More Than Just Specifications - The Research and Testing Behind Our Sensor Selections

We would like to give a big thank you to Dr. Lekan Popoola, Senior Research Associate at the University of Cambridge for assisting our science team with the following research. Also, a big thank you to our internal science team, led by Dr. Anika Krause, for their continued scientific rigour throughout this research process.

Within the next few months, we at AirGradient are preparing to release a new outdoor air quality monitor, the AirGradient Open Air Max. This monitor is intended for applications requiring reliable, standalone outdoor measurements, such as detailed urban air quality mapping or research projects. It builds upon our previous Open Air design by incorporating several new features and sensors that make it better suited to professional requirements.

Like our Open Air monitor, the Open Air Max will measure particulate matter, CO₂, TVOCs, temperature, and relative humidity (RH). Adding Nitrogen Dioxide (NO₂) and Ozone (O₃) sensors was also a direct response to common requests from customers, who needed data on traffic-related emissions and other key outdoor pollutants. Accurately measuring these gases is important for assessing outdoor air quality, particularly in urban areas. Crucially, we aimed to integrate these professional-grade sensors while keeping the Open Air Max relatively affordable.

However, integrating new sensors, especially for reactive gases like NO₂ and O₃, requires careful consideration. The market includes various sensors with different technologies, performance characteristics, and price points. Selecting appropriate sensors for the Open Air Max involved extensive research and testing to evaluate their suitability to measure reliably in real-world conditions. This is critical because unlike controlled laboratory environments – which often use fixed temperatures, humidity levels, and only the target gas – real-world ambient conditions are complex. Therefore, rigorous testing against reference instruments in these dynamic outdoor environments was essential. While this required significant time and effort, we wanted to ensure the sensors we and our customers rely on have acceptable accuracy and reliability.

“Drawing on my years of research in atmospheric science and air quality, it’s clear that selecting a low-cost sensor (LCS) for monitoring air pollutants in ambient environments demands thoughtful consideration. The right choice often depends on the intended application—be it source identification, citizen science, community engagement, or scientific research.

While original equipment manufacturer (OEM) sensors may perform adequately under controlled test conditions, their integration into LCS platforms for ambient air quality monitoring can lead to suboptimal results. This is often due to environmental influences and, at times, substandard device configuration. Therefore, sensor selection should go beyond manufacturer specifications and take into account real-world performance.

To support the LCS community, it is essential to share insights into how these sensors perform in practical settings. This blog highlights our evaluation of LCS sensors in measuring two key air pollutants: nitrogen dioxide (NO₂) and ozone (O₃).”

• Dr. Lekan Popoola, Senior Research Associate at the University of Cambridge.

As we’ve now concluded our research and decided on the sensors we will be using, we wanted to share our results with you. We wanted to do this not only to show just how much research goes into each sensor decision we make but also to discuss transparently what led to our final decision. If you’re interested in our evaluation process and the data from our research, please read on!

The Contenders

Our initial research surveyed several commercially available electrochemical sensors for NO₂ detection. Finding a reliable and affordable sensor suitable for widespread deployment required balancing performance claims, technical specifications, cost, and practical considerations like manufacturer support. Here’s a brief overview of the candidates we initially considered and why we decided either to progress further with assessing each sensor or why we excluded them from the running:

1. SPEC Sensors (NO2 - 110-508): These sensors are very inexpensive, which would have been useful for our target price. However, available information and preliminary test results (including those from AQMD) indicated potentially poor performance for ambient monitoring. Their specifications listed a very high Limit of Detection (LOD) of 30 ppb, which is often above typical urban NO₂ background levels, and confirmed they measure both NO₂ and O₃ (an Ox sensor). Coupled with difficulties getting technical feedback from the manufacturer, we did not pursue these further for the Open Air Max.

2. Amphenol SGX Sensortech (PS1-NO2-2): This is another relatively low-cost option. At a good price point and with decent specifications, this sensor was a contender for the Open Air Max. Unfortunately, they never responded to our queries about the sensor, and when we asked for test data, we received no reply, making it difficult to evaluate their real-world performance.

3. Winsen (ZE12A-NO2): This sensor is much more expensive (the most expensive we looked into). It also has no Ozone filter, meaning it functions as an “Ox” sensor, measuring a combined signal from both NO₂ and O₃. While Ox sensors have their uses, accurately deriving the NO₂ concentration requires a separate, reliable O₃ measurement and a robust correction algorithm. For broader applicability, we prioritised sensors designed specifically for NO₂ measurement or those with established correction methods.

This initial review highlighted two leading contenders that seemed most promising for deeper evaluation based on their specifications, market reputation, and potential suitability for our requirements:

1. EC Sense (TB600B-NO2-2): A relatively new entrant to the market at a medium price point. Specifications-wise, these sensors are comparable to established sensors (LOD 5 ppb, Resolution < 5 ppb) and EC Sense has claimed potential advantages like longer lifetime (>3 years) and stability due to their use of polymer electrolytes instead of traditional aqueous ones. These claims warranted practical investigation.

2. Alphasense (NO2-A43F): A well-established and widely used sensor in air quality monitoring applications with a similar price to EC Sense. This specific model is designed with an internal Ozone filter, aiming to minimise cross-sensitivity and provide a more direct NO₂ measurement. Its reputation and the built-in filtering made it a strong benchmark candidate.

Based on this information, we decided the next step was to carry out a side-by-side comparison under real-world conditions. Our detailed testing phase, therefore, focused primarily on evaluating the performance of EC Sense versus Alphasense sensors for both NO₂ and the corresponding O₃ sensors from each manufacturer. The rest of this article details the setup and results of that comparative deployment.

Our Evaluation Setup

To make a final (and informed) decision, theoretical specifications and manufacturer claims are not enough. We needed to see how these sensors performed in real-world, ambient conditions. We conducted a co-location study to achieve this, deploying the candidate sensors directly alongside high-grade reference instruments.

• Location: The sensors were installed at an established air quality monitoring site operated by the University of Cambridge. This provided access to reliable, research-grade reference measurements for NO₂ and O₃, serving as the baseline for our comparison.

• Duration: The deployment ran through typical winter conditions (UK) from early December 2024 to late February 2025. Note that this period doesn’t cover high temperatures, which might also affect the sensor performance. But it’s sufficient for an initial comparison.

• Sensors Tested: We deployed two units of each primary sensor type under evaluation:

  • Two Alphasense NO₂ sensors (model NO2-A43F)
  • Two Alphasense O₃ sensors (model OX-A431)
  • Two EC Sense NO₂ sensors (model TB600B-NO2-2)
  • Two EC Sense O₃ sensors (model TB600B-O3)

Deploying pairs of sensors helps assess unit-to-unit variability and ensures conclusions aren’t based on a single potentially anomalous sensor. However, we tested two of each sensor.

• Data Collection & Analysis: The Airgradient low-cost nodes that incorporated these sensors were configured to record measurements every 5 minutes. This high-resolution data was important for observing rapid changes and identifying characteristics like signal noise. However, we primarily aggregated the sensor data into hourly averages to compare overall trends and calculate performance metrics against the reference instruments (which often report at hourly intervals). This approach provides clearer visualisations and aligns with common air quality reporting standards. We will specifically refer back to the 5-minute data when discussing phenomena best observed at that finer timescale. Note that the data gaps in the time series results for the EC Sense are due to poor Wi-Fi connectivity affecting data transmission.

With this setup, we could directly compare the output of the Alphasense and EC Sense sensors against trusted reference measurements under identical environmental conditions.

First Impressions

Before delving into hourly performance metrics, examining the raw, high-resolution (5-minute) data provides valuable initial insights into sensor behavior. During this phase, two key characteristics immediately stood out: signal noise and data truncation.

Please note that for the later comparisons in this article, we opted to use the better-performing sensor from each pair from this initial testing.

Signal-to-noise Ratio: When comparing the 5-minute readings, a noticeable difference in signal stability emerged. The EC Sense NO₂ sensors exhibited considerably more noise in their output compared to the Alphasense sensors. While all electrochemical sensors have some inherent signal noise, the fluctuations in the 5-minute EC Sense NO₂ readings were significant, on the order of 5-8 ppb peak-to-peak (as observed in the second set of graphs above).

In contrast, the raw voltage signals from the Alphasense sensors (as seen in the first set of graphs) appeared smoother relative to the reference trends. While hourly averaging smooths out much of this noise, high raw signal noise can sometimes indicate underlying instability, thereby limiting the sensor’s ability to detect small, rapid changes in concentration.

Data Truncation in EC Sense Ozone: A more critical observation arose from the EC Sense O₃ sensors’ 5-minute data. We consistently found that the sensors did not report any values below a threshold of approximately 9-10 ppb (as clearly shown in the second set of graphs above). The data output was effectively “flat-lining” at this level, even when the reference instrument simultaneously measured O₃ concentrations well below this value – levels which are still environmentally relevant.

These initial observations from the 5-minute data already flagged important differences. The higher noise in EC Sense signals suggested potential challenges in resolving finer concentration changes, while the O₃ data truncation represented a major limitation in capturing the full range of ambient ozone levels.

Hourly Performance (Out-of-the-box)

Moving to hourly averaged data allows for a clearer comparison of how well the sensors track ambient concentration trends against the reference instruments (since most reference instruments only report on an hourly basis). In this step, we evaluated the sensors using generic scaling factors derived from the deployment data but without applying specific corrections for cross-interferences between NO₂ and O₃. This involved a simple comparison of the peak-to-trough ratios of the plots shown above relative to the reference signals for each species. This represents a baseline or “out-of-the-box” performance level.

Nitrogen Dioxide (NO₂) Comparison

Alphasense: The Alphasense NO₂ sensors performed well right from the start. As shown in the time series and scatter plots above, the hourly Alphasense readings tracked the reference NO₂ concentrations closely. We calculated strong Pearson’s correlation coefficients (r) of around 0.88-0.89, indicating a good linear relationship.

The Root Mean Square Error (RMSE), a measure of the average prediction error, was also reasonably low at approximately 3-4 ppb. This good initial performance is likely aided significantly by the built-in ozone filter in the Alphasense model, which is designed to minimize interference from ozone, a known cross-interferent for electrochemical NO₂ sensors.

EC Sense: The results for the EC Sense NO₂ sensors, without any correction, were starkly different. The time series showed little resemblance to the reference NO₂ trends, and the scatter plots revealed virtually no correlation (Pearson’s r ≈ 0), as seen above. The RMSE was consequently very high, around 14-19 ppb. This demonstrates that the EC Sense NO₂ sensor, despite its name, does not selectively measure NO₂ out-of-the-box. Its signal appears heavily influenced by other pollutants, strongly suggesting it behaves primarily as an Ox (NO₂ + O₃) sensor. The inability to distinguish NO₂ from O₃ renders its uncorrected output unusable for determining actual NO₂ levels. More importantly, the comparative statistics are hugely negatively impacted by the truncations of the EC Sense signal below 10 ppb. A significant proportion of the true NO₂ values from the reference falls below the redaction value; this limits the data available to establish a reliable relationship with the reference.

Ozone (O₃) Comparison

Alphasense: The Alphasense O₃ sensors also showed generally good agreement with the reference instrument on an hourly basis. Correlations were strong (r ≈ 0.85-0.87), and the RMSE was acceptable at around 5-6 ppb. While some deviations were visible, particularly at lower concentrations, the overall tracking was satisfactory for an uncorrected sensor.

EC Sense: The performance of the EC Sense O₃ sensors was mixed and generally lagged behind Alphasense. While one sensor showed a reasonable correlation (r ≈ 0.85), the other was weaker (r ≈ 0.66), and RMSE values were higher (≈ 7-13 ppb). More importantly, the impact of the data truncation below ~10 ppb, first observed in the 5-minute data, remained clearly visible in the hourly averages. Both the time series and scatter plots show a distinct lack of data points below this threshold, confirming the sensor’s inability to capture lower O₃ concentrations. This truncation significantly limits the sensor’s usefulness, as it misses a relevant part of the typical ambient O₃ range.

The out-of-the-box comparison revealed significant performance differences. Alphasense NO₂ sensors provided immediately useful NO₂ data, while EC Sense NO₂ sensors do not output any useful raw measurements due to strong O₃ interference.

For O₃, Alphasense showed better baseline performance, while EC Sense was critically hampered by its low-concentration data truncation. This led us to investigate the potential improvements offered by cross-sensitivity corrections.

Cross-Sensitivity Corrections

Electrochemical sensors are often susceptible to interference from gases other than their target analyte. For NO₂ and O₃ sensors, O₃ can interfere with NO₂ measurements, and NO₂ can interfere with O₃ measurements. Applying correction algorithms can potentially improve accuracy, but the effectiveness and practicality depend on the sensor’s behavior and the availability of reliable data for the interfering gas.

Correcting EC Sense NO₂ for Ozone Interference

Given the EC Sense NO₂ sensor’s behavior observed in the out-of-the-box results, applying an ozone correction was essential to even attempt extracting a meaningful NO₂ signal. We investigated this by subtracting a 100%contribution of ozone from the sensor’s signal.

Initially, we attempted to use the data from the co-located EC Sense O₃ sensor for this correction. However, as established earlier, the EC Sense O₃ data was severely compromised by the truncation below ~10 ppb. This missing data made it unreliable for accurately correcting the NO₂ sensor, especially during periods of low ozone.

Therefore, to assess the potential best-case performance of the EC Sense NO₂ sensor, we applied a correction using the reference O₃ instrument data. The results (shown below) were dramatic. The correlation with reference NO₂ improved significantly (Pearson’s r > 0.8), and the RMSE dropped considerably (to around 4.5 ppb).

This confirms that the EC Sense sensor can provide a reasonable estimate of NO₂ for cases where the NO₂ reading is well above the truncation value, but only when its strong ozone interference is actively compensated for using accurate, concurrent ozone measurements. The critical limitation here is the reliance on external, reference-grade O₃ data to achieve this. In a standalone deployment of the Open Air Max, such reference data is not available. The sensor’s own compromised O₃ data was insufficient, making independent, accurate NO₂ measurement with this sensor problematic.

Correcting Alphasense O₃ for NO₂ Interference:

Ozone sensors can also exhibit some cross-sensitivity to NO₂. We evaluated the potential to improve the Alphasense O₃ readings by correcting for this interference. We could perform this correction using data from the co-located Alphasense NO₂ sensor (the A43F model), which already provides a good NO₂ estimate.

Applying this correction yielded a noticeable improvement in the Alphasense O₃ sensor’s performance. The correlation with the reference O₃ increased further (Pearson’s r > 0.93), and the RMSE decreased to approximately 3-4 ppb, indicating enhanced accuracy across the measured range.

The key advantage here is independence. This correction utilizes data readily available from the paired Alphasense sensor (NO₂ and O₃) within the same monitoring device. It doesn’t rely on external reference instruments, making it a practical method for improving accuracy in real-world, standalone deployments like the Open Air Max.

This analysis highlighted a vast difference in practicality. While the EC Sense NO₂ sensor required correction using secondary and often unobtainable (in situ) reference O₃ data to be useful, the Alphasense O₃ sensor benefited from correction using readily available data from its paired Alphasense NO₂ sensor. This capability for independent, self-consistent correction significantly favors the Alphasense sensor pair for standalone monitoring applications.

Final Assessment

After analyzing the out-of-the-box performance and the impact of corrections, we conducted a final head-to-head comparison based on the best achievable results for each sensor type under practical conditions.

NO₂ Performance Summary

We compared the Alphasense NO₂ sensor (using its internal filter, no external correction needed) against the EC Sense NO₂ sensor after it was corrected using reference O₃ data (its best-case, though impractical, scenario).

• Alphasense NO₂ (Independent): Demonstrated good correlation (r ≈ 0.89) and accuracy (RMSE ≈ 3.2 ppb) independently, thanks to its built-in filter.
• EC Sense NO₂ (Corrected w/ Ref O₃): Achieved reasonable correlation (r ≈ 0.81) and accuracy (RMSE ≈ 4.4 ppb) only when corrected using external reference O₃ data.

While the corrected EC Sense performance approaches that of Alphasense, the crucial difference is independence. Alphasense delivers reliable NO₂ data on its own, whereas EC Sense requires external data unavailable in typical deployments, confirming its fundamental Ox behavior in this configuration.

O₃ Performance Summary:

We compared the Alphasense O₃ sensor (corrected using its paired Alphasense NO₂ data) against the EC Sense O₃ sensor (using its standard scaling/correction).

• Alphasense O₃ (Self-Corrected): Showed excellent correlation (r ≈ 0.93) and improved accuracy (RMSE ≈ 4.0 ppb) using data available within the sensor pair.
• EC Sense O₃: Performance was significantly lower (r ≈ 0.85, RMSE ≈ 7.0 ppb in the better example) and, most importantly, was fundamentally compromised by the data truncation below ~10 ppb. This truncation leads to a substantial loss of valid environmental data.

The comparison for O₃ is clear. The Alphasense sensor, especially when self-corrected, provides significantly more accurate and complete data across the relevant ambient concentration range. The EC Sense O₃ sensor’s truncation issue severely limits its utility for ambient monitoring.

Overall Conclusion

Based on the comprehensive data collected during the Cambridge co-location study (December 2024 - February 2025), the Alphasense sensors demonstrated superior overall performance and practicality for integration into the AirGradient Open Air Max:

1. Accuracy & Reliability: Alphasense provided more accurate readings, particularly for O₃, and captured the full range of observed concentrations without truncation.
2. Independence: The Alphasense NO₂ sensor (A43F) works effectively out-of-the-box due to its filter. The Alphasense O₃ sensor’s accuracy can be further improved using data from its paired NO₂ sensor, maintaining operational independence.
3. Data Integrity: Alphasense sensors provided continuous concentration data, whereas the EC Sense O₃ sensor suffered from significant concentration data loss at low concentrations due to truncation.
4. EC Sense Limitations: The EC Sense NO₂ sensor behaved as an Ox sensor, requiring external reference data for correction. The EC Sense O₃ sensor was unusable for low-concentration measurements due to truncation. The higher noise at 5-minute resolution was also a minor concern, limiting its ability to resolve low varying ambient concentrations which are encountered in background locations.

Therefore, based on this rigorous evaluation, we selected the Alphasense NO2-A43F and Alphasense OX-A431 sensors for integration into the AirGradient Open Air Max.

Feedback and Suggestions for EC Sense

We believe in constructive feedback and transparency. While the EC Sense sensors showed some potential, particularly regarding their new technology, which might offer benefits in longevity or stability (though not verifiable in our winter deployment), several aspects hindered their performance in our ambient air quality monitoring tests. We offer the following suggestions for consideration:

1. Address Data Truncation: The most critical issue observed was the hard coded truncation of O₃ data below ~10 ppb. In our understanding this is a data cleaning step from the manufacturer to mark the limit of detection and prevent physically impossible (i.e. negative) concentration values. However, we believe that there is still very useful data below the truncation line, especially if the factory calibration is wrong. For instance if the factory calibration is wrong by say a factor of 2, twice in reality the default “10 ppb” truncation will actually be truncating values below 20 ppb. We recommend providing access to the full, raw, untruncated sensor signal output.

2. Clarify NO₂ Sensor Behavior: Given its strong cross-sensitivity to ozone, labelling the TB600B-NO2-2 sensor simply as “NO₂” can be misleading for ambient applications. We suggest more transparently classifying it as an “Ox (NO₂ + O₃)” sensor or providing clear guidance and algorithms for ozone correction, acknowledging the potential need for a reliable, co-located O₃ measurement.

3. Investigate Signal Noise: While less critical than truncation, exploring the reasons for the higher signal noise observed at 5-minute resolution compared to other sensors could lead to further improvements in signal stability and responsiveness.

We appreciate the innovation EC Sense is bringing to the sensor market and hope these feedback contributes to the ongoing development and refinement of their products for demanding applications like ambient air quality monitoring.