The winter 2019 air pollution (PM 2.5 ) measurement campaign in Christchurch, New Zealand

. MAPM (Mapping Air Pollution eMissions) is a project whose goal is to develop a method to infer (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) airborne (cid:58) particulate matter (PM) emissions maps from in situ PM concentration measurements. In support of MAPM, a winter ﬁeld cam-paign was conducted in New Zealand in 2019 (June to September) to obtain the measurements required to test and validate the MAPM methodology. Two different types of instruments measuring PM were deployed: ES-642 remote dust monitors (17 instruments) and Outdoor Dust Information Nodes (ODINs; 50 instruments). The measurement campaign was bracketed 5 by two intercomparisons where all instruments were co-located, with a permanently installed Tapered Element Oscillating Membrane (TEOM) instrument, to determine any instrument biases. Changes in biases between the pre-and post-campaign intercomparisons were used to determine instrument drift over the campaign period. Once


Introduction
Airborne particulate matter (PM) comprises particles that can be solid, liquid or a mixture of both.The solids comprising PM can include both organic and inorganic constituents, such as sea salt, dust, pollen, and soot.Particle sizes and composition vary with location, origin and in situ chemical processes (Adams et al., 2015).There are health concerns associated with PM emissions, as PM remains suspended in the air where, if it is inhaled, the risk of developing cardiovascular and lung-related diseases increases (Anderson et al., 2012;Pizzorno and Crinnion, 2017).The World Health Organization estimates that PM air pollution contributes to approximately 800,000 premature deaths each year, ranking it the 13 th leading cause of mortality globally (Anderson et al., 2012).Pope et al. (2009) show that by decreasing the ambient PM 2.5 concentration by 10 µgm −3 life expectancy can be increased by 0.6 years.PM can be described by its aerodynamic equivalent diameter (AED) and particles are generally subdivided according to their size: < 10, < 2.5, and < 1 µm (PM 10 , PM 2.5 , and PM 1 , respectively).Particles with a diameter greater than 10 µm have a relatively small suspension half-life and are largely filtered out by the nose and upper airway if inhaled.Particles with diameters between 10 and 2.5 µm (PM 10−2.5 ) are referred to as 'coarse', less than 2.5 µm as 'fine', and less than 1 µm as 'ultrafine' particles.It is important to note that PM 10 encompasses ultrafine (PM 1 ), fine (PM 2.5−1 ), and coarse (PM 10−2.5 ) fractions.
During winter, towns and cities in New Zealand suffer from elevated levels of PM :::::::: primarily resulting from the burning of wood and coal for home heating (Ministry for the Environment & Stats NZ, 2018).Poor air quality is a more frequent problem in cities and towns that are located in the South Island.This reflects the climatologically colder winters, that occur in the South Island, resulting in greater use of solid fuel for home heating and the formation of capped boundary layers that restrict the dispersion of pollutants being more likely.This study presents measurements of PM made during a winter field campaign in Christchurch in 2019.Christchurch is New Zealand's third largest city (population of 385,500 as at June 2019) and is one of the most polluted cities in New Zealand.To provide regional councils :: the ::::::: regional ::::::::::: government ::::::::: responsible ::: for ::::::::: managing :::::::: emissions ::: of ::: PM : with legislative tools to address poor air quality, the New Zealand government defined national environmental standards (hereafter NES) for air quality in 2004 and updated these in 2011.The standards include five main air contaminants, viz.PM 10 , sulphur dioxide (SO 2 ), carbon monoxide (CO), nitrogen dioxide (NO 2 ), and ozone (O 3 ).Each contaminant is monitored in 89 geographical regions surrounding urban areas known as airsheds, Christchurch lies within a single airshed (Fig. 2).Within each identified airshed, a limited number of PM 10 exceedances of a daily mean limit of 50 µgm −3 are permitted each year (one for some airsheds, three for others).However, the PM standard is currently under review with the expectation that the primary standard for PM pollution will shift from PM 10 to PM 2.5 in recognition of PM 2.5 being more relevant for assessing health impacts, since it penetrates deeper into the lungs than PM 10 .This proposed change will bring New Zealand's air quality standards up to : in :::: line :::: with those suggested by the World Health Organization (WHO Regional Office for Europe, 2017).As such, while PM 10 , PM 2.5 and PM 1 were measured during the field campaign, this paper focuses primarily on PM 2.5 .

The Mapping Air Pollution eMissions (MAPM) project
The goal of the MAPM project, funded through the New Zealand Ministry of Business, Innovation and Employment, is to develop a method for inferring daily, high spatial resolution (< 100 m) PM 2.5 emissions maps for cities.The MAPM method uses an inverse model that takes as input in situ PM 2.5 mass concentration measurements and the meteorological data required to calculate trajectories from sources to receptors (instrument locations) and generates PM 2.5 emissions maps and their uncertainties :::::::: (hereafter ::::::: referred :: to :: as :::: 'the ::::::: MAPM :::::::::::: methodology').Several linked lines of development, conducted in parallel, form the basis of the MAPM research: 1.A field campaign to generate the data required to test and validate the MAPM methodology.The purpose of this paper is to describe in detail this field campaign and the resultant data.
2. A forward model that simulates the local meteorology over the duration of the campaign.This model is used to drive Lagrangian particle dispersion trajectories and produce source-receptor relationships between the PM 2.5 sensors and the emissions sources.
3. An inverse model that takes the source-receptor relationships, in situ PM 2.5 concentration measurements and a prior emissions map as input to generate daily maps of sources of PM 2.5 emissions and their uncertainties.
4. Several Observing System Simulation Experiments that are being used to explore the effects of different (i) instrument configurations, and (ii) instrument types and associated measurement uncertainties.
Between May and November 2017 an additional 10 low cost nephelometers units were deployed to focus on denser measurement networks to investigate the prevalence of spikes and airshed boundary gradients using the 2016 spatial characterisation of the airshed.Both the 2017 and the 2016 campaign found significant spatial and peak PM differences with : to : the data from the 3 permanent monitoring sites.
Within MAPM, the measurements from the 2016 and 2017 measurement campaigns were combined using a regression model to create high resolution hourly PM 2.5 maps for Christchurch, which were then used as input to an algorithm that selected locations for the placements of Outdoor Dust Information Node (ODIN) and ES-642 instruments for the 2019 campaign ( :::: refer :: to Sect.3).
Another measurement campaign was undertaken in autumn 2016 by Huggard et al. (2019).18 ODIN nephelometers were installed in Rangiora, a small town 20 km north of Christchurch.Data from these were compared to measurements made by a permanent TEOM also installed in Rangiora.Huggard et al. (2019) analysed several methods of correcting ODIN PM data against a TEOM reference.They found little benefit in increasing the instrument co-location period beyond seven days and that a correction based on relative humidity was optimal.

Description of Christchurch meteorology and sources of particulate matter
Christchurch is the main urban centre of the Canterbury region, which is situated on the east coast of New Zealand's South Island.It is located on the eastern fringe of the Canterbury Plains that slope gently from the coast to the Southern Alps that rise to elevations well above 3000 m.While Christchurch is situated on generally flat terrain, immediately south of the main urban area, the Port Hills form the northernmost side of the volcanic landscape of Banks Peninsula, provide a local orographic feature that reaches elevations of up to 450 m (Fig. 1).
Dwellings in the urban area of Christchurch are mainly single story houses and buildings higher than 5 stories are rare in the city centre.The current tallest building in Christchurch rises to 86 metres.Many of the high-rise buildings were demolished following a series of major earthquakes in 2010 and 2011.Christchurch has a relatively low population density (270 km −2 compared to 1, 510 km −2 for London, UK).In the centre of Christchurch is Hagley park with an area of 1.65 km 2 in this area, very little PM emissions occur.
Christchurch has a temperate maritime climate with warm dry summers and winters in which it is common for temperatures to fall below 0 • C overnight.There are, on average, 70 days of ground frost per year.Snowfalls occur on average once or twice a year on the Port Hills and about once every two years on the plains.The dominant topography that modifies the synoptic flow around Christchurch are the Southern Alps which form a roughly perpendicular obstacle to the predominant westerly wind.
The resultant foehn-type winds lead to Christchurch having relatively low rates of rainfall that limit rainout of airborne PM pollution.The second most common wind in Christchurch is an onshore easterly wind that flows parallel to the Port Hills, which also induces the majority of the rainfall.
During winter, the main source of PM 2.5 emissions in Christchurch is burning wood and coal for home heating.Further minor anthropogenic sources result from industry and transport with natural sources including dust and sea salt.ECan monitors PM 10 at two locations in Christchurch (Woolston and St Albans) to provide the data needed to detect exceedances of the NES permitted thresholds.High pollution days can often be related to several precursor states occurring in concert such as meteorological conditions, topography influencing air mass movement, and short-term emission sources such as passing heavy or poorly serviced vehicles (Mukherjee and Toohey, 2016).
In 2019, Christchurch reported seven days where the daily mean PM 10 concentration exceeded the 50 µgm −3 NES permitted threshold (i.e.four days more than is currently permitted; from 1 September 2020, only a single exceedance is permitted each year).The proposed new limits for any airshed are: (i) no more than three exceedances of 25 µgm −3 for daily mean PM 2.5 and (ii) an annual mean PM 2.5 concentration of no more than 10 µgm −3 .During winter, 90 % of all particulates measured as PM 10 comprise particles smaller than 2.5 µm (Aberkane et al., 2010).A series of major earthquakes occurred in 2010 and 2011 in Christchurch, resulting in major structural damage, which substantially increased the reliance on woodburning for home heating.This, together with intensive construction and demolition activities elevated several sources of PM pollution (Tunno et al., 2019).On the other hand, major damage led to many homes being removed, people moving away and, older wood burners being replaced with lower emission burners or electrical heating, leading to reduced PM emissions.
On the 1 January 2019 the use of 'old style' wood burners was banned on any property smaller than 2 ha within the Christchurch Clean Air Zone (Fig. 2).After this date the installation of a burner that did not meet the 'ultra low' emissions standard was also banned on properties smaller than 2 ha within the Christchurch Clean Air Zone.Ultra low emissions burners must not exceed 38 mg of emissions per MJ of useful energy output and must have a thermal efficiency greater than 65 %.
Sources of PM in Christchurch's surrounding areas include agricultural fires and agricultural dust, as well as sea salt from the nearby ocean.Agricultural fires occur predominantly between February and March and are often forbidden during summer for safety reasons.Golders Associates (2014) investigated the impact of burning of crop residue and found that while agricultural fires were not likely to cause an exceedance of the NES, large spikes in PM 10 were possible at hourly timescales and recommended that agricultural fires are not burned within 6 km of an urban area.
This paper describes each of the instruments used in the campaign (Sect.2), the algorithm used to decide where to locate the sensors (Sect.3), how the sensors were inter-calibrated and the QA/QC (Quality Assurance/Quality Control; Sect.4), the method used to derive the uncertainties on the PM 2.5 measurements (Sect.5), with a final description and presentation of the data in Sect.6. Concluding remarks regarding the intended use of the data are provided in Sect.7.
Measurements from these AWSs were complemented by measurements from AWSs operated by the Meteorological Service of New Zealand (MetService) and the National Institute of Water and Atmospheric Research (NIWA), as well as meteorological measurements made by the public and submitted to the United Kingdom Met Office weather observation website (WOW; https://wow.metoffice.gov.uk/).A micropulse lidar and a ceilometer installed on top of a building (45 m altitude above surface) measured vertical profiles of aerosol concentration.To investigate the stability of the boundary layer, its height, and to identify the occurrences of temperature inversions, 12 balloon-borne radiosondes were also deployed during the field campaign.

ES-642 remote dust monitor
The ES-642, produced by Met One Instruments, Inc., is a type of nephelometer which automatically measures real-time airborne particulate matter concentrations using the principle of forward laser light scatter.The sensor has a prescribed accuracy of ±5 % and a sensitivity of 1 µgm −3 (Met One Instruments, Inc, 2019).Air is drawn into the sensor through a sharp-cut cyclone to prevent particles larger than 2.5 µm entering the sensor.The accuracy of a nephelometer is hindered by water vapour present within the sample air.As relative humidity increases above 50 % particles begin to aggregate and increase in size due to water absorption (Di Antonio et al., 2018).To mitigate these effects, a 10 W inlet heater is used to warm the incoming air and thereby lowering the relative humidity of the air entering the sensor, preventing the intake of water vapour.The heater turns on when the ambient relative humidity reaches values above 40 %.The sampled air then passes through the laser optical module where the suspended particles in the air stream scatter the laser light through reflective and refractive properties.This scattered light is collected onto a photodiode detector at a near-forward angle, and the resulting electronic signal is processed to derive a continuous, real-time measurement of airborne PM concentrations.
The ES-642 instruments were provided by MOTE Ltd. and were coupled with data modems to transmit data in near real-time.
Although automatic data transmission can be enabled, this functionality was not used during the MAPM field campaign to improve instrument reliability.Instead, data were periodically retrieved from the SD card.Power is drawn from an on-board battery that is charged by a small solar panel, allowing for units to be installed in remote locations, independent of a power source.

Tapered Element Oscillating Microbalance (TEOM)
Three Tapered Element Oscillating Microbalance Filter Dynamics Measurement System (TEOM-FDMS, hereafter referred to as TEOM) instruments were running in Christchurch during the MAPM field campaign as part of the permanent observing system installed by ECan and provided data at hourly resolution.The TEOM instruments were co-located with and :: an : ES-642 and an ODIN instrument at the Woolston and St Albans sites and with an ES-642 at the Riccarton Road site (Fig. 2).The TEOM continuously measures PM 2.5 and PM 10 concentrations and are classified as equivalent to gravimetric measurements by the US Environmental Protection Agency (Charron, 2004).Gravimetric measurements are based on weighing the mass of particulate matter that accumulates on a filter after air has passed through the filter over a prescribed time period, generally 24 hours.The TEOM measures PM concentration by passing air through an oscillating filter (Patashnick and Rupprecht, 1991).As PM accumulates on the filter, the inertia of the filter and thus the frequency of oscillation of the filter changes.The instrument therefore measures particulate matter mass directly.

Automatic Weather Station (AWS)
Three temporary AWSs were installed specifically in support of the MAPM field campaign.These were deployed to supplement measurements from AWSs operated by MetService, NIWA and by members of the public who made their data available through the Weather Observation Website (WOW ) ::::: WOW maintained by the United Kingdom Met Office.While data from all of these AWSs (a total of 30 instruments) have been used in the MAPM project, only the three dedicated MAPM AWSs will be described and here.Measurements were made using a Unidata LM34 temperature sensor, a Vector W200P Potentiometer wind vane to determine the wind direction and a Vector A101 anemometer to measure wind speed.The data were logged using a Unidata Starlogger 6004D-2, which averaged 3-second data to a 10-minute resolution and recorded the averages, the standard deviation and the minimum and maximum values measured within the preceding 10 minutes.
The instrument locations were chosen to complement the network of permanently installed AWSs.Observations at the exterior of the city were preferred to provide information on any inflow of PM across the perimeter of the city.Two AWSs were located in rural fields just outside the suburban city area, while the third was located in a park within the ::::::: installed :: on ::: an ::::::::: abandoned :::::: airfield :::::: towards ::: the :::::::: perimeter :: of ::: the : city.The instruments were installed 2 m above the local foliage (one instrument was located in a field containing a 1.5 m tall crop so was installed 3.5 m above the surface).All AWSs were installed at least 50 m from the nearest tall obstruction.
Extensive quality control was performed on all AWS data, which is described detail in Sect. 4.

Vertical profile measurements
The vertical stability of the atmospheric column has a strong effect on the distribution of aerosols.During night-time, radiative cooling at the surface of the atmosphere causes temperature inversions to form in the lower layers of the atmosphere.These regions of stable air prevent mixing of aerosol above the boundary layer.Therefore, to accurately simulate the transport of aerosol across a city, it is essential for any transport model to correctly represent the planetary boundary layer height (BLH).

Mini micro pulse lidar (miniMPL)
A Sigma Space mini micro pulse lidar :::::::: miniMPL : was installed on the roof of the Rutherford Regional Science and Innovation Centre at the University of Canterbury (43.5225 • S, 172.5841 • E) at an altitude of 45 m above sea level.This building is approximately 30 m high and is surrounded by several buildings of similar height.The university campus is otherwise surrounded by a residential area of primarily single-and two-story houses.The miniMPL was installed on 17 July 2019 and operated by the University of Canterbury until the end of the MAPM field campaign.
The MiniMPL is a dual-polarisation micro pulse lidar operating at a wavelength of 532 nm at pulse repetition frequency of 2.5 kHz, with a maximum range of 30 km (Spinhirne et al., 1995;Campbell et al., 2002;Flynn et al., 2007).The MiniMPL is an aerosol backscattering lidar and a detailed description of the lidar instrument can be found in Ware et al. (2016).The MiniMPL operates similarly to other lidars and operates continuously with a temporal resolution of 2 minutes.
The instrument produces native binary files with backscatter and housekeeping meta-data, which can be converted to netCDF files using manufacturer supplied software (SigmaMPL).The measurements from this campaign have been used in Kuma et al. (2020) to demonstrate the potential of a ground-based lidar simulator for model evaluation of cloud properties.The instrument is also sensitive enough to measure aerosol backscatter on a continuous basis and can therefore be used to infer boundary layer height.

Radiosondes
Radiosondes are small balloon-borne instruments that measure the vertical profile of temperature, relative humidity, and pressure.Depending on the radiosonde type, pressure is either directly measured or inferred from the altitude of the instrument.
Altitude, wind direction and wind speed are calculated from the Global Positioning System (GPS) location of the sonde.
As part of the MAPM field campaign 12 GRAW DFM-9 radiosondes were launched.The radiosonde measurements were used to identify stable inversion layers that typically form during cold and calm periods, particularly at night-time.A thermistor is used to measure the temperature with an accuracy of ±0.occurring (Fig. 7).The primary goal of the balloon launches was to sample the air within the boundary layer.To increase the sampling rate in the boundary layer, all balloons were underinflated with a target ascent rate of 3 ms −1 compared to the commonly used 5 ms −1 .

MAPM Field campaign design
We sought an optimal set of 50 sites around Christchurch city whose pollution measurement times series would be as different as possible from those at every other site.This :::::: design philosophy would maximise the information content of the time varying PM concentration field sampled at the 50 sites.To accomplish this we first developed a method for generating hourly spatiallyresolved PM 2.5 concentration maps over the domain from point source PM measurements and model output.

Hourly concentration maps
The measurements used in the concentration maps were made by MOTE over the winters of 2016 and 2017 (Sect.1.2), extreme outliers were removed and hourly averages were then calculated.We fitted a least squares regression model to every winter day over 2016, and 2017 separately using the hourly PM 2.5 measurements.The basis functions in the regression model contained spatially resolved, modelled winter maximum and winter average concentrations expanded into six Fourier terms.
The modelled winter maximum and winter average of PM 2.5 concentration fields were obtained from Golders Associates (2016), and compromised 137x137 grid cells over Christchurch.For every hour the residuals of the fits were calculated and then kriging was used to interpolate this field across the whole model domain, creating the delta map.Finally the regression model was evaluated at each grid point, and combined with the delta map, : producing the gridded hourly maps of PM 2.5 concentration over Christchurch during the 2016 and 2017 winters.These maps then guided the process for locating the instruments deployed during the campaign.
4.2 PM 2.5 QA/QC and correction All PM 2.5 measurements were corrected using data collected during two co-location periods: i a pre-campaign co-location that ran from 6 June 2019 1700 NZST to 12 June 2019 1700 NZST ii a post-campaign co-location that ran from 30 August 2019 1900 NZST to 8 September 2019 1900 NZST For both co-location periods, all PM instruments together with the TEOM instrument were located at the Woolston site (43.5572• S and 172.6811 • E).The instruments were mounted on a scaffold approximately 3 m above the ground.

Smoke barrel tests
Smoke barrel tests were performed on all the ES-642 instruments before the initial co-location and after the final co-location.
For these tests groups of six ES-642s were set up so that their inlets were drawing air from a closed barrel.Fans were used inside the barrel to ensure the air inside was well mixed.Wood smoke was introduced to the barrel and the concentration of PM 2.5 was measured by each ES-642 as the smoke gradually dissipated.These measurements were made as a potential alternative to the co-location periods as a method of calibration.
However, the measurments made during the co-locations were used for the correction of PM 2.5 measurements instead of the smoke barrel tests because; -The co-location periods were considerably longer than the smoke barrel tests, allowing for a more statistically certain calibration.
-The co-location periods occurred over a larger range of meteorological conditions, allowing for a more sophisticated correction to be applied.
-The smoke barrel tests were composed of three separate tests.This would result in three groups of ES-642s that may be calibrated well against each other, but there would be potentially large, unknown variations between the three separate groups.
-By using the co-locations as our method of correction we are able to apply the same methodology to the ODINs and the ES-642s ensuring consistency between the two instrument types.
-While the smoke barrel tests only ensure internal consistency among ES-642s, using the co-location periods allows us to correct the ES-642 measurements against the TEOM instrument ensuring we have consistency with a gravimetric equivalent reference.
A potential downside of using the co-location periods over the smoke barrel tests is that the co-locations only cover a limited range of PM concentrations.This means that for periods of high PM concentration during the deployment period the calibration may have to use an extrapolation.This is of particular concern as the initial co-location period coincided with a period of low PM 2.5 concentrations.The smoke barrel test however, would span PM values from zero to much greater than would be expected to occur during the campaign period.

ODIN time retrievals
The ODIN instruments had no built in absolute reference for time.The time was set each time the instrument was installed and the instrument required constant power to the board in order to keep time.This meant that if an ODIN restarted during the campaign the time on the instrument would reset to the time that the instrument was originally started at.During the campaign ODINs restarted for a variety of reasons, presumably due to either low battery voltage (and then restarting once the solar panel recharged the battery), or due to a short on the circuit board due to ingress of debris or moisture.This resulted in several large sections of data being recorded that were unusable due to the timing of the data being unknown.

Automatic weather station (AWS)
After applying coarse limit tests on each of the AWS data streams (Appendix A), measurements of i air temperature ii relative humidity iii wind speed iv wind gust speed v air pressure from the 30 AWSs were tested for internal consistency.The purpose of the tests was to identify data that was recorded erroneously.Before conducting these internal consistency checks, for air temperature, all measurements were reduced to sea-level temperatures assuming a moist adiabatic lapse rate of 6 • Ckm −1 .For air pressure, the values were reduced to sea-level using the hydrostatic approximation assuming a layer mean temperature of 9.85 • C. For air temperature and wind speed, comparisons between sites were challenged by some sites providing measurements as 1-minute means and other sites providing measurements as 10-minute means.As such, 10-minute 'synchronised' means were calculated for all data across all locations, i.e. means were calculated in common 10-minute blocks centred on 5, 15, 25, 35, 45 and 55 minutes past the hour.
An important difference between the two uncertainty estimates is the temporal resolution at which they can be derived.

Instrument type accuracy
The second component of the measurement uncertainty corresponds to answering the question of: "How likely is it that the average of measurements taken by the ensemble of all instruments of the same type are the same as the measurement from a reference instrument?".
To derive the second component of the overall uncertainty on a measurement (ε x ), the differences between the expected measurements of an instrument type (or the average of the individual instrument measurements -the cohort average) and the measurements from a reference instrument, in this case the TEOM, were calculated.With these differences the dependencies with environmental factors can be determined.
There was no strong correlation in the instrument type accuracy of either the ODINs or ES-642s with either hourly mean temperature or relative humidity, nor was there any correlation of the uncertainty estimates with higher measured concentrations.
As a result, this second component of the measurement uncertainty can be added as a constant to the more dynamic intra-instrument variability.The instrument type accuracies from pre-and post-campaign co-location data were again slightly different.Therefore, this uncertainty type was interpolated over the deployment period, but was the same for any date-time for each instrument of type ODIN or type ES-642 respectively.
6 Data and analysis Temperature and relative humidity profiles were measured on 12 radiosonde flights during the two intensive sub-campaigns as detailed in Sect. 2. The boundary layer is of specific interest as its stability influences the concentration of pollutants such as PM 2.5 at the ground level.The ODIN instruments measured both PM 2.5 and PM 10 .Although the goal of the campaign was to measure PM 2.5 , the PM 10 data were used as adiagnostic tool for the PM 2.5 measurements.We define the dimensionless value R as the ratio of PM 2.5 /PM 10 .
In Fig. 11, R derived from measurements as two ODIN sites is compared: ODIN 172, a site near the centre of the city (Fig. 11b; 43.517 • S, 172.615 • E) and ODIN 156, a site on the eastern coastline (Fig. 11d; 43.498 • S ::::: 43.506 • S, 172.734 • E).The histograms of R for the city centre site (ODIN 172;Fig. 11b) show that under all wind directions the distribution of R had a mode of approximately 0.8 with values of R rarely falling below 0.6.This indicates that the majority of particles smaller than 10 µm were measured to also be smaller than 2.5 µm.PM sources such as home heating and transport primarily produce particles smaller than 2.5 µm.The histograms of R for the coastal site (ODIN 172; Fig. 11d) show that R has large variations that are dependent on the wind direction.During periods of westerly, offshore winds (red and green)the R distributions closely resemble to those at the city centre site with modes of approximately R = 0.8.However, during periods of easterly, onshore wind (blue and orange) the distribution of R has a mode of approximately 0.45 with R exceeding 0.6 less than 10.0 % of the time.This is consistent with a population of larger particles, primarily made up of natural sea-salt, entering the city from the ocean.ODIN 172 was 9.36 km at 257 • from ODIN 156.Although the distance between these sites was small the inland site rarely saw values of R smaller than 0.6.This highlights the increased rate of deposition that occurs in larger particles compared to smaller (< 2.5 µm) particles ::: this ::::::::: highlights ::: the :::::::::: constraining ::::: effect :::: that :::::::: inversions :::: have ::: on ::::::: aerosols.The fit coefficients calculated from the pre-and post-campaign co-location periods used to correct the PM 2.5 data forming version 1 of the dataset are shown in Fig. 8.For instruments whose data was corrected against a single co-location period, due to a failure during the other co-location period, the stationary coefficient used is plotted as either a square (corrected against co-location 1) or a triangle (corrected against co-location 2).The a fit coefficients (Fig. 8a) decreased from the first 635 co-location to the second for all instruments except one.Similarly, the b fit coefficients decreased for all ODINs (Fig. 8b; red) and increased slightly for all ES-642s (blue).These coefficient drifts are likely due to the differing conditions that occurred during the two co-location periods.The two co-location periods occurred at different times of the year, the PM sources would differ at these times due to seasonality of natural sources as well as differences in human activity.The synoptic time scale weather patterns that occurred during the co-locations would also have an effect on the sources of PM at the co-location site.

640
Differing PM sources will change the size distribution and chemical make-up of the PM which may result in a change of the sensitivity of the sensor.Huggard et al. (2019) showed that although the fit did improve as the amount of the training data was increased, when training a regression model between ODIN data and TEOM data, increasing the training period from 7 to 14 days only reduced the mean squared error (MSE) by 3.8 %.This gain is minimal considering that it requires the sacrifice of valuable deployment period data.Huggard et al. (2019) also found that some time periods produced anomalous calibration values.Because of this we recommended that for future campaigns data are corrected using a series of short co-locations.If weather patterns present during the co-locations are anomalous for the given season, the co-location should be repeated as it may not be a fair representation of the seasonal PM emissions that are to be measured.
With the exception of one ES-642, all ES-642s generally showed a smaller change in magnitude of both coefficients between the two co-locations.ES-642s are able to heat incoming air, preventing the relative humidity of the incoming air exceeding 40 %.This reduces the errors caused by the misidentification of water vapour as PM.ES-642s also used sharp-cut cyclones to prevent PM greater than 2.5 µm entering the sensor.These factors mean that ES-642s are less susceptible than ODINs to environmental changes such as changes in humidity or particle size distribution.This is likely the reason why the change in fit coefficients, from the pre-to post-campaign co-location, for the ES-642s is smaller than that for ODINs.
A comparison of the ::::::::   and version 2 for both types of instruments and how the bias depends on the temperature and relative humidity measured by the instrument.The best agreement between the ODIN and the TEOM occurred with the version 2 correction (Table 2).ODINs do not have a built in mechanism to reduce uncertainty resulting from water, which causes particles to aggregate and increase in size.The uncertainty of ODIN measurements is therefore increased during periods of high ambient relative humidity (Fig. 9gi).The version 2 correction includes a correction based on relative humidity; this is, in part an explanation for why the version 665 2 performed better.The mean bias between the raw ODIN data and the TEOM at St Albans is 0.42 µg −3 (Fig. 9a) this is less than that of the version 2 (Fig. 9c).However, the mean of the raw data differs significantly from the mode of the distribution and the bias shows strong asymmetry in its distribution.While the mean bias does not appear the depend on temperature, the variance on the bias, and therefore the uncertainty of the measurements made with this ODIN, increases at lower temperatures (Fig. 9d-f).Similarly, the variance in PM 2.5 bias increases when the relative humidity exceeds 80 %.These two trends may be related, as the relative humidity will generally increase as air cools.
In contrast, the ES-642s performed best when corrected using the simpler version 1 correction (Table 2).The version 2 performed worse than version 1 but was still an improvement on the raw data set.::: This :::::::: suggests ::: that ::: the :::::::: additional :: fit :::::::::: coefficients during the co-locations and deployment period.We thank Teresa Aberkane from ECan for providing us with TEOM data and offering advice on the PM corrections.We also thank Sam Edwards from NIWA who assisted in the deployment of ES-642 instruments.We thank the volunteers who hosted instruments on their private property including the Air Force Museum of New Zealand.We would like to thank the 735 Christchurch City Council for allowing us to install ODINs on the cities light poles.We acknowledge Marwan Katurji and Laura Revell for providing insight and local knowledge on picking instrument sites.We acknowledge Metservice, Niwa, and the United Kingdom Met Office for providing AWS data from during the campaign.We also thank Hamesh Patel and Johnny Lowis for assistance during the field campaign.

Figure 1 .
Figure 1.The geographical context for Christchurch showing the Southern Alps to the west, Banks Peninsula to the east, and the Canterbury Plains between the city and the Southern Alps.The inset shows a typical PM2.5 distribution around the city.Background image: © Google, Maxar Technologies.

Figure 2 .
Figure 2. Locations of the instruments deployed during the MAPM field campaign and the AWSs operated by MetService and NIWA.The solid black line indicates the boundary of the Christchurch Clean Air Zone, the black dotted line indicates the boundary of the Christchurch Airshed.Locations of AWSs operated by members of the public are not shown.© OpenStreetMap contributors 2020.Distributed under a Creative Commons BY-SA License.

Figure 3 .
Figure 3. Examples of typical co-located ES-642 and ODIN installs.Note the instrument on the left is a Dust Met Mote with an additional sonic anemometer.An ODIN was co-located with every ES-642 instrument for intercomparison purposes.

Figure 4 .
Figure 4.An example of a typical ODIN installation on a light-pole.