Isoprene is a reactive hydrocarbon emitted to the atmosphere in large quantities by terrestrial vegetation. Annual total isoprene emissions exceed 300 Tg a −1 , but emission rates vary widely among plant species and are sensitive to meteorological and environmental conditions including temperature, sunlight, and soil moisture. Due to its high reactivity, isoprene has a large impact on air quality and climate pollutants such as ozone and aerosols. It is also an important sink for the hydroxyl radical which impacts the lifetime of the important greenhouse gas methane along with many other trace gas species. Modeling the impacts of isoprene emissions on atmospheric chemistry and climate requires accurate isoprene emission estimates. These can be obtained using the empirical Model of Emissions of Gases and Aerosols from Nature (MEGAN), but the parameterization of this model is uncertain due in part to limited field observations. In this study, we use ground‐based measurements of isoprene concentrations and fluxes from 11 field sites to assess the variability of the isoprene emission temperature response across ecosystems. We then use these observations in a Metropolis‐Hastings Markov Chain Monte Carlo (MHMCMC) data assimilation framework to optimize the MEGAN temperature response function. We find that the performance of MEGAN can be significantly improved at several high‐latitude field sites by increasing the modeled sensitivity of isoprene emissions to past temperatures. At some sites, the optimized model was nearly four times more sensitive to temperature than the unoptimized model. This has implications for air quality modeling in a warming climate.

Low-cost air quality sensors have shown great promise as a complement to high-cost reference and equivalent methods. Though not currently as accurate, their low barrier of entry and smaller form factor allow them to be deployed in greater numbers, thus enabling air quality measurements to be made at a far higher spatial and temporal resolution than previously possible. However, their measurements require corrections as they suffer from both short-term biases (e.g., changes in environmental conditions such as temperature and humidity), and long-term measurement drift due to degradation. Many studies have focused on calibration and re-calibration of sensors, but fewer focus on correcting pre-calibrated sensor measurements. Correcting measurements is a likely scenario for people buying off-the-shelf devices, as they will not have access to the raw data that underpins the measurements, such as sensor voltages. Previous studies focused on a small range of correction techniques, without accounting for the variances that can occur between devices or locations. This work aimed to perform a comprehensive assessment of different correction techniques applied to air quality sensor systems. More than 470,000 unique measurement corrections were tested across two sites to determine best practices for correction campaigns going forward, resulting in a far more robust study than previous works. It highlights the large variances in results that occurred between sites, particularly for NO2, with results often more impacted by device type and location than the regression technique used. Simpler linear models were also found to perform just as well as, and sometimes better than, more complex non-parametric techniques. This study highlights that, though a strong focus is often put on comparing different regression methods, the choice of technique has less impact than the configuration of the device or the conditions of the co-location site. Therefore, future studies should focus less on small-scale comparisons of regression techniques and more on how to improve the transferability and applicability of results from a co-location campaign to another.

We propose operational definitions and a classification framework for air quality sensor-derived data, thereby aiding users in interpreting and selecting suitable data products for their applications. We focus on differentiating independent sensor measurements (ISM) from other data products, emphasizing transparency and traceability. Recommendations are provided for manufacturers, academia, and standardization bodies to adopt these definitions, fostering data product differentiation and incentivizing the development of more robust, reliable sensor hardware.

Large passenger ships are characterised as enclosed and crowded indoor spaces with frequent interactions between travellers, providing conditions that facilitate disease transmission. This study aims to provide an indoor ship CO2 dataset for inferring thermal comfort, ventilation and infectious disease transmission risk evaluation. Indoor air quality (IAQ) monitoring was conducted in nine environments (three cabins, buffet, gym, bar, restaurant, pub and theatre), on board a cruise ship voyaging across the UK and EU, with the study conducted in the framework of the EU HEALTHY SAILING project. CO2 concentrations, temperature and relative humidity (RH) were simultaneously monitored to investigate thermal characteristics and effectiveness of ventilation performance. Results show a slightly higher RH of 68.2 ± 5.3 % aboard compared to ASHRAE and ISO recommended targets, with temperature recorded at 22.3 ± 1.4 °C. Generally, good IAQ (<1000 ppm) was measured with CO2 mainly varying between 400 and 1200 ppm. The estimated air change rates (ACH) and ventilation rates (VR) implied sufficient ventilation was provided in most locations, and the theatre (VR: 86 L s−1 person−1) and cabins (VR: >20 L s−1 person−1) were highly over-ventilated. Dining areas including the pub and restaurant recorded high CO2 concentrations (>2000 ppm) potentially due to higher footfall (0.6 person m−2 and 0.4 person m−2) and limited ACH (2.3 h−1 and 0.8 h−1), indicating a potential risk of infection; these areas should be prioritised for improvement. The IAQ and probability of infection indicate there is an opportunity for energy saving by lowering hotel load for the theatre and cabins and achieving the minimum acceptable VR (10 L s−1 person−1) for occupants' comfort and disease control. Our study produced a first-time dataset from a sailing cruise ship's ventilated areas and provided evidence that can inform guidelines about the optimisation of ventilation operations in large passenger ships, contributing to respiratory health, infection control and energy efficiency aboard.

Conventional Palmes Diffusion Tubes (PDTs) are extensively employed by UK Local Authorities for measuring NO2 in air quality monitoring studies. These devices are known to suffer from biases due to from the effects of wind speed. Modified PDTs with wind protective filters have also been developed for deployment in the UK Urban NO2 Network (UUNN) with an improved measurement accuracy and repeatability. We report the performance of the two designs and also when enclosed in additional shelters. The comparison was carried out against simultaneous reference measurements and was evaluated through a statistical and modelled uncertainty. The model incorporated the individual components of the measurement uncertainty to provide an estimate of the total measurement uncertainty and identified which elements could be reduced across mean values of multiple measurements. We found that conventional PDTs could be adversely affected by wind speed and that the incorporation of shelters delivered improved repeatability and better accuracy across multiple diffusion tube measurements. The UUNN style diffusive samplers were more accurate than the PDTs and had better repeatability. The additional use of shelters with UUNN style samplers made no discernible difference to the measurements.

Isoprene is a key trace component of the atmosphere emitted by vegetation and other organisms. It is highly reactive and can impact atmospheric composition and climate by affecting the greenhouse gases ozone and methane and secondary organic aerosol formation. Marine fluxes are poorly constrained due to the paucity of long-term measurements; this in turn limits our understanding of isoprene cycling in the ocean. Here we present the analysis of isoprene concentrations in the atmosphere measured across the Southern Ocean over 4 months in the summertime. Some of the highest concentrations ( >500 ppt) originated from the marginal ice zone in the Ross and Amundsen seas, indicating the marginal ice zone is a significant source of isoprene at high latitudes. Using the United Kingdom Earth System Model we show that current estimates of sea-to-air isoprene fluxes underestimate observed isoprene by a factor >20. A daytime source of isoprene is required to reconcile models with observations. The model presented here suggests such an increase in isoprene emissions would lead to >8% decrease in the hydroxyl radical in regions of the Southern Ocean, with implications for our understanding of atmospheric oxidation and composition in remote environments, often used as proxies for the pre-industrial atmosphere.

A growing number of low-cost sensors (LCS) have been used to monitor air pollution in outdoor air. The benefit of utilizing LCS lies in its ability to offer increased spatial coverage, which provides real-time measurements at a reduced cost. The selection and combination of low-cost sensors represent the primary challenge in conducting observations using such sensors. This paper employs a sensor quality ranking strategy, utilizing random forest (RF) for aggregating the selected LCS combination, followed by evaluating the correction results using various model evaluation metrics. The LCS used in this study, regardless of their quality grades, achieves a coefficient of determination of 0.93 or higher after model calibration, indicating the effectiveness of employing RF for aggregation. It is found that using a pair of top and averaged LCS can significantly enhance the measurement quality by 25% in RMSE. Using RF to calibrate a single LCS increases the measurement performance at least two times in terms of MSE, RMSE, and MAE. Using paired LCS with RF aggregation for measuring PM2.5, the aggregated observation significantly approximates the reference measurement with $R^{2}=0.986$ .

Low-cost air quality sensors are a promising supplement to current reference methods for air quality monitoring but can suffer from issues that affect their measurement quality. Interferences from environmental conditions such as temperature, humidity, cross-sensitivities with other gases and a low signal-to-noise ratio make them difficult to use in air quality monitoring without significant time investment in calibrating and correcting their output. Many studies have approached these problems utilising a variety of techniques to correct for these biases. Some use physical methods, removing the variability in environmental conditions, whereas most adopt software corrections. However, these approaches are often not standardised, varying in study duration, measurement frequency, averaging period, average concentration of the target pollutant and the biases that are corrected. Some go further and include features with no direct connection to the measurement such as the level of traffic nearby, converting the initial measurement into a modelled value. Though overall trends in performance can be derived when aggregating the results from multiple studies, they do not always match observations from individual studies, a phenomenon observed across many different academic fields and known as “Simpson’s Paradox”. The preference of performance metrics which utilise the square of the error, such as root mean squared error (RMSE) and r2, over ones which use the absolute error, such as mean absolute error (MAE), makes comparing results between models and studies difficult. Ultimately, comparisons between studies are either difficult or unwise depending on the metrics used, and this literature review recommends that efforts are made to standardise the reporting of calibration and correction studies. By utilising metrics which do not use the square of the error (e.g., MAE), models can be more easily compared within and between studies. By not only reporting the raw error but also the error normalised by multiple factors (including the reference mean and reference absolute deviation), the variabilities induced by environmental factors such as proximity to pollution sources can be minimised.

Formic acid is an intermediate of the steam methane reforming process for hydrogen production. According to International Standard ISO 14687, the amount fraction level of formic acid present in the hydrogen supplied to fuel cell electric vehicles must not exceed 200 nmol·mol−1. The development of formic acid standards in hydrogen is crucial to validate the analytical results and ensure measurement reliability for the fuel cell electric vehicles industry. NPL demonstrated that these standards can be gravimetrically prepared and validated at 4 to 100 µmol·mol−1, with a shelf-life of 1 year (stability uncertainty < 7%; k = 2). Stability was not affected over 1 year or by low temperature or pressure. At sub-µmol·mol−1 level, formic acid amount fraction was found to decrease due to adsorption on the gas cylinder surface; however, it is possible to certify the formic acid amount fraction after a period of 20 days and ensure the certified value validity for 1 year with an uncertainty below 7% (k = 1) confirmed by thermodynamic investigation. This study demonstrated that formic acid in hydrogen gas reference materials can be prepared with reasonable uncertainty (>7%, k = 1) and shelf life (>1 year). Potential applications include the calibration of analysers and for studying the impact of formic acid on future application with relevant traceability and accuracy.

Isoprene is a hydrocarbon emitted in large quantities by terrestrial vegetation. It is a precursor to several air quality and climate pollutants including ozone. Emission rates vary with plant species and environmental conditions. This variability can be modeled using the Model of Emissions of Gases and Aerosols from Nature (MEGAN). MEGAN parameterizes isoprene emission rates as a vegetation-specific standard rate which is modulated by scaling factors that depend on meteorological and environmental driving variables. Recent experiments have identified large uncertainties in the MEGAN temperature response parameterization, while the emission rates under standard conditions are poorly constrained in some regions due to a lack of representative measurements and uncertainties in landcover. In this study, we use Bayesian model-data fusion to optimize the MEGAN temperature response and standard emission rates using satellite- and ground-based observational constraints. Optimization of the standard emission rate with satellite constraints reduced model biases but was highly sensitive to model input errors and drought stress and was found to be inconsistent with ground-based constraints at an Amazonian field site, reflecting large uncertainties in the satellite-based emissions. Optimization of the temperature response with ground-based constraints increased the temperature sensitivity of the model by a factor of five at an Amazonian field site but had no impact at a UK field site, demonstrating significant ecosystem-dependent variability of the isoprene emission temperature sensitivity. Ground-based measurements of isoprene across a wide range of ecosystems will be key for obtaining an accurate representation of isoprene emission temperature sensitivity in global biogeochemical models.

Atmospheric aerosols are important drivers of Arctic climate change through aerosol-cloud-climate interactions. However, large uncertainties remain on the sources and processes controlling particle numbers in both fine and coarse modes. Here, we applied a receptor model and an explainable machine learning technique to understand the sources and drivers of particle numbers from 10 nm to 20 μm in Svalbard. Nucleation, biogenic, secondary, anthropogenic, mineral dust, sea salt and blowing snow aerosols and their major environmental drivers were identified. Our results show that the monthly variations in particles are highly size/source dependent and regulated by meteorology. Secondary and nucleation aerosols are the largest contributors to potential cloud condensation nuclei (CCN, particle number with a diameter larger than 40 nm as a proxy) in the Arctic. Nonlinear responses to temperature were found for biogenic, local dust particles and potential CCN, highlighting the importance of melting sea ice and snow. These results indicate that the aerosol factors will respond to rapid Arctic warming differently and in a nonlinear fashion.

Novel traceable analytical methods and reference gas standards were developed for the detection of trace-level ammonia in biogas and biomethane. This work focused on an ammonia amount fraction at an upper limit level of 10 mg m-3 (corresponding to approximately 14 μmol mol-1) specified in EN 16723-1:2016. The application of spectroscopic analytical methods, such as Fourier transform infrared spectroscopy, cavity ring-down spectroscopy, and optical feedback cavity-enhanced absorption spectroscopy, was investigated. These techniques all exhibited the necessary ammonia sensitivity at the required 14 μmol mol-1 amount fraction. A 29-month stability study of reference gas mixtures of 10 μmol mol-1 ammonia in methane and synthetic biogas is also reported.

This report describes work to evaluate the performance of different commercial and proprietary cylinder treatments in improving the stability of ammonia reference materials in high pressure cylinders. Gas mixtures of 100 µmol/mol and 10 µmol/mol ammonia in nitrogen were prepared gravimetrically at both NPL and VSL. Comparative measurements at each amount-of-substance fraction were used to assess which passivation technique minimised the loss of ammonia upon preparation. The results indicate little difference between the commercial treatments, except at lower amount-of-substance fractions (10 μmol/mol). The variation observed in performance might be explained by the different abilities of the various treatments to prevent the adsorption of ammonia molecules on the internal surfaces of the cylinder, although the role of residual water on the cylinder surface in reacting with ammonia is unclear.

Abstract. The Southern Ocean is a critical component of Earth’s climate system, but its remoteness makes it challenging to develop a holistic understanding of its processes from the small to the large scale. As a result, our knowledge of this vast region remains largely incomplete. The Antarctic Circumnavigation Expedition (ACE, austral summer 2016/2017) surveyed a large number of variables describing the dynamic state of the ocean and the atmosphere, the freshwater cycle, atmospheric chemistry, ocean biogeochemistry and microbiology. This circumpolar cruise included visits to twelve remote islands, the marginal ice zone, and the Antarctic coast. Here, we use 111 of the observed variables to study the latitudinal gradients, seasonality, shorter term variations, the geographic setting of environmental processes, and interactions between them over the duration of 90 days. To reduce the dimensionality and complexity of the dataset and make the relations between variables interpretable, we applied a sparse Principal Component Analysis (sPCA), which describes environmental processes through 14 latent variables. To derive a robust statistical perspective on these processes and to estimate the uncertainty in the sPCA decomposition, we have developed a bootstrap approach. We identified temporal patterns from diurnal to seasonal cycles, as well as geographical gradients and “hotspots” of interaction. Our results establish connections of oceanic, atmospheric, biological and terrestrial processes in an innovative way, while confirming many well known relations of the Southern Ocean system. More specifically, we identify: the important role of the oceanic circulations, frontal zones, and islands in shaping the nutrient availability that controls biological community composition and productivity; that sea ice predominantly controls sea water salinity, dampens the wave field, and is associated with increased phytoplankton growth and net community productivity possibly due to iron fertilization and reduced light limitation; and clear regional patterns of aerosol characteristics emerged, stressing the role of the sea state, atmospheric chemical processing, as well as source processes near “hotspots” for the availability of cloud condensation nuclei and hence cloud formation. A set of key variables and their combinations, such as the difference between the air and sea surface temperature, atmospheric pressure, sea surface height, geostrophic currents, upper ocean layer light intensity, surface wind speed and relative humidity, played an important role in the majority of latent variables, highlighting their importance for a large variety of processes and the necessity for Earth System Models to represent them adequately. In conclusion, our study highlights the use of sPCA to identify key ocean-atmosphere interactions across physical, chemical, and biological processes and their associated spatio-temporal scales. The sPCA processing code is available as open-access and we believe that our approach is widely applicable to other environmental field studies.

The Southern Ocean is a critical component of Earth's climate system, but its remoteness makes it challenging to develop a holistic understanding of its processes from the small scale to the large scale. As a result, our knowledge of this vast region remains largely incomplete. The Antarctic Circumnavigation Expedition (ACE, austral summer 2016/2017) surveyed a large number of variables describing the state of the ocean and the atmosphere, the freshwater cycle, atmospheric chemistry, and ocean biogeochemistry and microbiology. This circumpolar cruise included visits to 12 remote islands, the marginal ice zone, and the Antarctic coast. Here, we use 111 of the observed variables to study the latitudinal gradients, seasonality, shorter-term variations, geographic setting of environmental processes, and interactions between them over the duration of 90 d. To reduce the dimensionality and complexity of the dataset and make the relations between variables interpretable we applied an unsupervised machine learning method, the sparse principal component analysis (sPCA), which describes environmental processes through 14 latent variables. To derive a robust statistical perspective on these processes and to estimate the uncertainty in the sPCA decomposition, we have developed a bootstrap approach. Our results provide a proof of concept that sPCA with uncertainty analysis is able to identify temporal patterns from diurnal to seasonal cycles, as well as geographical gradients and “hotspots” of interaction between environmental compartments. While confirming many well known processes, our analysis provides novel insights into the Southern Ocean water cycle (freshwater fluxes), trace gases (interplay between seasonality, sources, and sinks), and microbial communities (nutrient limitation and island mass effects at the largest scale ever reported). More specifically, we identify the important role of the oceanic circulations, frontal zones, and islands in shaping the nutrient availability that controls biological community composition and productivity; the fact that sea ice controls sea water salinity, dampens the wave field, and is associated with increased phytoplankton growth and net community productivity possibly due to iron fertilisation and reduced light limitation; and the clear regional patterns of aerosol characteristics that have emerged, stressing the role of the sea state, atmospheric chemical processing, and source processes near hotspots for the availability of cloud condensation nuclei and hence cloud formation. A set of key variables and their combinations, such as the difference between the air and sea surface temperature, atmospheric pressure, sea surface height, geostrophic currents, upper-ocean layer light intensity, surface wind speed and relative humidity played an important role in our analysis, highlighting the necessity for Earth system models to represent them adequately. In conclusion, our study highlights the use of sPCA to identify key ocean–atmosphere interactions across physical, chemical, and biological processes and their associated spatio-temporal scales. It thereby fills an important gap between simple correlation analyses and complex Earth system models. The sPCA processing code is available as open-access from the following link: https://renkulab.io/gitlab/ACE-ASAID/spca-decomposition (last access: 29 March 2021). As we show here, it can be used for an exploration of environmental data that is less prone to cognitive biases (and confirmation biases in particular) compared to traditional regression analysis that might be affected by the underlying research question.

Major cities such as London are increasingly becoming targets for reducing greenhouse gas emissions by policy makers. This is due in part to their higher rate of emissions compared to more rural areas, but also due to the political powers of city level government. To ensure that emission reduction policies are successful, policy makers require accurate knowledge of how emissions change over time. The London Greenhouse Gas Project aims to provide top-down emission estimates for London, adding a London measurement network to expand upon the UK’s existing national top-down measurement infrastructure. The national network has proved useful in contributing to the UK’s national emission reports, and the new local network will provide useful data targeted to London’s policy makers. A series of in-situ atmospheric concentration instruments are being installed across the city and will be used to estimate London’s emissions of methane initially, with carbon dioxide emissions to follow. A medium-density urban network provides challenges in instrument calibration and siting, as well as the development of new modelling approaches to capture the urban environment and link the measurements to policy-relevant emissions estimates. There are also opportunities to link with remote observations of London, including satellite and ground-based FTIR instruments. We present considerations of setting up the new network, and results from the initial instrument installation and model development.

A greenhouse gas monitoring network is being developed across London that will allow independent evaluation of reported emissions based on atmospheric data. The first site is operational at the Thames Barrier, and in this work, two atmospheric dispersion models (NAME and ADMS-URBAN) are compared to observed methane concentrations between 5 May 2018 and 31 July 2018. We find that the models simulate some of the major features in the data, with consistent data–model discrepancies suggesting errors in the emissions inventory. Under the United Nations Framework Convention on Climate Change (UNFCCC), developed countries are required to report their national emissions using greenhouse gas (GHG) inventories, which combine data on GHG-producing activities (e.g. energy production or waste management) with emissions factors for each activity. While it is considered best practice for these ‘bottom–up’ methods to be evaluated using atmospheric data-based ‘top–down’ techniques, currently, only the United Kingdom (UK), Switzerland and Australia have included such methods in their National Inventory Reports (Brown et al., 2019). Cities are becoming a more important modelling scenario as various sub-national groups, including cities, universities and large companies, declare climate emergencies and produce policies designed to reduce GHG emissions in line with a +1.5 degC target (Masson-Delmotte et al., 2018). The mayor of London has developed and begun enacting plans to make London carbon neutral by 2050 (Greater London Authority, 2018). These policies, the size of the city and the presence of a national-scale network (Stanley et al., 2018) make London a prime case study for the development of top–down, urban inverse modelling techniques. Inverse models calculate the unknown cause of an observed effect rather than the more straightforward problem of predicting the effect of a known cause. For estimating GHG emissions, this means using observations of atmospheric composition, with atmospheric transport and statistical models, to calculate the causal GHG emissions. Atmospheric transport models can introduce substantial uncertainty into the process. For this reason, the models need to be evaluated in the context of new scenarios such as cities. The two most abundant long-lived GHGs are carbon dioxide and methane, and both gases have been the subject of top–down investigations in other cities (using different models than those used in this study), such as Boston, USA (McKain et al., 2015; Sargent et al., 2018); Los Angeles, USA (Verhulst et al., 2017); and Paris, France (Staufer et al., 2016). Because emissions inventories only quantify anthropogenic sources, any influence from the biosphere must be accounted for. However, it is an ongoing research challenge to isolate anthropogenic carbon dioxide emissions in top–down studies as the biosphere (the photosynthesis and respiration of plants) can dominate the observations (White et al., 2019). For this reason, our study focuses on modelling methane, whose emissions are overwhelmingly anthropogenic in London and the southeast of the UK. Furthermore, methane inventories are thought to be more uncertain than those of carbon dioxide (±16.7% for methane vs ±2.9% for carbon dioxide in the UK inventory (Brown et al., 2019)) and could benefit more from top–down evaluation. London's methane emissions are estimated in the 2016 National Atmospheric Emissions Inventory (NAEI) to be 62kT per year, which is 3.1% of the UK total, despite London only occupying 0.64% of the UK’s land surface (using Office for National Statistics data). Cities across the UK are similar hotspots of methane emissions due to the density of gas transmission infrastructure and waste management facilities. Verified emissions reductions in London may provide evidence for similar policies to be used in other cities. The London GHG network ‘London GHG’ will comprise around 10 high-frequency instruments distributed across the city. To minimise difficulties in modelling the urban roughness layer, instruments are primarily being set up on buildings that lie high above the local urban canopy, such as lone tower blocks and tall church spires. In this paper, we will present and discuss early results from the measurements and modelling of a test site in central London. These results will inform future stages of the London GHG project. We have established an initial measurement site at the Thames Barrier in central London (51.497°N, 0.037°E). This site measures carbon dioxide and methane using a Picarro G2401 cavity ringdown spectrometer, which performs a measurement every 5s with a precision of approximately 50 parts per billion (ppb) for carbon dioxide and 1ppb for methane. These measurements are known as mole fractions, which is the atmospheric concentration of the gas measured as the fraction of particles in the air of the gas being observed. This instrument is similar to those installed in the national-scale UK Deriving Emissions linked to Climate Change (UK DECC) network (Stanley et al., 2018). In this article, we will examine the initial period of data collected from 5 May 2018 to 31 July 2018. We combine two bottom–up inventories to use as our emissions in this work, with NAEI used over the UK and Emissions Database for Global Atmospheric Research (EDGAR) used in surrounding countries. The NAEI is a gridded inventory produced by the UK government and provides a resolution of 1km × 1km, which can identify emissions within London, while EDGAR is produced by the European Commission Joint Research Centre at 0.1° × 0.1° (approximately 10km × 10km in the UK). The latest versions of both inventories available at the time of writing are used, which are 2016 for the NAEI and 2012 for EDGAR. Both inventories provide annual mean estimates but unfortunately do not include any seasonal or diurnal time variations nor spatial uncertainty estimates. Within London, NAEI methane emissions are predominantly due to waste water treatment and leakages in the domestic gas distribution system. Emissions from the gas network are roughly distributed by population in the inventory, while waste emissions are centred on multiple emission hotspots across the city, as shown in Figure 1. These hotspots may provide a challenge for atmospheric modelling as they are of a size similar to, or smaller than, the model resolution. Two models are required to infer GHG emissions from atmospheric concentrations: a physical model and a statistical model. The physical model is usually an atmospheric or chemical transport model that estimates the atmospheric concentration at a given location and time using emission (flux) data and meteorological input. The statistical model compares the modelled and observed concentrations and calculates the emissions field that enables the model to best replicate observations, subject to various constraints (Ganesan et al., 2014). In this work, we focus on analysing and comparing the performance of the two physical models in an urban environment. We use two models that work quite differently in order to identify the best path forward for future modelling in the London GHG project. The first of the two physical models used in this work is the Met Office Lagrangian particle dispersion model, the Numerical Atmospheric-dispersion Modelling Environment (NAME) (Jones et al., 2007). Atmospheric transport is simulated in NAME as the advection and diffusion of thousands of particles, which are tracked backwards in time from the measurement location, recording where they pass near (within 40m of) the surface – the assumed source of emissions (Manning et al., 2011) (Figure 2). The model provides estimates of observation sensitivities known as ‘footprints’, which are 2D fields that map how much the different regions in the emissions field contribute to the observed atmospheric concentration of the gas for each measurement. The model also estimates where and when particles leave the domain so that boundary conditions can be accounted for. Mole fractions at the measurement site can be estimated as the product of each footprint and the emissions field, plus any contribution from the mole fraction at the boundary of the domain. The domain and boundaries used in this work are shown in Figure 3. The boundary conditions are taken from the Copernicus Atmosphere Monitoring Service global methane products, which use satellite measurements and models to produce global four-dimensional methane fields (Inness et al., 2019), adjusted to better match background measurements at Mace Head, Ireland. The NAME model was run offline using Met Office Unified Model meteorology. We use the high-resolution (1.5km) UKV meteorological data, where available, and the approximately 12km-resolution global dataset elsewhere. While the UKV meteorology has a high enough resolution to resolve urban-scale phenomena such as the urban heat island, NAME itself does not explicitly account for urban turbulent transport. Footprints and emissions are combined in a multiple-resolution grid shown in Figure 1(c), with London and its surroundings in a high-resolution (0.032° × 0.021°, ~2.5km) grid embedded in a low-resolution (0.352° × 0.234°, ~25km) grid used for previous national modelling (Lunt et al., 2016). The second physical model used is ADMS-URBAN produced by Cambridge Environmental Research Consultants (Stocker et al., 2012; Hood et al., 2018). This model is designed specifically to model urban environments at a very high (street level) resolution, taking account of complex features such as the effect of buildings. ADMS-URBAN differs from NAME in several key ways: ADMS-URBAN can explicitly represent large numbers of individual sources, including point sources (with specified heights) and road sources, but is limited in domain, and the concentration downstream of each source is represented by an analytic distribution that, for point sources, is Gaussian in neutral and stable conditions and skewed Gaussian in unstable conditions but has other more complex forms for road sources. The concentration distribution is stationary in time for each successive hour and may use single-site or gridded meteorology to calculate the footprint. Here, we drive ADMS-URBAN with meteorological measurements from Heathrow Airport. These measurements are internally modified according to the difference in roughness lengths from the urban landscape at Heathrow and the Thames Barrier, resulting in a lower windspeed. This is the same setup that has been successfully used for modelling air quality in London (Hood et al., 2018). The boundary layer height is calculated internally as opposed to NAME, which uses the value diagnosed in the Unified Model. In this study, the domain for ADMS-URBAN is the same as that used by the London Atmospheric Emissions Inventory, which encompasses all London boroughs and everything within the M25. As ADMS-URBAN does not estimate the influence of fluxes outside London or regional boundary conditions, the ADMS-URBAN footprint requires additional information so that the total methane concentration can be simulated. In this study, we embedded ADMS-URBAN footprints within the larger-scale NAME footprints. The ADMS-URBAN footprints are coarsened to match the NAME high-resolution grid (~2.5km) and thus loses some spatial information as the grid cartographic projections are otherwise incompatible. The geographic extent of London used throughout the paper is taken from the OpenStreetMap London administration polygon, rasterised onto the NAME high-resolution grid. Examples of NAME and ADMS-Urban footprints are shown for two different meteorological conditions in Figure 4. The top row shows footprints under steady westerly winds at 1500 utc, 10 May 2018, whereas the bottom row shows footprints at 1500 utc, 24 May 2018, under more complex conditions, with fronts passing over London (Figure 5). Under the steady westerly winds, both footprints are qualitatively similar, with observations at the Thames Barrier being influenced by fluxes from western and central London, although the ADMS-URBAN footprint is four times more sensitive to emissions when both models are integrated over London. Under the more complex meteorological scenario, the NAME footprints indicate sensitivity to a wider area of London with nearly twice the total London sensitivity as ADMS-URBAN, presumably reflecting the range of wind directions experienced by the model particles, whereas ADMS-URBAN shows sensitivity to a narrower region upwind of the measurement site. On average, ADMS-URBAN is about twice as sensitive to London fluxes as NAME, with a mean (5th–95th percentile) total London sensitivity of 0.97 (0.24–2.96) (molm−2s−1)−1 compared to 0.43 (0.09–1.39) (molm2s−1)−1 for NAME. One possibility for the difference in sensitivity is the internal boundary layer height used by each model. Figure 6 shows a histogram of the boundary layer heights, demonstrating ADMS-URBAN's overall shallower boundary layers. The boundary layer is important in determining surface sensitivity as it limits the vertical mixing of air. In the models, this increases surface sensitivity, reflecting how low boundary layers trap GHGs and increase their atmospheric concentration near the surface. Figure 7(a) shows the hourly median and 33rd–66th and 5th–95th percentile ranges of methane observations at the Thames Barrier between 5 May 2018 and 31 July 2018 inclusive. Observed mole fractions are generally higher and more variable at night than during the day, and the lowest values observed are typically observed during the daytime. This difference is thought to be largely due to diurnal changes in atmospheric stability, with stable nocturnal boundary layers trapping locally emitted methane in contrast to strong mixing of nearby sources during the day (Stull, 1988). Figure 7(b) shows the mean observed mole fractions as a function of wind direction and wind speed (from the Met Office UM analysis meteorology as measurements were not made at the Thames Barrier), which highlights that the highest observed concentrations occur at low windspeeds and/or from an easterly direction, with a spot of high emissions from the northeast. There are several possibilities why easterly winds are associated with higher methane concentrations. The first reason is that these winds are likely to be carrying emissions from mainland Europe, with the Benelux region being particularly high in emissions according to the EDGAR inventory. In contrast, when winds come from the west, they arrive in the UK or Ireland with mole fractions consistent with the hemispheric background. A contribution from local sources is also possible, with several large methane emission hotspots within several kilometres of the Thames Barrier, according to the NAEI. For example, emissions from the Beckton Sewage Treatment Works approximately 4 km away may be consistent with the maximum rise in the mole fraction at around 50o. Mole fractions associated with this wind direction tend to be highly variable, suggesting a nearby plume impinging on the measurement site, rather than a more well-mixed regional source. Data from the addition sites planned around London could help distinguish between these two cases by providing different viewpoints on local emissions. By combining the footprints for NAME or ADMS-URBAN (embedded within NAME) with the NAEI and EDGAR emissions fields, we can produce a modelled time series that can be compared to the Thames Barrier data. An example for a typical 2-week period is shown in Figure 8. The modelled mole fractions are attributed to three different factors: fluxes from within London, fluxes outside London and contribution from the boundary conditions at the edge of our NAME domain. The two models only differ in their modelled London contribution as the ADMS-URBAN footprints are embedded into the NAME-derived regional footprints and boundary conditions. The full period mean and 5th–95th percentiles of the mole fraction due to sources within London for NAME and ADMS are 34.2 (4.37–121) ppb and 55.9 (9.30–173) ppb, respectively, compared to 45.2 (9.67–113) ppb from regional sources and 1921 (1910–1937) ppb from the boundary conditions. The modelled concentrations generally capture the observed diurnal cycle, although the magnitude of the night-time peaks can differ from the observed data by around a factor of two or more. NAME mostly underpredicts methane concentration, while the ADMS-URBAN model underpredicts on some nights and overpredicts on others. Figure 7(c–f) shows the hourly medians and wind dependence for the observed, NAME and ADMS-URBAN modelled mole fractions. From the hourly medians, the night-time underestimation seen in Figure 8 is more evident. Both models show an increase in mean mole fractions at low wind speeds, but at a much lower magnitude than in the observations. This finding could be because nearby sources (within a few km) are larger than estimated in the inventory, or it could show that the models tend to overestimate mixing during low-wind conditions, with both possibilities suggesting the high observations are not primarily due to the Benelux region. The hotspot to the northeast is also not captured in the models, which may indicate that a source in this direction is not present or underestimated in the inventory, or it could show that model transport is generally too dispersive for this wind sector. Figure 9 shows the modelled mole fractions plotted against the observations for the two dispersion models, for total concentrations and the London contribution only. For this analysis, the data were filtered to retain only points where the observational variability within each hour period was less than one half of the modelled London contribution. This removes points heavily influenced by local emissions that the models are not expected to capture accurately. Summary statistics are shown in Table 1. Overall, the models show broadly similar correlations with the data, despite their very different architectures. The NAME model has a slope of regression greater than 1, suggesting that the emissions or modelled sensitivities are underestimated. The opposite is true for the ADMS model, although the line of regression is skewed by a small number of points where the model greatly overpredicts methane concentrations. For both models, the R2 value decreases when looking at just the London contribution, perhaps because they struggle to accurately represent complex urban meteorology or because of errors in the distribution of nearby emissions sources in the NAEI. During the most well-mixed conditions (between 1100 and 1700, when hourly observation variability is below 5 ppb), the models are in closer agreement but show lower sensitivity to London emissions than at other times. Overall, model output from NAME correlates more strongly with the observations than ADMS-URBAN, perhaps due to the use of three-dimensional meteorology compared to single-site meteorology. However, ADMS-URBAN better captures the diurnal cycle present in the observations, possibly due to the different boundary layer height calculations used, although there could be many factors that contribute to both differences between the models. These simulations show that NAME and ADMS combined with the NAEI can capture some of the major features in a methane mole fraction time series at an urban site. The two models show similar features in their simulated mole fractions, despite a different modelling approach and driving meteorology, which suggests that a substantial portion of the model–measurement mismatch is due to the differences between the truth and inventory emissions magnitude, distribution and/or temporal variability. The next step in the development of a modelling system to support the London GHG network is to develop a new statistical model, an inverse modelling system that can determine whether changes in emissions and their distribution can improve the fit between the model and the data (Lunt et al., 2016). The differences between the models will lead to differences in inferred emissions from an inverse modelling system. These differences will capture some of the sensitivity of the inverse models to atmospheric transport error and can help better inform interpretation of inferred emissions as a result. As the first step in the development of a network for monitoring of London's carbon dioxide and methane emissions, we have established a continuous measurement site on the Thames Barrier. We analysed methane data from this site during the summer of 2018 and compared the observations to two distinct atmospheric transport models, NAME and ADMS-URBAN. Results showed that, over a 3-month period, the models could capture some of the broader features in the data, such as the diurnal cycle and wind direction dependence. The consistency of the difference between the model prediction of some of these features and the data suggests that a substantial proportion of the model–observation discrepancy is due to errors in the emission inventories. We will use both models in a future emissions estimation framework to provide some estimate of the sensitivity of the derived emissions to atmospheric transport model errors. Further work towards a London GHG monitoring network will involve the set-up of additional measurement sites across the city and the development of an urban-scale inverse modelling system that will use the transport models from this work to obtain top–down emissions estimates for London. Provided that the network can be supported over the coming years, the results from these estimates will be supplied to policymakers to help determine whether London's emissions reduction targets have been successful. The London GHG system also has the potential to identify missing sources or spatial discrepancies in the NAEI and may be able to give some insight into the temporal variability in emissions not accounted for in the bottom–up inventories. The authors are grateful to the Environment Agency for providing access to the Thames Barrier measurement site, particularly to Babatunde Adelakun for his help and support. The NAME model and United Model meteorological data are provided by the Met Office. NAME model runs were performed on the University of Bristol Advanced Computing Research Centre's BlueCrystal and JASMIN, the UK collaborative data analysis facility. This work is supported by the Natural Environment Research Council as part of the London GHG and MOYA projects. The Mace Head observations are funded by the Department for Business, Energy and Industrial strategy. Daniel Hoare is supported by a studentship from the NERC GW4+ Doctoral Training Partnership (grant no. NE/R000921/1).

The measurement of trace amounts of water in process gases is of paramount importance to a number of manufacturing processes. Water is considered to be one of the most difficult impurities to remove from gas supply systems and there is strong evidence that the presence of water contamination in semiconductor gases has a measurable impact on the quality and performance of devices. Consequently, semiconductor manufacturers are constantly reducing target levels of water in purge and process gases. As the purity of gases improves, the problem of quantifying contamination and ensuring that the gases are within specification at the point of use becomes more challenging. There are several established techniques for detecting trace water vapour in process gases. These include instruments based on the chilled mirror principle which measures the dew-point of the gas and the quartz crystal adsorption principle which measures the adsorption of water vapour into a crystal with a hygroscopic coating. Most recently, spectroscopic instruments such as those employing cavity ring-down spectroscopy (CRDS) have become available. The calibration of such instruments is a difficult exercise because of the very limited availability of accurate water vapour standards. This CCQM pilot study aims to assess the analytical capabilities of laboratories for measuring the composition of 10 μmol mol−1 water vapour in nitrogen. Main text To reach the main text of this paper, click on Final Report. Note that this text is that which appears in Appendix B of the BIPM key comparison database kcdb.bipm.org/. The final report has been peer-reviewed and approved for publication by the CCQM, according to the provisions of the CIPM Mutual Recognition Arrangement (CIPM MRA).