Shahin Tabandeh

Urban sensor networks can provide spatially and temporally resolved environmental information, but their measurement trustworthiness depends on maintaining an unbroken chain of traceability to the International System of Units (SI) and on quantifying the associated uncertainties under dynamic field conditions. This paper applies a Laplace-domain method for in situ calibration of city-wide sensor networks. The method estimates pairwise dynamic relationships between sensor signals, separates a medium-related spatial effect, and solves an overdetermined calibration problem with a known network reference. Measurement uncertainty is propagated using a Monte Carlo approach. The method is evaluated using two urban datasets: a Helsinki network linked to the SMEAR III station and an Antwerp network from the SensEURCity project. Calibration coefficients were obtained for temperature, relative humidity (RH), and pressure in both networks. Particulate matter (PM) calibration was feasible only under restricted sensor-subset conditions because of the localized and weakly correlated nature of PM signals. A transfer function-based malfunction indicator was also tested by introducing artificial shifts, showing that shifted measurement signals can be identified when the network geometry and signal correlations are sufficient. The results demonstrate the potential of Laplace-domain modeling as an uncertainty-aware route for dynamic calibration transfer in deployed sensor networks, while highlighting limitations related to interpolation, spatial heterogeneity, network geometry, and measurand-specific correlation.

Black carbon (BC) is an important urban air pollutant of emerging concern with documented health and climate effects, yet direct BC measurements remain unavailable at the majority of automatic monitoring stations in typical urban sensor networks. Virtual sensing has been recently proposed as a complementary approach, using statistical relationships between routinely measured air quality parameters to provide indicative estimates of unmeasured quantities. In this study, regression models based on multiple linear regression (MLR), random forests (RFs) and support vector regression (SVR) are explored for equivalent BC estimation using reference-grade signals from a newly established urban background pilot supersite (Ada Marina, Belgrade). Three seasonal cases were considered: heating season, non-heating season, and the complete ~1-year dataset. Predictors for the models were derived using two approaches: a greedy algorithm based on stepwise linear regression, and a non-greedy algorithm based on adaptive best subset selection, yielding: NOx and CO for the heating season; NO₂ and PM_2.5 for the non-heating season; and NOx and PM_2.5 for the complete period. Models achieved the following R² and RMSE performance metrics for 50/50 training/test split: heating season R² = 0.87–0.94 and RMSE = 0.65–0.95 μg/m³; non-heating season R² = 0.57–0.60 and RMSE = 0.87–0.90 μg/m³; and complete period R² = 0.70–0.72 and RMSE = 0.78–0.80 μg/m³. Model performance is discussed in the context of published BC virtual sensor results from other European cities. A network applicability assessment indicates that the majority of Belgrade monitoring stations record the predictor signals required by the developed models, suggesting potential applicability for indicative BC estimation across the monitoring network, pending site-specific validation. We also present a preliminary, exploratory assessment of the models’ transferability to two external sites, where models achieved 0.6–0.8 R² for the heating season, and 0.5–0.6 R² for non-heating and the complete period.

Low‐cost sensors are increasingly being used, and also considered for usage in regulatory contexts like air quality monitoring. Proper sensor calibration and uncertainty evaluation is necessary in such cases. With the advance of mobile sensors, for example, installed on top of a car, new test and calibration paradigms are required, in particular if the sensor and reference instrument are only compared during short collocation times, whereby the collocation time is shorter than the averaging time of the reference instrument and the underlying process is varying over time. In this paper it is shown how sensors can be tested and calibrated in such situations. A first result of this paper are analytical expressions for the variance and covariance of averages of Gaussian processes. Next, performance testing of sensors based on short collocation periods is addressed. It is shown that the process variability should be taken into account for best results in terms of either small sample sizes or preventing unnecessarily rejecting well functioning sensors. Finally, a new regression procedure, called corrected generalized distance regression (C‐GDR), is developed, and its performance is evaluated and compared with established methods in a numerical study. The parameters in this study are based on a dataset containing real measurement data of an sensor together with the data from a reference instrument. The parameters are then varied to cover more difficult cases as well. It is concluded that in a situation in which the uncertainty of the reference instrument is much lower than that of the other instrument, ordinary least‐squares regression performs properly. However, when uncertainties become more similar and the measured process starts varying more, the C‐GDR approach outperforms the other methods.

Trace water is one of the most critical matrix contaminants in ultra-high-purity (UHP) process gases, like argon (Ar), nitrogen (N₂), and many others. Even trace amounts can severely degrade the quality of many products that are reliant on these gases. Despite its importance to advanced technology sectors, notably semiconductor manufacturing, it has proven quite difficult to realize preparative or analytical trace water metrology over the full amount fraction range needed or in the broad spectrum of industrially relevant matrix gases. Within the EU-funded PROMETH2O project consortium, this challenge has been addressed through the development or significant improvement of traceable measurement methods and standards spanning 5 nmol⋅mol⁻¹ to 5 µmol⋅mol⁻¹, tailored for use in UHP process gas production, such as Ar, N₂ and clean dry air (CDA). The measurement ranges were extended and the uncertainties were improved while being consistent with the current best practice at primary humidity standard laboratories. The developed standards provide combined standard uncertainties ranging from approximately 0.4 % to 1.5 % in water vapor amount fraction and from 0.03 °C to 0.07 °C in frost-point temperature, while the comb-assisted CRDS system achieves detection limits in the sub-ppb to ppt range. These capabilities were validated in applications that are relevant to process instrumentation and the gas industry. A distributed metrological infrastructure at various European national metrology institutes and partner sites now provides SI-traceable trace water measurements in UHP gases, strongly supporting and extending the calibration capabilities for the gas and semiconductor industries and the associated stakeholders.

Black carbon (BC) is an important urban air pollutant of emerging concern with documented health and climate effects, yet direct BC measurements remain unavailable at the majority of automatic monitoring stations in typical urban sensor networks. Virtual sensing has been recently proposed as a complementary approach, using statistical relationships between routinely measured air quality parameters to provide indicative estimates of unmeasured quantities. In this study, regression models based on multiple linear regression (MLR), random forests (RF) and support vector regression (SVR) are explored for BC estimation using reference grade signals from a newly established urban background pilot supersite (Ada Marina, Belgrade). Three seasonal cases were considered: heating season, non-heating season, and the complete ~1 year dataset. Predictors for the models were derived using two approaches: greedy algorithm based on stepwise linear regression, and non-greedy algorithm based on adaptive best subset selection, yielding: NOx, CO for the heating season; NO₂, PM_2.5 for the non-heating season and NOx, PM2.5 for the complete period. Models achieved the following R² and RMSE performance metrics for 50/50 training/test split: heating season R² = 0.87–0.94, RMSE = 0.65–0.95 μg/m³; non-heating season R² = 0.57–0.60, RMSE = 0.87-0.90 μg/m³; and complete period R² = 0.70–0.72, RMSE = 0.78–0.80 μg/m³. Model performance is discussed in the context of published BC virtual sensor results from other European cities. A network applicability assessment indicates that the majority of Belgrade monitoring stations record the predictor signals required by the developed models, supporting the potential for indicative city-wide BC estimation as a complement to direct on-site measurements.

Trace water is one of the most critical matrix contaminants in ultra-high purity (UHP) process gases, like argon (Ar) and nitrogen (N₂), and many others. Even trace amounts can severely degrade the quality of many products reliant on these gases. Despite its importance to advanced technology sectors, notably semiconductor manufacturing, it has proven quite difficult to realize preparative or analytical trace water metrology over the full amount fraction range needed or in the broad spectrum of industrially relevant matrix gases. Within the EU-funded PROMETH2O project consortium, this challenge has been addressed through the development or significant improvement of traceable measurement methods and standards spanning 5 nmol mol⁻¹ to 5 µmol mol⁻¹, tailored for use in UHP process gas production, such as Ar, N2 and hydrogen (H2). The measurement ranges were extended and the uncertainties were improved, while being consistent with current best practice at primary humidity standards laboratories. These capabilities were validated in applications relevant to process instrumentation and the gas industry. A distributed metrological infrastructure at various European National Metrology Institutes and partner sites now provides SI-traceable trace water measurements in various UHP, strongly supporting and extending the calibration capabilities for the gas and semiconductor industries and the associated stakeholders.

Machine learning and its associated tools are an indispensable part of sensor network metrology at all levels. Modern machine learning methods are becoming increasingly sophisticated, albeit at the cost of human interpretability. Although the concurrent development of methods to add human-understandable explanations to Machine learning models helps alleviate this trade-off, the influence of uncertainty in the training data on the interpretability or explainability of the model itself requires further study. In this contribution we study the influence of measurement uncertainty on the explainability of a machine learning based soft-sensor that provides estimates for the ozone concentration in a region based on inputs from co-located meteorological and air-quality sensors. Using a customized uncertainty model and a Monte Carlo simulation, the influence of uncertainty on explanations in the form of feature attributions provided by the SHapley Additive exPlanations (SHAP) framework is studied. The analysis is performed for two different classes of tree-based ensemble methods and the influence of the geographical placement of sensors in the network is taken into account.

Current temperature sensors require regular recalibration to maintain reliable temperature measurement. Photonic/quantum-based approaches have the potential to radically change the practice of thermometry through provision of in situ traceability, potentially through practical primary thermometry, without the need for sensor recalibration. This article gives an overview of the European Partnership in Metrology (EPM) project: Photonic and quantum sensors for practical integrated primary thermometry (PhoQuS-T), which aims to develop sensors based on photonic ring resonators and optomechanical resonators for robust, small-scale, integrated, and wide-range temperature measurement. The different phases of the project will be presented. The development of the integrated optical practical primary thermometer operating from 4 K to 500 K will be reached by a combination of different sensing techniques: with the optomechanical sensor, quantum thermometry below 10 K will provide a quantum reference for the optical noise thermometry (operating in the range 4 K to 300 K), whilst using the high-resolution photonic (ring resonator) sensor the temperature range to be extended from 80 K to 500 K. The important issues of robust fibre-to-chip coupling will be addressed, and application case studies of the developed sensors in ion-trap monitoring and quantum-based pressure standards will be discussed.