Table of Contents

• Coverage
• Dataset Description
  • Methods & Sampling
  • Data Processing Description
  • BCO-DMO Processing Description
  • Problem Description
• Data Files
• Related Publications
• Related Datasets
• Parameters
• Instruments
• Project Information
• Funding

This dataset follows the methodology described in Chua et al. (2025). Continuous water quality monitoring was conducted at three observational open-water platforms located within the main marsh channel of SMIIL. Each platform was outfitted with Aqua TROLL 600 multiparameter sondes (In-Situ Inc.), configured to measure a suite of environmental variables including water depth, temperature, salinity, dissolved oxygen (DO), pH, chlorophyll a fluorescence, turbidity. Two sondes were mounted on a fixed pole at each platform and deployed near the bottom of the water column (in water depths <2 m). Salinity, temperature, DO, and pH were measured in duplicate to ensure data quality and redundancy. Each sonde was also configured with either a chlorophyll a or turbidity sensor, allowing coverage of all target variables across platforms. Sondes were retrieved and replaced with pre-cleaned, newly calibrated sondes every 4–12 weeks. Data were downloaded during each maintenance visit using the VuSitu Mobile App (In-Situ Inc.). Although sondes were initially configured for remote telemetry, persistent data dropout issues led to a switch to internal data logging by June 2022. Discrete surface water samples for DO, dissolved inorganic carbon (DIC), and total alkalinity (TA) were also collected during maintenance visits, starting in September 2022 (Supino et al., 2025). These samples were used to validate the Aqua TROLL DO and pH data (see Data Procession section).

Meteorological data were also collected at SMIIL to complement the water quality dataset. A ResponseONE™ Weather Transmitter (R.M. Young Company) measured wind speed and direction, air temperature, relative humidity, and atmospheric pressure. Due to data transmission issues, this record only spans September 2021–May 2023. Additionally, a nearby HOBO water level logger (Onset Computer Corp.) recorded air temperature and atmospheric pressure from August 2022–June 2024. Data from both sensors were combined to produce a continuous record of air temperature and atmospheric pressure spanning the full duration of water quality measurements. To estimate surface wind speed over the study period, hourly 10-m u- and v-component wind data were obtained from the fifth generation European Centre for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5; Hersbach et al., 2023) for the grid cell closest to 39.04^oN, 74.79^oW. The ERA5-derived wind speeds were used in subsequent analyses due to their close agreement with field measurements from the ResponseONE™ station.

Water quality data were processed using MATLAB (MathWorks) version R2023b. Time and depth corrections were applied to each deployment to account for instrument offsets, and all deployments were merged into a continuous time series. Quality control procedures were based on protocols adapted from the U.S. Integrated Ocean Observing System (IOOS) Quality Assurance/Quality Control of Real-Time Oceanographic Data (QARTOD) recommendations (Integrated Ocean Observing System, 2018; Palevsky et al., 2023). Data for depth, temperature, salinity, DO concentration, pH, turbidity, and chlorophyll a were subjected to a combination of manual and automated quality checks as follows.

First, the depth time series for each sonde/platform was manually inspected and “human-in-the-loop” (HITL) annotations were created to document known and identified bad data. Points flagged by HITL review were compared to service trip logs to diagnose the cause of bad data (e.g., from power or communications disruptions or when the sensor was out of water between deployments). Flagged points that arose from data dropouts or when the sonde was out of water were removed from further analysis for all parameters (not just depth). Next, a gross range test was performed to check whether data were within reasonable ranges for a specified parameter. Parameter-specific selection of gross range limits (Table 1) was guided by Aqua TROLL sensor output ranges, expert judgement, and local knowledge. Points that exceed these thresholds were flagged and removed as bad data. Then, a spike test was performed to evaluate whether a data point exceeded a threshold relative to adjacent points. The absolute value of the spike was tested to catch both positive and negative spikes. Parameter-specific spike test thresholds (Table 1) were selected by assessing the frequency distribution of the absolute difference between a point and the average of its nearest neighbor points for the parameter in question. For both the gross range and spike tests, the thresholds were kept consistent for a given parameter between open-water platforms. Following these initial quality control tests, the data were retimed to 10-minute bins to resolve irregular sampling times.

Following initial QC, a moving median test was applied to detect outliers in the remaining datasets (Leys et al., 2013). This test was not performed for depth or temperature because of the large diel oscillations in these parameters. A 25-hour moving window median was calculated, and then the absolute deviation of each value from the moving median was obtained. An upper median absolute deviation (MAD) was obtained as the median of the deviations above the moving median, and a lower MAD for the deviations below. This double MAD approach was employed because the data distributions for all of the parameters were asymmetric (Tan et al., 2022). Points that exceeded ±3 respective MADs from the moving median were flagged and excluded from further analysis. 

For parameters measured in duplicate (salinity, temperature, DO concentration, and pH), datasets were compared to produce a single, final quality-controlled time series. To assist this assessment, the absolute difference between duplicate datasets for a parameter was found for each time point and highlighted if it exceeded a certain threshold. The data were then visually examined and the “best guess” data was determined on a deployment-by-deployment basis. Several general principles were used to guide this selection. If the individual time series were offset for no apparent reason, and neither appeared to match up better with the previous and next deployment, then the mean time series was chosen (for salinity only, because the data were extremely noisy, the mean of the 24-hour moving median was used). If there were clear differences between duplicates for a given deployment (e.g., the individual time series were offset and one clearly did not match up with the previous and next deployments; one time series diverged from the other; or only one sonde had a complete record for that deployment), then the data from the “better” sensor was used. Data between related parameters were compared to aid this assessment (e.g., if pH dropped to relatively low values, the DO data were consulted to see if it coincided with a period of hypoxia or anoxia).

To further help in determining the best guess data, discrete Winkler samples were compared to the DO data, and DIC/TA samples were used to evaluate the pH data [reference discrete sample dataset in this project]. Discrete salinity samples were occasionally available for spot-checking. For the pH comparison, Aqua TROLL pH data were converted to hydrogen ion concentrations ([H+]) rather than evaluated on a logarithmic scale. Accordingly, the DIC and TA values were used to calculate discrete [H+] values using the CO2SYSv3 MATLAB program (Lewis & Wallace, 1998; Sharp, 2023; Van Heuven et al., 2011). Errors in the CO₂ system variables were propagated to estimate errors in the calculated [H+] (Orr et al., 2018). The discrete values were used to help guide judgment in two cases: 1) determining which of the duplicate sensors was better performing when they differed, and 2) correcting for offsets in the sensor data when offsets were observed at both the beginning and end of a deployment. While the discrete sample data were useful additions to help inform the duplicate sensor check, they were evaluated on a deployment-by-deployment basis and as few adjustments as possible were made, because of uncertainty in the discrete measurements themselves as well as potential spatial mismatch in the water mass sampled and the sonde location. Moreover, the discrete sample coverage did not span the entire open-water platform time series, and therefore were only applicable to certain deployments in the last two years of measurements.

Because the Aqua TROLL sondes produce DO concentration values based on the measured pressure, temperature, and salinity, any errors in these measurements will propagate into the output DO values. To address this, DO concentration values were recalculated prior to duplicate comparison based on the salinity and depth compensation equations of Aanderaa Data Instruments AS (2017) using the final quality-controlled depth, temperature, and salinity time series. DO % saturation was calculated by normalizing the measured DO concentration to the saturation DO concentration (DO % saturation = DO_measured / DO_sat​ ×100%), where DO_sat was calculated as a function of the final quality-controlled temperature and salinity (Garcia & Gordon, 1992).

Turbidity and chlorophyll a, which were only available from single sensors at a platform, were reviewed further for final manual removals. This final manual inspection was necessary to address known issues of biofouling, which strongly affects the performance of these optical sensors by obscuring the sensor and causing sustained upward drift and excessive noise (Palevsky et al., 2023), and which was not fully remedied by the prior quality control tests. All of the open-water platform sondes experienced heavy biofouling year-round; a recent study conducted at the SMIIL site also reported pronounced biofouling that impacted turbidity sensor performance (Perkey et al., 2024). Here, cross-platform data were examined on a deployment-by-deployment basis for evidence of biofouling, and suspect data were removed, resulting in the removal of large periods of data.

Table 1. Thresholds for the initial data quality control tests for the parameters specified. Data exceeding these thresholds were flagged as suspect and removed prior to subsequent quality control tests.

- Imported "gull_finalDataset.csv", "north_finalDataset.csv", and "south_finalDataset.csv" into the BCO-DMO system, replacing NaN with blanks
- Concatenated all three files, adding a column for "site", which is what delineated the three files
- Converted datetime to ISO 8601 format YYYY-MM-DDTHH:MM:SSZ
- Exported file as "971943_v1_marsh_channel_monitoring.csv"

Salt marshes are highly productive, dynamic coastal environments that experience large fluctuations in biogeochemical parameters such as dissolved oxygen and pH in response to both natural forcings and anthropogenic impacts. At present, we have a limited understanding of the magnitude of biogeochemical variability in coastal habitats, hindering our ability to predict how they will respond in the future to episodic events and long-term change. This incomplete picture owes to a lack of sustained water quality measurements in coastal and estuarine systems worldwide.

This project investigates biogeochemical cycling in the Seven Mile Island Innovation Laboratory (SMIIL), a network of tidal marshes and channels in coastal New Jersey that is a site of historic and contemporary dredging and a testbed for marsh restoration techniques such as beneficial use of dredged sediment. It leverages multiple data sets of continuous, high-frequency (10-minute) measurements of physical and biogeochemical parameters, including water depth, temperature, salinity, dissolved oxygen, pH, chlorophyll a, and turbidity, as well as precipitation, wind speed, wind direction, temperature, relative humidity, and atmospheric pressure data. Two long-term biogeochemical sensor data sets, comprising a three-year time series at three sites in the main marsh channel (from June 2021–June 2024) and 1-2 year time series at six distinct salt ponds (August 2022–June 2024), were collected. Additionally, short-term (~months long) dredging and sediment placement monitoring data were collected at five different locations in 2022 and 2023. Data collection was designed to capture a variety of natural (diel, tidal, seasonal, storm-related) and human-created (sediment dredging and placement) conditions. Factory-calibrated biogeochemical sensors were field calibrated and maintained every 4 to 12 weeks. Discrete samples for dissolved oxygen, total alkalinity, and dissolved inorganic carbon were collected at the start and end of each sensor deployment for additional sensor calibration and validation. Altogether, our high-quality, multiyear dataset provides critical insights into the inherent variability of biogeochemical conditions in temperate salt marshes on diel and seasonal timescales, as well as how they may respond to transient events (e.g., storms, dredging activities) and over the longer term.