Automated sensors reveal short-term nitrogen dynamics in coastal waters

Coastal scientists have long known that nitrogen can change rapidly in estuaries, but routine monitoring often sees only a fraction of that movement. Water samples collected once every few weeks may miss nutrient pulses caused by rain, river discharge, wind-driven sediment disturbance or sudden bursts of biological activity. A new study in Roskilde Fjord, Denmark, shows how automated chemical sensing can expose these hidden fluctuations, revealing that coastal nitrogen dynamics can shift substantially over just a few hours. The findings suggest that monitoring programs designed around fixed, infrequent sampling schedules may underestimate the natural variability that shapes eutrophication, productivity and ecosystem health.

Christian Lønborg, Peter A. U. Stæhr and Cátia Carreira deployed an automated nutrient analyser at a pier in the southern part of Roskilde Fjord during three campaigns: late October to early November 2023, February 2024 and April to May 2024. The shallow, temperate estuary is about 30 kilometres long, has a mean depth of roughly 3 metres and receives freshwater from a drainage basin dominated by agriculture. That runoff delivers an estimated 1,000 tonnes of nitrogen and 50 tonnes of phosphorus to the fjord each year. The fjord is therefore a useful natural laboratory for examining how river inputs, weather and biological processes interact in nutrient-rich coastal waters.

The researchers focused on nitrate plus nitrite, written as NO₃⁻ + NO₂⁻, and nitrite, NO₂⁻. These are inorganic forms of nitrogen that can be taken up by phytoplankton and other organisms, transformed by microbes or transported through the estuary. The sensor was programmed to sample every four to six hours at a depth of 1 metre. Instead of relying on ultraviolet light, as many nitrate sensors do, it used a reagent-based colorimetric method similar to automated laboratory flow analysis. Water was filtered through a 0.45-micrometre inlet filter before entering the analyser.

The instrument measured nitrite directly through a chemical reaction that produced a coloured compound. The colour intensity was then measured spectrophotometrically, with stronger colour corresponding to a higher nutrient concentration. To determine nitrate plus nitrite, vanadium(III) chloride first reduced nitrate to nitrite at moderate temperatures of about 40–60 °C. The combined nitrite signal then represented the total concentration of both nitrogen forms. This approach allowed the analyser to distinguish nitrite from the broader nitrate-plus-nitrite pool, while collecting measurements far more frequently than a research vessel could realistically manage.

The automated results generally agreed with conventional laboratory measurements. Across field samples and laboratory dilution tests, sensor readings tracked the one-to-one relationship expected with discrete samples analysed by segmented-flow analysis. The sensor’s average precision was 0.07 micromoles per litre for nitrate plus nitrite and 0.01 micromoles per litre for nitrite. Differences between successive readings exceeded the threshold for statistical significance in 91 per cent of nitrate-plus-nitrite measurements and 87 per cent of nitrite measurements. On average, however, sensor values were slightly lower than laboratory results—by about 5 per cent for nitrate plus nitrite and 9 per cent for nitrite—probably because the two methods used different chemical reduction procedures.

That agreement came with important qualifications. In a dilution experiment, the largest discrepancies for nitrate-plus-nitrite measurements occurred in samples from Hundested Harbour, the site with the highest salinity. The researchers suggest that the chemical “matrix” of seawater may have influenced the results. Salinity, dissolved organic material, refractive index and other substances can affect how efficiently nitrate is reduced and how the final colour is detected. Such effects are especially relevant in coastal waters, where freshwater and seawater mix over short distances and where the composition of the water can change after storms or runoff events.

The time series showed pronounced seasonal contrasts. Nitrate plus nitrite ranged from 4.3 to 107.8 micromoles per litre in sensor measurements, while nitrite ranged from 0.20 to 2.38 micromoles per litre. Concentrations were highest and most erratic during the February campaign, when nitrate plus nitrite in discrete samples reached 106.76 micromoles per litre. They were generally lower during the brighter, warmer April–May period. Chlorophyll a, used as an indicator of phytoplankton biomass, varied from just 0.1 microgram per litre in February to 48.8 micrograms per litre in spring. The contrast suggests that large nitrogen pools can build up during cold periods when biological uptake is limited, then decline as light and temperature stimulate growth.

The researchers linked the short-term changes to a combination of physical supply and biological demand. Higher nitrogen concentrations were generally associated with lower salinity, consistent with an increased influence of freshwater carrying nutrients from the surrounding land. Wind also appeared to raise nitrate-plus-nitrite and nitrite concentrations, probably by stirring the shallow seabed and resuspending particles or porewater enriched with dissolved nutrients. In contrast, concentrations tended to fall as solar radiation and chlorophyll a increased, indicating that phytoplankton and other organisms were removing inorganic nitrogen from the water as biological activity intensified.

February provided a particularly revealing example. Nitrogen concentrations were high even though salinity was not at its lowest, suggesting that riverine delivery alone could not explain the observations. Continued runoff, combined with weak biological uptake under cold and relatively dark conditions, may have allowed nitrogen to accumulate. Short-lived increases may also have followed individual precipitation and wind events. The February deployment recorded maximum winds of about 20 metres per second and 81.4 millimetres of precipitation, compared with maximum winds of 13 metres per second and 42.2 millimetres of rain during the autumn campaign. These bursts of forcing are precisely the kinds of events that weekly or monthly sampling can miss.

The sensor records also exposed a less obvious problem: when a sample is collected may matter almost as much as where it is collected. The period of day that best approximated the daily average changed with season. Night-time measurements were most representative during October and November, afternoon measurements during February and morning or afternoon measurements during April and May. The study did not identify a consistent day–night cycle in nitrate plus nitrite or nitrite, but the seasonal differences showed that a single sample can be unrepresentative even when it is collected at the same location. A monitoring program that always samples in the morning, for example, may systematically miss the average condition during some parts of the year.

To test how conventional monitoring might perform, the researchers repeatedly drew random subsets from the high-frequency records, simulating sampling frequencies of one, two, four, seven, 10 or 14 observations per deployment. They compared the resulting averages and 95 per cent confidence intervals with data from Denmark’s NOVANA monitoring program, whose nearby station is sampled much less frequently. Temperature and salinity were relatively robust to reduced sampling, although the precision of their averages improved with more observations. Biogeochemical variables were much more sensitive. Chlorophyll a, nitrate plus nitrite and nitrite showed large changes in estimated averages and uncertainty when the number of observations was reduced.

In one example from February, a single temperature measurement of 2.28 °C underestimated the period’s average of 4.03 °C by about 1.75 °C. A single nitrate-plus-nitrite measurement of 107.85 micromoles per litre contrasted with a campaign average of 86.57 micromoles per litre, a difference of 21.28 micromoles per litre. During April and May, one measurement gave 13.61 micromoles per litre compared with a full-period average of 18.94 micromoles per litre. These are not merely statistical curiosities: estimates of nutrient exposure, eutrophication status and biological response can change depending on whether a monitoring program happens to sample during a pulse or a lull.

The results do not mean that every coastal station needs an analyser recording every few minutes. The instrument used in this study was deliberately set to a four-to-six-hour interval to balance temporal resolution, deployment length and data quality. Sampling too frequently can accelerate sensor drift, consume reagents and generate large datasets that are difficult and expensive to validate. The study also revealed practical engineering limits. During the cold February deployment, when water temperatures fell below about 3 °C, the reagents crystallised and measurements were interrupted for roughly two weeks. Long-term deployments would also need protection against biofouling, calibration drift and interference from changing water chemistry.

The sensor measured only nitrate plus nitrite and nitrite, not the complete nitrogen cycle. In Roskilde Fjord, nitrate plus nitrite accounted for approximately 16 per cent of the total nitrogen pool, while dissolved organic nitrogen made up a much larger share. The automated measurements therefore provide a sensitive window into one important part of the system rather than a complete accounting of nitrogen. The authors say further testing across wider environmental conditions, longer deployments and different coastal habitats will be needed before such instruments can be incorporated fully into national regulatory monitoring. Parallel collection of discrete samples remains essential for quality control and for ensuring that sensor data meet the requirements of frameworks such as the European Union’s Water Framework Directive and Marine Strategy Framework Directive.

For now, the study points toward a hybrid future in which autonomous instruments fill the gaps between carefully analysed water samples. High-frequency observations could alert scientists to storm-driven nutrient pulses, sediment releases or rapid biological uptake that would otherwise disappear between scheduled visits. They could also help managers choose more effective sampling times and frequencies for different variables and seasons. The central message is that coastal nitrogen is not a slow-moving background signal: it is a rapidly changing currency exchanged among rivers, sediments, microbes, plankton and the atmosphere. Capturing that motion may be crucial for understanding when nutrient pollution causes the greatest ecological damage—and for measuring whether efforts to reduce it are actually working.