The measurement principle of a Palmes tube (Fig. 1) is based on molecular diffusion. The tube contains two fine stainless-steel grids impregnated with triethanolamine (TEA), a reagent that chemically binds NO₂ upon contact, forming nitrite (NO2-). Diffusion takes place as there is a concentration difference between the open bottom of the tube (at ambient NO₂ concentration) and the closed top (at zero concentration). After exposure, the tube is sealed and brought to the laboratory, where the accumulated nitrite on the grids is quantified through chemical analysis. This mass is then used to calculate the average ambient NO₂ concentration at the sampling site over the exposure period. Due to the slow diffusion rate, PDTs are typically deployed for a four-week sampling duration. Shorter intervals, such as two weeks, are feasible but result in an increased measurement uncertainty.

Kumasi is the capital of the Ashanti Region in southern Ghana and serves as a major metropolitan and commercial hub. As of 2024, the city has an estimated population of approximately 3.5 million, making it the second-most populous city in Ghana. Geographically, Kumasi is situated at approximately 6.69° N latitude and 1.62° W longitude, at an elevation of about 250 to 300 m above sea level. The city experiences a tropical rainforest climate, characterized by high humidity and relatively stable temperatures throughout the year, ranging between 21 and 31 °C. Kumasi has a bimodal rainfall pattern, with major rains typically occurring from March to July, and a shorter rainy season from September to November. Annual rainfall averages between 1300 and 1500 mm. The dry season, associated with the Harmattan winds, usually spans December to February, though residual precipitation may occur.

We strategically selected 20 monitoring sites across Kumasi to capture the spatial variability of NO₂ pollution, such as high-traffic areas around marketplaces, health-sensitive zones such as hospitals, major industrial locations, and some urban background areas (Table 1). This categorization was intended to reflect the range of anthropogenic activities contributing to NO_x emissions within the city and to assess potential exposure risks to different population groups. The location and classification of the sites is shown in Fig. 2.

The objective was to establish the first independent, locally operated air quality (AQ) laboratory in Kumasi, capable of preparing and analysing diffusion tubes without dependence on foreign laboratories. The lab was designed with a focus on simplicity and economy – minimizing the use of costly equipment and consumables and avoiding complex analytical techniques.

The methodology for PDT preparation, deployment, and analysis followed the best practices outlined by Targa and Loader (2008), which served as the foundation for the lab's protocols. These guidelines were adapted as needed to suit local conditions (see Sect. 2.3.1 and 2.3.2).

The laboratory was developed collaboratively by three teaching assistants from the Chemical Engineering Department at KNUST and three master's students from the Faculty of Industrial Design Engineering at TU Delft, Netherlands. A comprehensive lab manual was created (published as Eckert et al., 2025) detailing all procedures step by step. This manual, along with instructional videos, now forms the core training resource for new local staff.

A Hach DR5000 spectrophotometer was donated by the Utrecht University, Netherlands. A 10 g bottle of N-(1-naphthyl) ethylenediamine dihydrochloride (NEDA) was imported to Ghana, which is sufficient for extended use. All other chemicals could be sourced locally. Tube holders were made from two pieces of wood (around 20 cm long) nailed together in an L-shape, see Fig. 4. The top part has 2 cm diameter holes to accommodate the tubes. The lower part, mounted to the poles, includes two holes for zip ties to secure it to a pole at the measurement site.

TROPOMI is a nadir-viewing imaging spectrometer aboard the Sentinel-5P satellite (Veefkind et al., 2012). Launched in October 2017, it is orbiting the Earth in a sun-synchronous orbit at an altitude of approximately 800 km. The overpass time is around 13:30 local solar time.

The chemical composition of the atmosphere is obtained from hyperspectral measurements of sunlight backscattered by Earth's atmosphere. From the position and depth of the absorption lines the column density of trace gases like NO₂ can be determined (e.g. van Geffen et al., 2020). The retrieval algorithm converts the so-called slant column density into a total vertical column density and separates the total column density in a stratospheric and a tropospheric part.

The tropospheric column density Vtrop represents the total number of NO₂ molecules contained within a vertical column with unit area in the troposphere. If n(l) represents the number density at height l, then the column density is given by 4Vtrop=∫tropnldl In practice, the satellite retrieval algorithm provides an estimate V^trop of this value. The retrieved column depends on how sensitive the measurement is to concentrations at different altitudes. This vertical sensitivity is described by the averaging kernel A(l), such that 5V^trop=∫tropA(l)nldl The averaging kernel (AK) quantifies how a perturbation in the true (but unknown) atmospheric profile at height l affects the retrieved column (Eskes and Boersma, 2003). Values larger than 1 imply high sensitivity to that layer, while values near 0 indicate low sensitivity.

In discrete form, for an atmosphere divided into a finite set of layers, Eq. (5) becomes 6V^trop=∑lAlVl where Vl represents the partial column density in layer l. The elements Al are calculated using a radiative transfer model that incorporates an a priori profile, scattering and absorption by clouds and surface, as well as observation-specific parameters such as the solar zenith angle and the instrument's viewing geometry.

We use the latest level-2 retrieval product for TROPOMI tropospheric NO₂ column densities and their associated AKs, version 2.4 (Eskes et al., 2022; van Geffen et al., 2022), which can be downloaded at the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/, last access: 13 April 2026). The averaging kernel calculation of the retrieval is based on a priori profiles from the global chemistry model TM5-MP (Williams et al., 2017), run at a 1°×1° spatial resolution.

Maps of average tropospheric column densities V‾ corresponding to the Palmes measurement period are created by stacking all footprints of retrievals marked with a quality value >0.75, i.e. valid retrievals with cloud radiance fraction below 50 %. Each overpass, a given location is sampled by a different satellite footprint, see Fig. 3. As larger footprints are less representative for the area, we take a weighted average of individual retrievals V^i covering that location 7V¯=∑iwiV^i∑iwi taking weighting factor wi=1/Ai, with Ai representing the area of footprint i.

For the Kumasi measurement campaign, we used two batches of 50 tubes, prepared and analysed by Buro Blauw Luchtkwaliteit – an ISO 17025-accredited air quality laboratory based in Wageningen, the Netherlands. The tubes (referred to as “remote-lab tubes”) were prepared within two weeks prior to deployment and analysed within four weeks following exposure. In addition, 100 unprepared tubes of the same type and material were brought from Buro Blauw to Ghana. These were prepared and analysed at the newly established laboratory facility at KNUST (referred to as “local-lab tubes”). By conducting side-by-side measurements, the performance and reliability of the local laboratory could be evaluated (see Sect. 3.2). After use, the local-lab tubes were cleaned and recycled in subsequent measurement cycles.

To explore scalable solutions for Palmes networks in Ghana and other countries on the continent, we also tested diffusion tubes made from materials available at local markets. Development and validation of these locally made tubes is ongoing and will be presented in a separate study.

The tube holders were mounted on existing poles using tie wraps, positioned at a height of approximately 3.5 m (see Fig. 4). The remote-lab diffusion tubes were deployed across all 20 selected monitoring sites. To assess the precision of the method, duplicate (duplo) measurements were conducted at each location.

In addition, blank tubes were installed at 8 of the sites. These tubes were sealed (i.e., not exposed to ambient air) by keeping their end plugs in place. This allowed for the evaluation of potential degradation of the TEA coating over the course of the measurement period.

The measurement campaign consisted of two sampling cycles, each lasting approximately two weeks (see Table 2). At the start of Cycle 1, remote-lab diffusion tubes were placed in tube holders mounted on signposts or lamp posts found at the 20 monitoring sites. At the end of Cycle 1, the tubes were retrieved and replaced with those designated for Cycle 2. For this cycle, additional local-lab tubes were deployed alongside the remote-lab tubes at 10 of the sites to enable side-by-side comparison. All tubes were successfully retrieved at the end of both cycles, with none missing. No insects or spiders were found inside. Following the recommendations of Targa and Loader (2008), the tubes were stored in a dark and, as much as possible, cool location both before and after deployment.

A total of 40 duplo measurements were conducted across the monitoring locations. As shown in Fig. 6, 15 duplos (37.5 %) exhibited a difference of less than 1 µg m⁻³, while 26 duplicates (65 %) showed a difference of less than 2 µg m⁻³. Using the variance of the differences in the duplo measurements, we estimate the precision of the method at 3.0 µg m⁻³.

It should be noted that the exposure period was relatively short – only two weeks. For the more commonly used four-week exposure period, a higher level of precision is expected.

Eight on-site blank measurements were performed, 4 at each sampling cycle. One tube, 070-035, was kept as a lab blank, shielded from sunlight under room temperature conditions. The results are listed in Table 3.

The average of the on-site blank samples is 2.47 µg m⁻³ in Cycle 1 and 2.48 µg m⁻³ in Cycle 2. The nitrite mass of the lab blank was found to be below the detection limit of 0.02 µg. The nitrite formation found in the other tubes is therefore assumed to originate from degradation of the TEA once exposed to elevated temperatures and solar radiation at the sampling site. Assuming that a similar degradation effect also occurs in the exposed (non-blank) tubes, we account for this bias by subtracting 2.47 µg m⁻³ in the results.

Consistent exposure across sites was ensured by mounting all diffusion tubes on free-standing poles using identical holders. Nevertheless, the simple design of the tube holders provides little protection against radiation and wind. In particular, wind-induced turbulence can shorten the effective diffusion path and increase the uptake rate. Heal et al. (2019) identify wind as the largest contributor to bias, potentially leading to overestimations of several tens of percent. In addition, local temperature differences – such as between shaded and sun-exposed locations – may introduce site-specific bias. A better-designed mounting system that shields the tubes from wind and radiation should help reduce this bias in future measurements.

Table 4 lists the NO₂ concentrations found at the network locations, where the duplo measurements have been averaged and bias-corrected by 2.47 µg m⁻³. Differences between the two measurement periods are minimal. This is consistent with the nearly identical meteorological conditions during both cycles, in terms of temperature, wind speed and wind direction. Notably, the more frequent light rain in Cycle 2 hardly influences the NO₂ concentrations. Averaging the results for each site provides a more robust dataset, considered to be representative of air quality conditions in Kumasi during March 2025.

A wide range of NO₂ concentrations was observed across Kumasi, as illustrated in Fig. 7. The spatial distribution clearly indicates that NO₂ levels are predominantly influenced by road traffic.

The lowest NO₂ concentration – 7 µg m⁻³ – was recorded at KU-020 (Engineering Guest House), located within the green and relatively traffic-free KNUST campus. In contrast, the highest concentrations were found at traffic-dense locations, including KU-018 (Anloga Junction, 69 µg m⁻³), KU-013 (Sofoline Interchange, 77 µg m⁻³), and KU-014 (Suame Circle, 88 µg m⁻³).

Measurements taken at industrial sites (KU-004 and KU-005) did not show outstanding concentrations, suggesting that industrial emissions are not a major contributor to ambient NO₂ levels in these areas.

During Cycle 1, the air quality laboratory facility at KNUST was established and became operational. For Cycle 2, 10 locally prepared tubes were deployed alongside remote-lab tubes at 10 sites within the measurement network. The local-lab tubes were analysed shortly after their collection at the end of Cycle 2. Since they are made from the same materials and have identical dimensions as the remote-lab tubes, they are assumed to experience similar degradation and were therefore corrected by subtracting 2.47 µg m⁻³.

The scatter plots in Fig. 8 show data pairs closely aligned with the 1:1 line, indicating strong agreement between the two sets of measurements. The relatively low scatter demonstrates that the KNUST laboratory can reliably reproduce results comparable to those of the accredited laboratory.

The measured values for each data pair are presented in Table 5. Note that the remote laboratory results are generally lower than those from the local laboratory, possibly due to sample degradation during the 27 d delay before analysis, compared with 2 d for the local lab. The uncertainty in the differences, however, is too large to claim a real bias.

To estimate the uncertainty in NO₂ concentrations determined by the local laboratory, we calculate the variance of the differences Di between paired measurements. This allows us to approximate the error associated with the PDT results from the local lab, σlocal, using the following relation: 8σlocal2+σremote2=1n-1∑i=1nDi-D‾2 where D‾ is the mean of the differences, here 2.0 µg m⁻³, and σremote represents the estimated uncertainty of duplicate measurements analysed at the accredited (remote) laboratory, estimated in Sect. 3.1 to be 3.0 µg m⁻³. Based on this approach, the resulting uncertainty in NO₂ concentration for the local laboratory is estimated at 5.1 µg m⁻³.

It should be noted that this estimate is not very robust due to the small sample size (n=10) and the relatively short exposure period of two weeks.

Comparison of Figs. 7 and 9 shows no clear spatial relation between the satellite column measurements and the surface concentrations. The elevated NO₂ levels found in downwind regions are not readily apparent in the surface measurements, although they might be detectable with a more extended ground-based monitoring network that includes more urban background stations.

The main reason for the poor spatial relation is the scale mismatch: TROPOMI's footprint (5.5×3.5 km² at best) averages the sharp local gradients found near strong emission sources such as traffic arteries. The relationship is further complicated by the unknown vertical distribution of NO₂, as tropospheric columns integrate contributions from air masses of different origins, ages, and mixing states. Additional atmospheric modelling would therefore be needed to resolve fine-scale structures in surface air pollution from satellite observations (see, e.g., Mijling et al., 2025).

Other limitations for direct comparison include the once-daily overpass of TROPOMI, which captures NO₂ concentrations around 13:30 local solar time, which might not be representative for daily averages. Furthermore, since the satellite retrievals rely on sufficiently cloud-free conditions, the lack of data under cloudy conditions can introduce a sampling bias in monthly averages (see e.g. Glissenaar et al., 2025).

Nevertheless, it is interesting to look at the ratio between average tropospheric column densities and average ground concentrations on an urban scale, as TROPOMI retrievals of tropospheric columns have been shown to correlate strongly with in-situ surface measurements, particularly in urban environments (Cersosimo et al., 2020; Goldberg et al., 2021; Ialongo et al., 2020; Jeong and Hong, 2021). Understanding this relationship provides a basis to monitor surface concentrations using satellite measurements alone. Table 6 compares the column-to-surface ratio for Kumasi with those of European cities of similar size for March 2025.

Surface NO₂ concentrations were extracted from the European Environment Agency (EEA) database (EEA, 2025) for March 2025. Urban averages are calculated from all regulatory monitoring stations located within 10 km of each city centre, having at least 80 % completeness in its hourly measurements. To put our Kumasi measurements further in perspective, we included the average surface concentration for Accra in March 2020, measured by Ogawa passive samplers (Arku, 2021) from the study by Wang et al. (2022).

Average tropospheric column concentrations were obtained from maps of stacked tropospheric NO₂ columns (as described in Eq. 7), calculated within a 10 km radius of each city centre. The background column concentration is defined as the minimum value within a 15 km radius.

Table 6 shows that the column-to-surface ratios in Kumasi and Accra are significantly lower than those in the European cities. As illustrated in Fig. 10, this is a notable finding: small columns observed by TROPOMI over West Africa do not necessarily correspond to low surface concentrations – contrary to what one might intuitively conclude from NO₂ maps when presuming a fixed column-to-surface ratio.

The low columns over Ghanaian cities can be partly explained by the reduced sensitivity of the retrieval method near the surface in combination with coarse a priori NO₂ profiles from TM5-MP. Averaging kernels over Accra and Kumasi (Fig. 11a) confirm a reduced sensitivity throughout the troposphere. This reduction is closely associated with the high aerosol loading in the region, as reflected by elevated aerosol optical depth (AOD) values.

Table 7 presents the mean TROPOMI AOD at 494 nm (KNMI et al., 2025) in March 2025 for a 30×30 km² area centred on each city. Values well above 1 suggest the presence of substantial haze, resulting primarily from long-range dust transport from the Sahara and Sahel regions, as well as smoke from biomass burning, which remains active during March. The resulting decrease in atmospheric transparency effectively obscures near-surface urban pollution from satellite detection, thereby contributing to the decrease in vertical sensitivity towards the surface.

Another reason for the relatively low tropospheric column densities over Kumasi can be found in the NO_x chemistry. During the day chemical loss of NO_x is dominated by the oxidation of NO₂ by the hydroxyl radical (OH): R1NO2+OH+M→HNO3+M The associated lifetime of NO₂ is 9τ=1/kOH, where k is the reaction rate and OH the hydroxyl concentration, expressed in molec. m⁻³. The primary daytime source of OH is the photolysis of ozone (O₃) followed by a reaction with water vapor: R2O3+hvλ<320nm→O1D+O2R3O1D+H2O→2OH Compared to mid-latitudes, OH production is enhanced in the tropical lower troposphere due to stronger solar radiation, more abundance of water vapor, and higher temperatures. This is confirmed by simulations of the global chemistry model TM5-MP for March 2008, conducted around the satellite overpass time at a 1°×1° spatial resolution (Fig. 11b). Average OH concentrations between 1000 and 850 hPa are shown in Table 7. The corresponding lifetime τ is estimated using Eq. (9), assuming a typical rate constant k=1.1×10-11 cm³ molec.⁻¹ s⁻¹ (Sander et al., 2006). In the lower troposphere, k is only weakly dependent on temperature within the 250–300 K range. It should also be noted that these lifetimes represent averages over a relatively large 1°×1° grid cell, whereas lifetimes within urban areas can be considerably shorter (Beirle et al., 2011; Lorente et al., 2019).

From the table, it can be seen that lifetimes found in Kumasi are 1.1–4.9 times shorter than found in European cities. This means that while NO₂ concentrations found at the surface close to emissions sources can be high, NO₂ column densities will be lower, as the tropospheric column contains air parcels of different origins, with more aged chemistry.

To better quantify the differences in column-to-surface NO₂ ratios between European and Ghanaian cities, information on the vertical distribution of NO₂ is required. Although this distribution is not directly known without dedicated atmospheric modelling, we can approximate it by introducing a simplified height-distribution model (Fig. 12) and fitting it to surface measurements and tropospheric column densities, explicitly incorporating their averaging kernels in the retrieval.

The model assumes that background NO₂ is well-mixed within the planetary boundary layer (PBL) of height hPBL at the time of the satellite overpass, consistent with NO₂ sonde observations (Sluis et al., 2010). Contributions from free-tropospheric NO₂ to the tropospheric column density are considered negligible.

The urban NO₂ component, originating from local emissions, is assumed not to be fully mixed vertically. Observations from the Eiffel Tower, for example, show that NO₂ concentrations at 300 m height in central Paris are about 45 % lower than surface levels around the satellite overpass time (Dieudonné et al., 2013, Fig. 1b). Accordingly, we represent the urban plume as being well-mixed within an effective layer of height hmix, located below the PBL.

First, we determine the background surface concentration cbg. We estimate the background column density V^bg from the lowest TROPOMI retrieval value found within a 15 km radius around the city centre (see Fig. 10), and take hPBL taken from the ERA5 reanalysis (Hersbach et al., 2023). For a given concentration cbg (in ppb), the partial background column in an atmospheric layer l is 10Vbg,l=cbg×10-9NAMairgΔPlPBL, where NA is Avogadro's number, Mair the molar mass of dry air, and g the gravitational acceleration. The term ΔPlPBL is the pressure thickness of the layer within the PBL: for the layer containing hPBL, only the portion below the PBL top is used, and for layers entirely above the PBL, ΔPlPBL=0. Using Eq. (6), we find the relation between cbg, the observed V^bg, and the AK elements Al, which allows us to solve for cbg.

In the second step, we determine the mixing layer height hmix. We estimate the urban column density V^ from the mean TROPOMI retrieval within a 10 km radius of the city centre, and take the surface concentration csurf from the in-situ measurements. For the block profile (blue curve in Fig. 12), the partial column in layer l is 11Vl=Vbg,l+csurf-cbg×10-9NAMairgΔPlmix, where ΔPlmix is the pressure thickness up to the mixing-layer height hmix, and zero above it. Using Eq. (6), we determine the value of hmix for which the profile matches the observed V^, given the averaging kernel Al.

Note that surface concentrations must correspond to the satellite overpass time. For the European cities, hourly measurements at 13:00 UTC are used; these are lower than the daily mean values by factors ranging from 0.63 (Warsaw, Berlin) to 0.76 (Milan, Paris). For Kumasi and Accra, we apply a factor of 0.69 to estimate values at overpass time from mean surface concentrations.

Table 8 presents the fit results and the recalculated column densities, derived by vertically integrating the profiles without applying the averaging kernel (Eq. 4). Comparing with the results from Table 6, the recalculated tropospheric column densities for Paris, Milan, Brussels, Berlin in March 2025 increase by 30 %–40 %. This is consistent with Douros et al. (2023), who attribute the underestimation of retrievals over emission hot spots to the coarse (1°×1°) TM5-MP profiles taken as a priori.

In contrast, the recalculated tropospheric NO₂ column over Kumasi increases by a factor of 2.5. This larger correction can be attributed to a combined effect of the coarse a priori profile and a possible underestimation of the urban emissions in TM5-MP, and the reduced retrieval sensitivity caused by the high aerosol loading.

Despite this increase, the recalculated NO₂ columns over Kumasi (76 µmol m⁻²) and Accra (54 µmol m⁻²) remain substantially smaller than those observed over the European cities (132 µmol m⁻² on average). The lower vertical mixing height hmix suggest that the shorter chemical lifetime – discussed in the previous section – is likely a contributing factor.

The resulting lower effective mixing height might seem counterintuitive, as one would expect stronger convection due to higher surface temperatures. Note that the very low value over Accra may partly result from an overestimation of the surface concentration (March 2020 may not represent March 2025) or from the reduced air mass residence time over the city due to stronger winds.