A Low-Cost, Open-Source, In Situ, Near-Real-Time Sensor for the Detection of Nitrate and Ammonia in Environmental Waters

Abstract

Elevated levels of ammonia and nitrate in water bodies contribute to environmental issues such as eutrophication and pose serious risks to human health and aquatic life. Traditional monitoring methods rely on grab sampling, and existing in situ sensor technologies are often prohibitively expensive, which limits their widespread deployment. This study presents a low-cost, in situ, near-real-time sensor and a hand-held device for the detection and monitoring of ammonia and nitrate in water. The hand-held device demonstrated excellent performance during laboratory testing, achieving an R ² of 0.96 when estimating both ammonia and nitrate concentrations in real environmental samples. It reliably quantified both nutrients across a broad range, with accurate measurements observed up to 10 mg/L. The in situ system showed reliable performance for ammonia in environmental samples (R ² = 0.76) and detected ammonia down to 0.1 mg/L. However, below this threshold, signal variability increased, and measurements became less consistent. During four field sampling periods, it demonstrated reliable performance in measuring ammonia concentrations, achieving an R ² of 0.86 and a Nash–Sutcliffe efficiency (NSE) of 0.83. The results showed that the low-cost sensor can operate for 2 weeks by sampling every 40 min using just one kit of low-cost reagents (16 USD per kit, and 0.09 USD per sample).

1. Introduction

Anthropogenic activities significantly impact the distribution of ammonia and nitrate in water bodies globally. Agricultural practices, particularly fertilizer use, are major contributors to nitrogen pollution in rivers and groundwater. , Urbanization and industrial development further increase nitrogen loading in water bodies. In highly developed areas, nonpoint-source pollution has become the dominant source of nitrogen. These activities alter nitrogen cycling, affect microbial communities, and can lead to harmful algal blooms.

High levels of nitrate and ammonia pose significant risks to human health and the environment. Consuming vegetables with nitrate accumulation can lead to serious health hazards. Furthermore, high nitrate intake has been shown to impair liver function and morphology in animal studies. Chronic exposure to nitrate in drinking water has also been linked to an increased risk of certain cancers and adverse reproductive outcomes. Ammonia neurotoxicity can disrupt bioenergetics and alter neurotransmission, exacerbating its harmful effects. Beyond human health concerns, elevated levels of ammonia and nitrate contribute to the eutrophication and acidification of ecosystems. In aquatic organisms, prolonged exposure to ammonia and nitrite disrupts hematological parameters and weakens antioxidant defenses. Given these widespread impacts, accurate monitoring of nitrate and ammonia is essential for effective environmental and public health management, allowing for the identification and mitigation of contamination sources to safeguard ecosystems and human well-being.

Nitrate and ammonia detection in water can be categorized into two main methods: direct and indirect. Direct methods measure contaminant levels without chemical conversion, offering simplicity and real-time analysis. These include potentiometric methods (i.e., ion-selective electrodes (ISEs)), voltammetric detection, chromatography, and spectrophotometry, which are known for their accuracy and cost efficiency. However, they often suffer from lower sensitivity compared to analytical techniques, require frequent maintenance, and are unsuitable for long-term monitoring. , Indirect methods, on the other hand, involve chemical reagents to convert the target analyte into a more stable or detectable form, typically producing a colorimetric reaction for quantification. Techniques such as flow injection analysis (FIA) and colorimetric methods rely on expensive instruments and reagents, making them complex, labor-intensive, and less practical for continuous field applications. , Additionally, these methods often require sample collection and laboratory analysis, limiting their ability to provide high-frequency, temporal measurements. Consequently, there is a growing need for alternative, low-maintenance, and in situ monitoring solutions that can overcome the limitations of both direct and indirect methods. ,

ISEs offer a low-cost, portable solution for online nitrate measurement without requiring pretreatment, making them a promising tool for real-time monitoring. Kim et al. developed an advanced automated three-electrode ISE system controlled by an Arduino board for in situ nitrate detection, achieving a standard error of 15% compared to analytical nitrate concentrations. ISEs have notable limitations, including a high dependency on electrolyte stability and a reduced lifespan due to electrolyte depletion. Additionally, external factors such as interfering ions and environmental conditions can impact the sensor performance. Recent electrochemical studies have emphasized similar challenges, noting that while electrochemical platforms provide sensitive and portable detection for aquatic pollutants, issues such as oxygen interference, ion competition, and long-term stability remain critical bottlenecks for real-world deployment.

Recent advancements in nitrate detection technologies have introduced ultraviolet (UV) analyzers, such as the Submersible Ultraviolet Nitrate Analyzer (SUNA) and the TriOS OPUS UV optical nitrate sensor. These optical sensors enable high-resolution, in situ monitoring of nitrate concentrations across diverse aquatic environments. While commercial sensors of this type have demonstrated potential for continuous nitrate monitoring, in practice, these commercial optical sensors often fail to meet their stated detection limits. Snyder et al. reported discrepancies between grab sampling and sensor-derived results, particularly when the concentration of nitrate falls within 0.1 to 1.0 mg/L; only 37.5% of measurements were within the accuracy bounds of the manufacturer. Moreover, when the concentration of dissolved organic carbon (DOC) derived from leaf leachate increased to 25 mg C/L, the SUNA estimation was 2.5 mg N/L above the actual NO₃ concentration. These discrepancies are especially problematic in stormwater and surface waters, where typical ammonia and nitrate concentrations are frequently in the range of 0 to 2 mg/L. , Therefore, despite their technical potential, commercial sensors face significant practical challenges, including limited effective detection range, sensor drift, background interference, , and high operational costs.

Recent advancements in microprocessors and low-cost electronics have greatly enhanced the accuracy, accessibility, and practicality of water quality monitoring sensors. Portable, low-cost sensors that incorporate IoT integration now enable real-time, continuous monitoring of various water quality parameters, , offering more cost-effective, user-friendly, and scalable solutions compared to traditional laboratory-based methods. Additionally, commercially available chemical reagents for water testing have improved access to affordable colorimetric assays for nitrate and ammonia. While hand-held test kits relying on manual reagent dosing and microprocessor-driven or microfluidic platforms for continuous sensing have been reported, these approaches have not yet been combined into a continuous, IoT-enabled, reagent-based colorimetric system for field monitoring of nitrate or ammonia. Moreover, most previous optical ammonia sensors have focused on ammonia gas detection, laboratory-based configurations, or complex chemistries that are not readily transferable to practical water monitoring applications. ,

To help address the lack of low-cost, sensitive, and real-time nitrogen detection technologies suitable for environmental waters (0–2 mg/L), this study developed a hand-held colorimetric sensor for detection of both ammonia and nitrate, together with an in situ version dedicated to ammonia monitoring in water. To the best of our knowledge, this is the first open-source, hand-held, and in situ optical ammonia sensor designed for practical field deployment and near-real-time detection in environmental waters. The open-source nature of the system enables full transparency in both hardware and software design, supports community-driven innovation, and reduces costs by avoiding proprietary components, thereby overcoming key limitations imposed by expensive commercial technologies.

2. Methods

In this study, the colorimetric sensor for nitrate and ammonia was designed and developed through four key stages: (1) design considerations, (2) sensor development, (3) laboratory calibration using both synthetic and environmental samples, and (4) final validation of the sensor under real field conditions. Figure presents a schematic overview of this development process.

1.

Schematic overview of the steps involved in the design, development, calibration, and validation of the low-cost colorimetric sensor.

2.1. Sensor Design and Development

While multiple brands offer commercial reagents designed for fish tank water quality, we selected the API Ammonia and Nitrate Test Kit (API Fishcare, Pennsylvania, USA) as it is easy to access around the world, and it is low-cost (16 USD). The API test kits consist of reagents that require manual preparation: for ammonia testing, two reagents are mixed with 5 mL of the water sample, shaken for 1 min, and left to develop color for 5 min. The resulting color is then compared to a provided color chart to estimate the ammonia concentration. The nitrate test kit involves an additional step, requiring one reagent to be shaken for 30 s before being added to the water sample.

We aimed to develop two distinct setups: a hand-held device and an in situ sensor. The hand-held device consists of a TCS color sensor to digitize the intensity of the developed color, while a modified MADAS was employed to automate the sampling and mixing process to build the in situ sensor. Both the hand-held device and in situ sensor designs are all open source and open hardware, and together with the code to operate the systems, can be downloaded from ref .

The estimated hardware bill of materials (Table S1, Supporting Information) for the in situ sensor, including the BoSL board, TCS color sensor, LED, peristaltic pumps, tubing, 3D-printed PETG components, glass windows, O-rings, magnetic stirrer components, and SD-card logging hardware, was approximately USD 23.56 for the hand-held device and USD 126.55 for the in situ sensor, excluding labor and analytical reagents.

2.1.1. Hand-Held Device Sensor Setup

The hand-held device setup was designed as a 3D-printed chamber featuring a 2.7 V LED (Broadcom, California, USA) positioned at the bottom, with the TCS sensor mounted inside to measure the light emitted by the LED (Figure ). Additional components used included a BoSL 0.5 microcontroller board, and a TCS34725 (blue model) color sensor (Texas Advanced Optoelectronic Solutions Inc., Texas, USA), all sourced online. The TCS34725 is a light-to-digital converter with integrated red, green, blue and clear photodiodes and an infrared blocking filter, providing spectral responsivity across the visible range (∼400–700 nm) while suppressing IR interference.

2.

Hand-held device color sensor using a 3D-printed case and Eppendorf tubes for measuring RGB values, and schematic of the connections for the TCS sensor and LED to the BoSL board.

Ammonia and nitrate samples were manually prepared in 5 mL Eppendorf tubes and placed inside the chamber for measurement. Each reading was taken 20 times using the TCS sensor, with the data recorded via a BoSL board 0.5 and saved on an SD card for further analysis.

2.1.2. In Situ Sensor

The in situ sensor featured a compact 3.13 mL chamber designed for both mixing reagents and measuring color temperature values (Figure –part 4). To enhance water tightness, the chamber was 3D-printed using poly­(ethylene terephthalate glycol) (PETG), and subsequently coated with multiple layers of Water Shield (Selleys, Australia) sealant. The two open sides of the hollow chamber, allocated for the TCS sensor and LED light, were sealed using O-rings placed beneath transparent acrylic plates. These plates were securely fastened with bolts and nuts to ensure a reliable, watertight seal. A complete schematic view of all of the connections of the in situ sensor is provided in Figure S1.

3.

Colorimetric reading cycle of the sensor and a 3D model of the in situ sensor showing all of the parts as follow: 1: 3D-printed stand for water tube, reagent tubes, and connection wires, 2: BoSL FAL pump for water, 3: BoSL FAL pump for reagents, 4: mixing chamber, 5: magnetic stirrer case, 6: DC motor, 7: 3D-printed cover, 8: magnet for activating the small magnet inside the chamber, and 9: small magnet inside the chamber for mixing the solutions.

The in situ sensor utilized a novel 3D-printed automatic sampler known as MADAS for flexible pumping. MADAS operates by controlling the number of spins in a small 3D-printed peristaltic pump, using a Hall effect sensor to read signals from a small magnet embedded in the rotor of the pump (Figure –parts 2 and 3). The design was modified to allow two pumps to operate simultaneously, enabling the simultaneous pumping of both the water sample and the reagent into a mixing chamber.

We designed a magnetic mixing system to ensure consistent and thorough mixing of the reagents and water sample inside the chamber. The system included a small 4 mm long magnet placed within the chamber (Figure –part 9), which was activated externally by a magnetic activator located at the base. The activator comprised two magnets attached to either end of a rectangular 3D-printed component connected to a DC motor (Figure – parts 6 and 8). When the motor was activated at low speed, the internal magnet rotated, enabling uniform mixing of the solution. The overall setup aimed to facilitate effective reagent distribution and stable color development, improving the accuracy and reliability of the sensor readings.

In our case, the in situ sensor logged measurements as timestamped CSV files to a microSD card via the BoSL 0.5 board and transmitted them to a live online server using our integrated SIMCOM7000G device. For field deployment, the sensor could use a lower-cost Arduino R3-style board or BoSL Nanostyle microcontroller to acquire measurements and transmit data to a bank-side receiver/logger. The bank-side unit can then provide cellular connectivity for cloud upload, where required. The system was powered using two 3.7 V, 700 mAh lithium-ion rechargeable batteries and operated for more than 1 week at a 30 min sampling interval during testing. The chamber was sealed using O-rings and acrylic plates, while the 3D-printed components were coated with waterproofing spray to improve water resistance. The current prototype was designed for near-surface or bank-side deployment, and formal pressure testing to determine the maximum waterproof depth remains a subject for future work. No visible biofouling was observed inside the measurement chamber during the deployment period, potentially due to repeated exposure to the acidic/oxidizing reagent chemistry during each measurement cycle; however, longer-term biofouling evaluation is still required.

The in situ sampling process operated as follows: initially, both pumps were calibrated, and then the water was pumped to purge the previous solution and fill the chamber with a freshwater sample. Next, the reagents were injected into the chamber. The magnetic stirrer mixed the solution for 1 min, after which the LED light illuminated the chamber, and the TCS measured the color values through the designated hole. Once the measurements were recorded, the solution inside the chamber was purged again to restart the cycle. The in situ sensor completes one full measurement cycle in approximately 10 min and can be programmed to sample at user-defined intervals (default: 40 min), with results logged automatically to an onboard SD card and uploaded to an online cloud server, providing near-real-time data availability compared to conventional grab-and-analyze approaches.

2.2. Performance Optimization

To ensure robust performance and reliability of the colorimetric sensor, we developed some design tests to identify potential sources of variability and uncertainty in the colorimetric sensor. Their outcomes provide insights into the operational framework of this sensor.

2.2.1. Reagent Homogenization

To evaluate the effect of shaking the reagent bottles before injecting them into the water sample for color development as recommended by the reagent’s manufacturer, we allowed the reagents to remain stationary for 1 week in an upside-down position to minimize any unintentional mixing during handling. Subsequently, we tested a 5 mg/L solution of both ammonia and nitrate using both the stationary (still) reagent and the shaken reagent to determine if there were any differences in the resulting color development.

2.2.2. Reagent Ratio Tolerance

According to the manufacturer’s instructions, 8 drops (0.3 mL) of each reagent are required for 5 mL of an ammonia–water sample. We further tested reagent dosages at −50%, −25%, + 25%, and +50% of the standard volume to simulate underdosing and overdosing errors induced by peristaltic pump fluctuations. For instance, based on the manufacturer’s recommended optimum dosage of 8 drops from each reagent for a 5 mL water sample (8/8), we additionally tested 25% and 50% lower and higher reagent volumes to simulate potential under- and overdosing caused by peristaltic pump variability. This resulted in the following tested combinations: 8/8, 4/4, 12/12, 8/4, 8/12, 4/8, and 12/8, which were used to assess the impact of reagent ratio deviations on color development and measurement accuracy.

2.2.3. Effect of Ambient Light

When testing the TCS color sensor under direct sunlight, noise can be introduced into the readings, causing discrepancies in measurements in a field environment. To evaluate the effect of sunlight on the in situ sensor and determine whether the sensor measurement needs to be conducted in a dark environment, we conducted tests by placing the sensor under its protective cover or inside a box. Color temperature (CT) values were measured and compared to determine the extent of the impact sunlight has on sensor performance and whether the protective measures effectively mitigate these effects.

2.2.4. Reagent Delivery Volumes

To measure the precise reagent volume in the in situ sensor, the density of the reagents was first calculated by measuring the full and empty mass of a 10 mL volumetric flask. The milliliters per peristaltic rotation for each MADAS was then manually measured by filling a 10 mL volumetric flask. These measurements determined the number of rotations required for each pump to achieve the desired water-to-reagent ratio.

2.2.5. Optimal Flushing Volume

To determine the optimal volume of fresh water required to completely purge the previous solution and introduce a fresh solution for the next reading, an experiment was conducted using two different ammonia concentrations (1 mg/L and 13 mg/L). The color temperature was measured at two stages: before adding the reagent and mixing, and after adding the reagent, mixing for 1 min, and allowing a 5 min reaction time. Fresh water was then introduced into the chamber in increments of 3.13 mL (equivalent to one pore volume), followed by a reading. This process was repeated 20 times. The data collected from this experiment facilitated the determination of the optimal number of pore volumes required to ensure the complete purging of the chamber before each new measurement.

2.3. Lab Calibration Using Standard Solutions

To calibrate the hand-held device and in situ sensors, various concentrations of ammonia and nitrate solutions were prepared by diluting them with DI water. To do this, standard solutions of 10 mg/L ammonia and nitrate were prepared by dissolving 0.038 g of ammonium chloride (NH₄Cl) and 0.061 g of sodium nitrate (NaNO₃) into separate 1000 mL flasks. The salicylate–hypochlorite method quantifies ammonia nitrogen, reported here as total ammonia nitrogen (TAN = NH₃ + NH₄ ⁺, expressed as N). Throughout this manuscript, the term ammonia is used as shorthand for TAN unless otherwise stated. Sodium nitrate and ammonium chloride were obtained from Sigma-Aldrich, USA. To ensure that the inherent background color of the water sample did not affect the readings, the TCS sensor recorded baseline measurements before the injection of reagents. Additionally, the impact of turbidity on sensor measurements was evaluated by using samples with turbidity levels ranging from 0 to 250 NTU. To achieve this, a standard turbidity solution was created by mixing DI water with dry garden soil to obtain specific turbidity levels of 10, 20, 35, 80, 160, and 250 NTUs. As garden soil may contain trace amounts of nitrogen, a control sample (DI water mixed with the soil) was tested to confirm that the background nitrogen levels were negligible and did not interfere with the ammonia or nitrate measurements. These turbidity solutions were then used to dilute various concentrations of ammonia and nitrate, enabling the simultaneous assessment of the effects of turbidity and target pollutant concentrations on the hand-held device colorimetric sensor. The R, G, B, C, lux, and color temperature values were measured and calculated by using the Adafruit_TCS34725.h library.

To compensate for turbidity interference, a multiple linear regression model was developed where NH₃ concentration was expressed as a function of two inputs: (1) The color temperature value after reagent addition (CT₁), and (2) the blank color temperature value before reagent addition (CT₀). This relationship is described in eq

$${\mathrm{NH}}_3(\mathrm{mg}/\mathrm{L})=\alpha +{\beta }_1\times {\mathrm{CT}}_1+{\beta }_2\times {\mathrm{CT}}_0$$ 1

where α, β₁, and β₂ are regression coefficients derived from the calibration data set. This model enables the sensor to isolate the chemical signal from turbidity-related background noise, substantially improving accuracy under variable field conditions.

2.4. Lab Validation Using Field Samples

River water samples were collected from the Yarra and Bunyip rivers in Melbourne, Victoria. These samples were used to validate the correlation between color temperature and nitrate/ammonia concentrations and to calibrate both hand-held devices and in situ sensors. These samples were tested without any pretreatment, while simultaneously being filtered through a 45 μm filter for laboratory analysis of actual ammonia and nitrate concentrations. For reference measurements, flow injection analysis (FIA) was conducted using a Hach Lachat QuickChem 8500 system (Colorado, USA).

To cover a wide range of ammonia and nitrate concentrations, we not only tested these samples directly after collecting them from the waterways but also spiked these environmental samples with wastewater. Before testing each sample, turbidity was measured, and a blank reading was taken to account for the potential influence of background color in environmental water. Both the hand-held device and in situ sensors were evaluated using these environmental samples to assess their accuracy and reliability in detecting ammonia and nitrate concentrations.

2.5. In-Field Validation

To further evaluate the performance of the in situ sensor under real-world conditions, field testing was conducted at the Dandenong Creek sampling location and a nearby wetland (Figure ). Five locations within the wetland were selected to cover a range of ammonia concentrations (Figure ). The laboratory analysis of nitrogen species concentrations is provided in Table S2 of the Supporting Information. Field validation was conducted from November 2024 to March 2025, comprising 27 field experiments across 4 sampling campaigns (Figure S2). The sampling frequency varied by month, with no field campaign conducted in January 2025. The initial two rounds of fieldwork involved collecting hourly samples from a wetland near Dandenong Creek (location 1) and Dandenong Creek, while subsequent sampling efforts included daily collections from all locations under both wet and dry conditions.

4.

Study site for field validation, including sampling locations for Yarra River, Dandenong Creek, Bunyip River, and five sampling locations of the study wetland. The in situ sensor is located in an insulated cooler box for field validation at the Dandenong Creek location. The water was pumped up using a peristaltic pump into a bottle, and it was used by the sensor to measure the ammonia level.

To minimize the impact of sunlight on measurements (Figure S3), particularly during this experiment, where the standard protective cover was not used to match the condition of the test to the lab testing condition, the sensor was housed inside a cooler, with water samples pumped into a collection bottle using a peristaltic pump. After thorough manual mixing to ensure homogeneity, the sample was subsampled into five replicate bottles by repeatedly dispensing 10 mL aliquots into each bottle until a final volume of approximately 100 mL per bottle was obtained. Two aliquots were randomly selected, one filtered for laboratory analysis and the other tested in the field using the in situ sensor. Although this setup was not intended as a model for typical field deployment, it enabled a comparable condition under which the lab correlation was made. Therefore, the field results presented here should be interpreted as field-based validation using homogenized samples, rather than continuous real-time in situ operation, but provide the evidence that this system can be further optimized to create a real-time, automated, in situ, semicontinuous sensor system in the future. In routine installations, however, the sensor would be enclosed within its waterproof and lightproof protective cover, with water pumped directly into the measurement chamber to ensure continuous, hands-free operation and protection from environmental interference.

3. Results and Discussion

3.1. Stability of Solutions

According to the safety data sheets (SDS), nitrate test kits are based on the Griess method, using sulfanilamide and hydrochloric acid to form a colored azo dye for nitrite detection, while ammonia test kits rely on the salicylate–hypochlorite method, using sodium salicylate, nitroprusside, and sodium hypochlorite to produce an indophenol blue complex. − After 1 week without shaking the reagent bottles, ammonia samples prepared with unshaken reagents showed negligible variation from those with freshly shaken reagents at 2 mg/L. In contrast, nitrate tests were inconsistent: the first test showed the expected color, but the next two failed to develop the characteristic red color (Figure ).

5.

Results of the developed color of three replicates from a single sample after not shaking the reagent container for 1 week.

Our one-week “no-shaking” test illustrates that static storage is sufficient for the PEG-based nitrate reagent to undergo noticeable stratification, producing inconsistent or incomplete color development. Although we did not test shorter or longer durations, the observation that a single week of undisturbed storage caused measurable changes indicates that even shorter periods may allow partial settling, while longer storage would be expected to increase stratification, consistent with the behavior shown in Figure . These findings align with broader studies of Griess-type nitrate assays. Pai et al. demonstrated that premixed Griess reagents are highly sensitive to reagent ratio, order of addition, incubation temperature, and reaction time, and that deviations from optimal conditions lead to reduced or unstable color formation. Murray et al. further showed that nitrate reduction followed by the Griess reaction is strongly influenced by mixing intensity and standing time, with inadequate or excessive mixing reducing the analytical signal. Together, these results indicate that both storage time and temperature-dependent reaction conditions can meaningfully affect nitrate color development, and that the reagent-handling steps in automated systems require careful control. To mitigate settling in the in situ sensor, we incorporated a 3 V vibrator to mix Reagent 2 in the reagent reservoir before each test.

3.2. Optimizing Reagent Ratios

Figure illustrates the impact of varying reagent ratios on the resulting color temperature across three ammonia concentrations. The primary aim of this test was to evaluate the flexibility and tolerance of the sensor system to deviations from the recommended reagent ratios, thereby assessing how much inaccuracy in reagent dosing can be tolerated before compromising measurement reliability. The two reagents, solution 1 (containing sodium salicylate and sodium nitroprusside) and solution 2 (containing sodium hypochlorite and sodium hydroxide), must be combined in the correct ratio to ensure full color development through a redox reaction, forming the colored complex.

6.

Color temperature results of various ammonia reagent ratios on three different concentrations are shown as the number of drops of reagent 1/number of drops of reagent 2. Error bars show the standard error.

At lower ammonia concentrations (0.5 and 2 mg/L), the system showed a high degree of robustness: even with moderate deviations in the reagent ratio (±25%), the resulting color temperature values remained largely close to the expected target. This suggests that the method is forgiving of small dosing inaccuracies in typical environmental applications, where ammonia concentrations are generally below 10 mg/L.

At the 2 mg/L level, underdosing solution 2 to 50% (e.g., using 8/4 or 12/4 ratios) resulted in lighter color development and significantly lower color temperature values. This result highlights a critical sensitivity to the oxidizing component of the reaction, suggesting that a shortage of sodium hypochlorite can limit the full conversion of salicylate to its oxidized chromophore form. In contrast, overdosing either solution or underdosing solution 1 had a much smaller impact on the final result, with readings remaining within the acceptable range.

At the highest concentration tested (13 mg/L), variability became more pronounced, but this is outside the typical range encountered in surface waters and thus does not affect the sensor’s expected real-world performance. Overall, the test confirms that the system tolerates moderate dosing inaccuracies up to 50%, and it is particularly sensitive to the underdosing of the oxidizing reagent.

3.3. Hand-Held Sensor

Ammonia solutions with concentrations ranging from 0 to 1 mg/L were tested in the lab using the hand-held device sensor, and the color temperature values were compared against the NH₃ concentrations. The results showed a positive linear relationship between ammonia concentrations and color temperature values (Figure a). However, when the same experiment was conducted at different turbidity levels, the calibration line shifted, indicating the influence of turbidity on the measurements. To address this, the sensor’s blank reading (without reagents) was correlated with turbidity levels, establishing a multilinear relationship between color temperature values obtained with and without reagents and the ammonia concentrations. Using this multilinear model (see eq ), the predictive capability of the hand-held device was R ² = 0.96 (Figure c).

7.

(a) Lab calibration results of ammonia using hand-held device sensor with different turbidity values resulted in a multilinear relationship (n = 6 for each turbidity), (b) lab calibration of nitrate samples with different turbidity levels using the hand-held device sensor resulting in a logarithmic relationship (n = 35), (c) actual versus predicted values of the lab data and environmental samples from Yarra River and Lang Lang river in Melbourne, Australia dosed with different percentage of wastewater to increase the range of test (n = 33), and (d) actual versus predicted values of lab data and environmental samples from Yarra River and Lang Lang River in Melbourne, Australia (n = 15). NH₃ refers to TAN.

While the hand-held device provided a low-cost proof-of-concept platform for evaluating colorimetric nutrient detection, it was not intended as the final optical configuration for field deployment. The bottom-mounted LED arrangement was compatible with standard Eppendorf tubes and compact prototyping, but it increased sensitivity to particle settling during the color-development period. This limitation was reflected in the turbidity experiments and informed the design of the continuous in situ sensor, in which the LED and TCS sensor were positioned opposite each other across the chamber to provide a direct transmission path and reduce sedimentation-related optical interference.

Nitrate samples (Figure b) showed a negative logarithmic relationship with color temperature values. However, when different turbidity levels were tested, it was observed that the results closely followed the lab calibration curve for nitrate. Using the developed calibration models from the lab data, predictions were made for environmental samples collected from two rivers in Melbourne (Yarra River and Lang Lang River). These predictions achieved near-perfect accuracy, with an R ² value of 0.96 for both nitrate and ammonia (Figure c,d).

These results demonstrate that the hand-held device sensor performs reliably with environmental samples, which are more complex than synthetic laboratory samples. Additionally, the sensor performed well across a broader concentration range (up to 12 mg/L) for both of the contaminants. However, one outlier was observed in each case, where the predictions underestimated ammonia and overestimated nitrate concentrations, highlighting areas for further refinement of sensor performance.

3.4. In Situ Sensor

3.4.1. Pumping Volumes Required to Flush the Chamber

By pumping up to 20 pore volumes (3.13 mL) of fresh water into predeveloped ammonia solutions (13 mg/L and 1 mg/L), we determined that using 8 pore volumes, equivalent to 25 mL, was sufficient to reset the color temperature value back to that of the blank solution, as shown in Figure . Repeated measurements during laboratory optimization and field operation (more than 400 cycles across five months) showed no progressive drift or accumulation effects, indicating that 8 pore volumes remained adequate over many consecutive cycles. The CT values in this flushing experiment should not be interpreted as a calibration response, as early flushing stages may contain mixed residual and fresh solutions, and CT is derived from the RGB balance rather than direct color intensity. The main result is that after approximately eight pore volumes, the CT response converged close to the blank baseline, indicating sufficient flushing before the next measurement cycle.

8.

Number of pore volume (3.13 mL) injected into the chamber with 1 and 13 mg/L NH₃ to evaluate the optimum number of pore volume needed to purge the previous solution out of the chamber and fill it with the fresh solution. Note that this experiment evaluates flushing efficiency rather than sensor calibration. Early CT values may reflect mixed residual and fresh solutions, and the RGB-derived CT response can be affected by residual color and reagent background. After approximately eight pore volumes, the CT values converged close to the blank baseline. NH₃ refers to TAN.

Consequently, for each measurement cycle, the sensor generates approximately 25 mL of a waste solution, which must be disposed of properly. Therefore, with a 2 h sampling frequency, the sensor can operate continuously for 14 days using a 4.2 L waste container and a single 37 mL bottle of reagent, without requiring any maintenance. The estimated in situ sensor hardware cost was approximately USD 126.55, and one USD 16 reagent kit supported approximately 2 weeks of operation, corresponding to approximately USD 0.09 per sample. In comparison, commercial in situ nutrient analyzers and optical nitrate sensors are typically sold at substantially higher capital costs, often on the order of several thousand to tens of thousands of USD, before consumables, maintenance, and servicing are considered. However, this also highlights a limitation of the current system, as the waste collection unit must be emptied and maintained every 2 weeks to ensure uninterrupted operation. Future work should focus on minimizing the volume of tubing and chambers to reduce waste production, ultimately enhancing system efficiency and sustainability while lowering maintenance demands.

3.4.2. Lab Calibration

The results of the in situ sensor showed that, unlike the hand-held device, ammonia measurements did not exhibit significant variability across different turbidity levels; this is evidenced by the strong correlation between CT values and ammonia concentration shown in Figure a. While the slope of the regression line for ammonia increased with the in situ sensor, the slope of the regression line for nitrate decreased significantly (Figure b). Additionally, the relationship between nitrate concentration and color temperature shifted from logarithmic to linear. Neither ammonia nor nitrate demonstrated significant differences in regression or relationship trends across varying turbidity levels and solution concentrations; however, the uncertainties for both measurements increased.

9.

Lab calibration data for in situ sensor using 0 to 250 NTU turbidity samples for (a) ammonia (n = 17), and (b) nitrate sensors (n = 17), (c) actual versus predicted values of ammonia from environmental samples spiked with various wastewater percentages to increase the range (n = 8), and (d) actual versus predicted values of nitrate from environmental samples (n = 8). Predicted values were obtained by measuring the color temperature in real samples and applying the calibration equation derived in Figure a,b. These predictions were then compared against laboratory-measured ammonia and nitrate concentrations to assess the model’s accuracy. NH₃ refers to TAN.

These differences in the sensor response are largely attributable to the physical and optical distinctions between the two systems. In the hand-held device, the LED is mounted at the bottom of the Eppendorf tube, while the TCS sensor is positioned on the side. After mixing and the 5 min color-development period required by the API method, suspended particles can settle toward the bottom of the tube and accumulate directly in front of the upward-facing LED. Because the detector receives primarily side-scattered rather than transmitted light, this settling alters both the intensity and the angular distribution of the light reaching the sensor. These effects amplify turbidity-related variability and necessitate the use of the multilinear correction model.

In contrast, in the in situ sensor, the LED and TCS sensor face each other across the 3.13 mL chamber, establishing a clear transmission path. Settled solids are therefore far less likely to interfere with the optical path, making this design markedly less susceptible to turbidity. These geometric and operational differences explain why ammonia measurements in the in situ system exhibited minimal variability across turbidity levels, unlike the hand-held configuration, and why the nitrate calibration relationship shifted from logarithmic to linear in the in situ design.

Repeatability of the continuous in situ sensors was assessed using replicate measurements across the calibration range. For the continuous in situ ammonia/TAN sensor, the relative standard deviation (RSD) ranged from 0.66% to 7.31% for nonzero standards, while the blank showed an RSD of 9.36%. For the continuous in situ nitrate sensor, the RSD ranged from 0.36% to 4.01% across the tested standards. These results indicate the generally acceptable short-term repeatability of the optical measurements under the tested conditions.

3.4.3. Validation in Lab Using Field Samples

Figure c shows the performance of the ammonia sensor when testing environmental samples spiked with varying percentages of wastewater (0 to 10%). At lower concentrations, the sensor exhibited better performance with reduced uncertainty. However, at higher concentrations, the sensor tended to overestimate the ammonia concentration. Overall, the in situ sensor achieved a high R² value of 0.94, which is particularly impressive given that the predictions were based solely on calibration using laboratory data. Taylor et al. reported that ammonia concentrations in urban stormwater typically range from 0.20 to 0.29 mg/L during both baseflow and storm events. The sensor demonstrated measurable and consistent responses within this concentration range, suggesting its potential applicability to urban stormwater monitoring.

In contrast, the nitrate sensor performed poorly in predicting nitrate concentrations in the environmental experiments (Figure d), showing high levels of uncertainty. This uncertainty may be attributed to the reagent from bottle two, which requires proper shaking before use.

Although a 3 V vibrator was added to reagent 2 to mitigate settling, this proved insufficient. The PEG-based nitrate reagent requires vigorous whole-volume shaking, whereas the small vibrator generated only local motion and could not homogenize reagent already in the tubing. As a result, the effective concentration of reagent 2 likely varied between measurement cycles.

In our system, the in situ sensor further differs from the hand-held arrangement in ways that exacerbate these sensitivities: sequential peristaltic delivery, weaker internal mixing, a shorter and more complex optical path, and potential adsorption or aging of viscous nitrate reagents in plastic tubing. Additionally, API nitrate test kits themselves may be unreliable in natural waters, particularly at low concentrations, suggesting that matrix effects compound the limitations of automated delivery.

Together, these factors explain why the hand-held device produced accurate nitrate measurements while the in situ system performed poorly (R ² = 0.07). Future work will require redesigned reagent reservoirs and mixing mechanisms, improved chamber geometry, or alternative chemistries to achieve reliable continuous nitrate sensing.

3.4.4. Validation of Sensor In Situ and In-Field

According to Figure a, the ammonia concentrations measured in the field, using a calibration curve derived from laboratory experiments, were underestimated, particularly at higher concentrations (≥4 mg/L). Despite this, the field validation showed strong predictive performance, with an R ² of 0.80 and a corresponding NSE of 0.74. The ammonia R ² values obtained in this study are consistent with performance commonly reported for practical low-cost environmental sensors operating under real-world conditions. , Calibration of low-cost environmental sensors is particularly challenging because sensor response is strongly influenced by environmental conditions, cross-sensitivities, sensor drift, and variability in sensing materials, all of which can reduce accuracy when sensors are deployed outside controlled settings. These practical limitations mean that even well-calibrated sensors frequently show weaker correlations in complex, unsteady field environments compared with laboratory characterization. These results highlight the sensor’s robustness, especially considering that it was calibrated solely with laboratory data yet performed well under real field conditions.

10.

Results of measured NH₃ in the field experiment using the in situ sensor versus the actual values measured with lab analysis using (a) correlation developed lab calibration (n = 27), and (b) correlation developed using field calibration (n = 27). NH₃ refers to TAN.

To further improve the correlation, we used data from our initial field experiment to develop a field-specific calibration curve. Applying this new calibration curve to the rest of the field experiment measurements (Figure b) resulted in improved accuracy. The revised approach yielded an R ² value of 0.86 and an NSE of 0.83, demonstrating a closer match to actual field conditions compared with the lab-calibrated values. This improvement primarily reflects the use of field-derived data for both calibration and validation, which better captures site-specific conditions rather than indicating an inherent enhancement in sensor capability.

These findings highlight the importance of using a field-specific calibration curve when deploying sensors in real-world environments. Unlike controlled lab conditions, field deployments are subject to various external factors, such as temperature fluctuations, interference from other compounds, and long-term sensor stability. , Future work should focus on further enhancing sensor performance in the field by mitigating these environmental influences, improving calibration methods, and optimizing sensor design, such as temperature compensation, to enhance both accuracy and longevity.

3.5. Limitations and Future Work

While the proposed sensor demonstrates reliable continuous monitoring of ammonia under field conditions, some limitations should be acknowledged. Although a formal fluid dynamics analysis of the magnetic stirrer was not conducted, the stable ammonia response over more than 400 laboratory and field measurement cycles indicates that mixing was sufficient for this chemistry; nevertheless, future work should quantitatively evaluate mixer efficiency. While the hand-held device produced reliable nitrate measurements (R ² = 0.93), the in situ sensor suffered from performance degradation due to reagent settling, incomplete homogenization, and chamber-scale optical constraints. In addition, the in situ sensor was evaluated under field-based but not fully autonomous conditions, as stormwater samples were manually homogenized prior to analysis; thus, true hands-free operation with direct environmental intake remains to be demonstrated. Finally, although turbidity was explicitly tested as a controlled optical interference, the sensor was also evaluated using real environmental water samples that contained naturally occurring matrix variability, including background color, suspended matter, dissolved organic matter, and coexisting ions. Therefore, the validation data set provides a practical assessment of sensor performance under realistic mixed-matrix conditions. However, individual chemical interferences and spiked recoveries across specific wastewater or industrial matrices were not systematically assessed. Future work should include targeted interference testing and recovery experiments across a wider range of wastewater matrices.

Another limitation is that the in situ sensor measured unfiltered water, whereas reference FIA analysis was performed on samples filtered through a 45 μm filter. Filtration was required in the field to preserve the dissolved nutrient fraction and minimize biological transformation of ammonia and nitrate before laboratory analysis; however, paired filtered and unfiltered measurements were not collected in this study. Therefore, any potential bias introduced by comparing unfiltered sensor measurements with filtered laboratory reference concentrations could not be quantified. Future work should include paired filtered/unfiltered sampling to evaluate whether particulate matter or microbial activity influences the comparability between in situ and laboratory measurements.

While reagent settling effects were examined after 1 week without shaking, the impact of longer storage times, temperature variations, and aging pathways was not systematically assessed. Finally, although no tubing or pump failures were observed during the multimonth deployment, occasional faults related to 3D-printed components occurred; this is attributed to the use of 3D printing for rapid prototyping, and future work will transition to more robust, commercial-grade materials and manufacturing processes to improve durability and support long-term deployment.

4. Conclusion

This study demonstrated the potential of a low-cost, hand-held device and a 3D-printed colorimetric sensor for monitoring ammonia and nitrate levels in water using commercial aquarium test kits. Field validation showed promising results for ammonia detection, with an R ² of 0.86 and an NSE of 0.83, highlighting the sensor’s suitability as an affordable alternative to expensive commercial systems. Further investigation revealed that the reagent ratio plays a critical role in measurement accuracy, particularly for ammonia, where reagent 2 must be used in at least 50% of the manufacturer-recommended amount (8 drops) to ensure reliable results. In addition, exposure to sunlight was identified as a source of optical noise, emphasizing the need to shield the sensor during deployment. Field experiments also indicated that applying a calibration curve developed specifically from field data improved the sensor’s performance. This is likely due to additional environmental factors such as temperature variations and the presence of other ions that influence the chemical reactions. Overall, the platform shows promise for detecting a broader range of pollutants using alternative reagents. Future work should focus on minimizing chemical consumption, optimizing reagent handling, reducing the size of the reaction chamber and tubing, and evaluating the long-term stability and durability of the sensor in diverse environmental conditions.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsestwater.6c00349.

• A complete schematic view of all of the connections of the in situ sensor (Figure S1); field validation results from November 2024 to March 2025 (Figure S2); effect of sunlight on color temperature blank reading (Figure S3); estimated hardware bill of materials (Table S1); laboratory analysis results of nitrogen species (Table S2); and replicate measurement data and calibration results (Tables S3, and S4) (PDF)

M.J.: Conceptualization, methodology, software, formal analysis, data curation, writingoriginal draft, visualization. B.S.: Conceptualization, formal analysis, data curation, resources, writingreview & editing. D.M.: Conceptualization, methodology, formal analysis, writingreview & editing, supervision.

The authors declare no competing financial interest.