future studies should prioritize near-synchronous sampling, improved characterization of local hydrodyamics, and the use of sensor-specific or regionally tuned algorithms for optically complex waters. Additionally, expanded in situ measurements (e.g., K_d,λ) would improve the ability to diagnose sources of uncertainty in satellite-derived products.

While the CubeSat utilized in this research, the proof-of-concept SeaHawk-HawkEye, demonstrated relatively low bias vs. in situ measurements in comparison with other satellites and offered significant advantages in spatial resolution, its long revisit period and difficulties with atmospheric correction due to a mechanical defect present challenges for its application in dynamic coastal settings. Future ocean color missions may be able to overcome the temporal challenge while taking advantage of spatial resolution improvements of SeaHawk-HawkEye by launching constellations of CubeSats.

Overall, this study provides practical insights into the constraints and considerations necessary for effective integration of in situ and satellite observations. The lessons learned from this work can inform the design of future coastal remote sensing studies and contribute to improved monitoring of complex nearshore environments.