Research ProposalPDF Available

Particulate matter measurement with low-cost sensors - investigation of data quality and the benefit of data correction approaches

Authors:

Abstract

The transmission and analysis of data is one of the challenges of the 21st century. In the field of environmental measurement technology, existing broadband and wireless technologies have not been able to transmit data reliably and cost-effectively over long distances and in hard-to-reach places. LoRaWAN, an IoT technology, could be an energy-efficient, cost-effective and secure alternative as a narrowband technology in combination with battery-powered sensors and thus make an important contribution to the intelligent, largely wireless networking of objects, plants and machines (IoT), for example in the 5 municipal sector. In addition to ecological and economic benefits, the quality of life in modern, intelligently networked cities can be enhanced by real time data acquisition. However, the prerequisite is that the quality of the data acquired via this method is sufficiently good. This paper therefore addresses the question of the quality of particulate matter data collected by low-cost sensors. To determine this, an SDS011 particulate matter sensor from Nova Fitness was ported to LoRaWAN. The sensor was installed next to a governmental measurement station. In a test that lasted five weeks, data from the SDS011 sensor were 10 compared with those from the governmental station. Differences were identified and a correction approach was developed and applied. The efficiency of the approach was verified. Based on the results, it can be seen that the use of the low cost sensors has weaknesses. Problems can only be partially reduced. Nevertheless, the use of the low-cost sensors can be helpful for a flexible and cost effective collection of environmental data.
Particulate matter measurement with low-cost sensors - investigation
of data quality and the benefit of data correction approaches
Tobias von Kuyck-Studzinski*, Thoralf Buller** and Alexander Conrad*
Abstract
The transmission and analysis of data is one of the challenges of the 21st century. In the
field of environmental measurement technology, existing broadband and wireless techno-
logies have not been able to transmit data reliably and cost-effectively over long distances
and in hard-to-reach places. LoRaWAN, an IoT technology, could be an energy-efficient,
cost-effective and secure alternative as a narrowband technology in combination with
battery-powered sensors and thus make an important contribution to the intelligent,
largely wireless networking of objects, plants and machines (IoT), for example in the
municipal sector. In addition to ecological and economic benefits, the quality of life in
modern, intelligently networked cities can be enhanced by real time data acquisition.
However, the prerequisite is that the quality of the data acquired via this method is
sufficiently good. This paper therefore addresses the question of the quality of particula-
te matter data collected by low-cost sensors. To determine this, an SDS011 particulate
matter sensor from Nova Fitness was ported to LoRaWAN. The sensor was installed
next to a governmental measurement station. In a test that lasted five weeks, data from
the SDS011 sensor were compared with those from the governmental station. Differences
were identified and a correction approach was developed and applied. The efficiency of
the approach was verified. Based on the results, it can be seen that the use of the low
cost sensors has weaknesses. Problems can only be partially reduced. Nevertheless, the
use of the low-cost sensors can be helpful for a flexible and cost effective collection of
environmental data.
JEL-Classification:
Keywords: Fine dust measurement (PM2.5, PM10), LoRaWAN, Internet of Things, low-
cost sensors, SDS011, level-indication measurements, sustainability.
* Eberswalde University for Sustainable Development, Germany (contact: aconrad@hnee.de)
** bbw University of Applied Sciences, Germany
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
© 2022 by the author(s). Distributed under a Creative Commons CC BY license.
1 Introduction
Data collection, data transmission and data persistence in and from areas that are diffi-
cult to access, such as forest and agricultural areas, but also in municipal and rural areas,
could not be obtained in an economically or ecologically way until now. So far it was
only possible to collect data in a cumbersome way, for a limited time or a topic-specific
project. Up to now, scientific measurement data from the forest and environmental area
have been realised using cellular radio technologies (e.g. GSM, UMTS, LTE, 4G, 5G)
or WLAN technology. None of the existing data transmission technologies has been sa-
tisfactory in terms of range, energy consumption and data bandwidth. This is because,
with the current state of knowledge, it is not possible for these transmission technologies
to implement all three conditions equally well. WLAN technology, for example, is cha-
racterised by a good energy balance and a very good bandwidth with which the data is
processed. However, WLAN technology does not have a high range, so that measurements
in the forest and environmental area always depend on the good availability of a WLAN.
The radius in which measurement results can be obtained is also limited and the choice
of valid, reliable and objective comparative measurements is severely restricted. For the
cellular technologies in the mobile radio sector currently on the market, on the other
hand, efforts can be seen to be able to implement the range, the energy consumption
and the data bandwidth equally well. However, this is partly based on open standards
or proprietary rights.
The Internet of Things (IoT) can be a solution to this problem, especially the Lo-
RaWAN technology, which was developed in 2015. Using LoRaWAN, data collected by
sensors or actions performed by actuators can be collected, transmitted and stored by
radio over a distance of several kilometres in an energy saving and at the same time cost
effective manner. LoRaWAN is an open LPWAN system architecture that operates in
the non licensed radio spectrum and thus has the potential to collect data from areas
that could not be accessed before. This makes this technology suitable for industrial (e.g.
smart metering in the energy sector) and municipal (smart parking in France and Dubai)
examples of applications, as well as for the individual technology enthusiastic for private
use (e.g. beekeeper’s scales to transmit the weight of a beehive as a measured value).
This technology also offers many advantages for scientific research due to the low ener-
gy consumption, the measuring ranges for the transmission of data, as well as the cost
savings due to the royalty-free radio technology. The constant technical development of
the hardware and software, also with regard to the costs for these, enables further future
application examples that can be implemented with and through LoRaWAN.
1
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
One area of application that has become more significant for humans and nature in
recent years is the particulate matter (PM) pollution at the respective place of residence.
Not only because recent research is showing an interaction of air pollution and COVID-19
mortality (Wu et al. 2020), but also due to the recently established scoreboard on air
quality in European cities and diverse discussions about summer driving bans in case of
high PM pollution, it must be the common goal of all municipalities and cities to reduce
PM pollution to protect humans and nature. Due to the far reaching health effects,
air quality is monitored by state institutions with a network of air monitoring stations.
But due to the high costs per air monitoring station, these are spatially distributed in
small numbers. However, particulate pollutants are mobile and subject to spatial and
temporal variance. Citizens’ initiatives and individual projects are trying to determine
the environmental pollution of individual areas more adequately with the help of digital
tools and citizen science communities. It is their aim to shape their own city with this
additional information in a healthier and more livable way. If and to what extent these
measurement results are credible, reliable and objective has not yet been scientifically
determined in a larger study. Indeed, only a view comparative measurements of state
measuring stations for PM2,5, PM10 and PM measurement on LoRaWAN basis exists so
far in the German-speaking area which takes place under conditions comparable in time
and space. Benabbas et al. 2019 provide one of the few studies available.
This study presents further analysis on this issue. It demonstrates a pragmatic way
of collecting scientific data using LoRaWAN which makes it possible to collect, store
and evaluate data in municipal sectors (and others) economically, efficiently and in an
energy-saving manner.
This is examined within the framework of this elaboration using the example of
PM measurement in Eberswalde (Brandenburg, Germany) at a comparative measuring
station with a reference source from the “Landesamt für Umwelt Brandenburg” (LfU). For
this purpose, the PM sensor including the necessary infrastructure was designed, built
and installed to enable the transmission of the data using the LoRaWAN technology.
A limiting factor of the sensors currently available on the market is the inaccuracy
of the measurement results in certain situations (e.g. weather). Within the framework,
approaches to overcome problems of inaccuracy will be presented. The approaches will
be statistically evaluated.
The following questions will be answered: Are the PM data measured using low-cost
sensor technology comparable with those of the LfU? Can data adjustments contribute
to comparability?
To answer the questions the paper is organized as follow: Chapter two describes the
2
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
current state of the art with regard to environmental measurement techniques, the go-
vernmental measurement system and non-governmental initiatives. Chapter three deals
with the methods and the materials used like the workflow, the hardware and softwa-
re, the study area, the monitoring as well as the case study for the particulate matter
measurement. The results are statistically evaluated and summarised in the fourth chap-
ter. Approaches to deal with inaccuracies are presented and discussed. In chapter five, a
summary is made and an outlook on further possible research is given.
2 Environmental measurement and particulate matter mea-
surements
2.1 Health protection against particulate matter as a state task
Maintaining air quality where it is good and improving it where it is not - this is the
basic idea of the EU Directive on ambient Air Quality and cleaner Air for Europe (Air
Quality Directive, see EUR-Lex 2008), which was implemented in Germany with the 39th
statutory order on the Implementation of the Federal Immission Control Act (BlmSchG,
see ELAW 2002). It is the aim of the BlmSchG to achieve levels of air quality that do not
have negative significant impacts on human health and the environment. Further mea-
sures and guideline values have been defined, which are intended to protect citizens from
exposure to particles and ozone in the air, as well as ecosystems from acid disposition,
excessive nitrogen accumulation and ozone (EUR-Lex 2005). The WHO estimates that
approximately seven million people die each year due to air pollution (WHO 2021a). A
particular dangerous and harmful emission for humans and the environment is PM. PM,
also known as aerosol particles, are solid particles that exist as suspended dust in the
atmosphere (MLUK 2021a). The chemical composition of PM is diverse and depends on
the respective emission source. For example, PM can have a natural origin through forest
and bush fires, sandstorms and volcanic eruptions. However, the following anthropogenic
causes are responsible to a considerable extent for the formation of PM: motorised road
traffic, agriculture, the energy sector, industrial facilities and processes, as well as private
and commercial heating systems (see WHO 2021b; UBA 2020; MLUK 2021a). In addi-
tion to the emission source, the size of the particles is a decisive differentiating factor. A
distinction is made between coarse particles with an aerodynamic diameter smaller than
10 µm (PM10), fine particles with a diameter smaller than 2.5 µm (PM2.5) and ultrafine
particles with a diameter of less than 0.1 µm. The smaller the particles, the further they
can penetrate into the respiratory tract. Particles with a diameter of more than 10 µm
3
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
only reach the throat and nasal mucous membranes, also called inhalable airborne dust,
and are usually coughed up again. Particles smaller than 2.5 µm, on the other hand,
can reach the lungs and bronchi. They are referred to as respirable airborne dust (see
UBA 2009; MLUK 2021a; MLUK 2021f). Particles smaller than 0.1 µm can pass through
the lungs into the bloodstream, the tissue and the entire body. According to current
knowledge, they are also of health significance. Due to a lack of reliable standardised
measurement methods suitable for measuring ultrafine particles, and thus a sufficient
number of studies on the exposure-impact relationship for ultrafine particles is missing,
no guideline and limit values for ultrafine particles exist to date (see MLUK 2021a; UBA
2018). That´s why these particles will not be considered here further. Diseases caused by
PM are multifaceted (e.g. asthma, cardiovascular diseases) and can lead to an increase
in mortality and morbidity (WHO 2018). A concentration threshold in the ambient air
below which no harmful effect is to be expected does not exist for PM. PM is therefore
fundamentally different from other pollutants, such as nitrogen dioxide, and is always
harmful. Local and time-limited actions against PM, such as imposing driving bans, is
not very effective in the long term, since air currents transport PM over greater distances
and the PM itself can remain in the atmosphere for a certain time until it sinks to the
ground (see UBA 2009; UBA 2021c, p. 22). In order to improve the air quality in the
long term, the relevant authorities in the EU countries therefore jointly measure, moni-
tor and assess air pollutants and compliance with the specified limit and target values in
accordance with the Air Quality Directive (UBA 2019a).
2.2 Official air quality measurement systems
The monitoring and assessment of pollutants present in the air has been measured by
hundreds of air quality measuring stations in Germany for several decades (UBA 2021b).
In addition to PM10 and PM2.5, the most important air pollutants monitored are carbon
monoxide (CO), ozone (O3), sulphur dioxide (SO2) and nitrogen dioxide (NO2), which
are measured by the measuring stations of the ministries of environment of the respective
German federal states and by the German Federal Environment Agency (UBA 2021a).
There are legal requirements for the location, number of measuring stations and the mea-
suring methods used (see EUR-Lex 2008; MacDonnell et al. 2013, ES-1). An example of
this is the requirement to set up measuring stations where the population is exposed to
the highest concentration (UBA 2019b). This type of data collection results in expensive,
large, stationary measuring stations that, apart from being purchased, need to be regu-
larly maintained as well as calibrated and this is partly done by technical staff stationed
4
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
on site (see Williams et al. 2014, p. 50f.; UBA 2013, p. 50f.; Eickelpasch and Eickelpasch
2004, p. 61).
The location and number of air quality measuring stations in Germany corresponds
to the European requirement that the territory of a Member State is to be divided and
assessed into agglomerations, i.e. cities with more than 250,000 inhabitants, and other
assessment areas (MLUK 2021b). Area-wide coverage is not envisaged and a measurement
obligation for individual cities and municipalities cannot be derived (UBA 2019b). The
combination of the number of inhabitants and the pollution situation in each assessment
area determines the number and location, i.e. type of monitoring station, to be operated
for each pollutant in the respective monitoring area. The pollution situation is the place
where the population is most exposed to a certain pollutant, e.g. nitrogen dioxide is
mainly caused by traffic. In addition, concentration data is collected, i.e. representative
data for the exposure of the population to (certain) pollutants. This is done by measuring
stations in typical urban residential areas, also referred to as station type background
(UBA 2021c, p. 7). In addition to the two station types traffic and background, there is a
third measuring station type in industrial areas industry. Furthermore, the surroundings
of a monitoring station are classified as urban, suburban and rural / regional (MLUK
2021f). The combination of environment, number of inhabitants and station type thus
determines the number of monitoring station locations. The exact location of a monitoring
station depends on various local conditions and legal requirements. As a general rule, the
measuring stations should never be located in the direct vicinity of strong emission sources
such as traffic routes or industrial facilities (see EUR-Lex 2008, Annex III, IV). For
the sampling point, both large-scale and small-scale local regulations must be observed.
Large-scale means that the measurement of very small-scale environmental conditions in
the immediate vicinity should be avoided. Air samples from road sections, for example,
are only representative if the road section is at least 100m long (EUR-Lex 2008, Annex
III, B). In the case of measuring stations close to traffic, it should be noted in the sense
of small scale that the measuring stations should be set up no further than 10m from
the edge of the carriageway, at least 25m from a busy intersection and that interference
factors such as safety, power supply, accessibility and visibility of the measuring station
should be taken into account (see MLUK 2021b; EUR-Lex 2008, Annex III, C).
Although this classification and scientific categorisation of the measuring stations
appears to make it possible to compare the types of measuring stations, unfortunately
observation data can hardly be combined and compared with each other (Snyder et
al. 2013). The respective environment of the measuring stations (Helbig et al. 2013,
p. 208), weather data, geographical location and technical equipment used, which is
5
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
often also manufacturer-specific, are too different. In comparison with other national air
quality measuring stations, these are usually country-dependent and difficult to assess
and reproduce. Combining air quality data from heterogeneous sources is therefore a
major challenge, especially in urban and suburban areas (see Kotsev et al. 2016, p. 2;
MacDonnell et al. 2013, ES-1).
The measurement technology used to measure PM is an essential instrument for asses-
sing air quality (Eickelpasch and Eickelpasch 2004, p. 1). The performance of immission
studies in Germany is regulated by state provisions and is carried out on the basis of
state-recognised standards and guidelines. Part of these regulations concerns quality ass-
urance, which is ensured, among other things, by the use of standardised measurement
procedures according to VDI (Verein Deutscher Ingenieure) guidelines and standards,
suitability-tested measuring instruments, and reference, equivalence and calibration pro-
cedures (see Williams et al. 2014, p. 2; UBA 2021d; Eickelpasch and Eickelpasch 2004, p.
97; UBA 2021d). The measurement procedures are to be divided into discontinuous and
continuous. During discontinuous measurement methods, sampling usually takes place in
the field and analysis in the laboratory, and thus in two separate processes. This method
is advantageous for sample measurements, when many measuring points in the field,
different substances and also substances for which there are no automated measuring
methods yet, are to be investigated. In contrast, sampling and analysis take place in
one procedure in continuous procedures by means of automatically operating measuring
instruments, whereby a measurement without time gaps can be carried out. This measu-
rement procedure offers the possibility of stationary air monitoring without gaps in time
and can thus reveal air pollutants that only arise due to higher temporal rather than
spatial fluctuations (Eickelpasch and Eickelpasch 2004, p. 61). This procedure is main-
ly used for the implementation of government regulations (Eickelpasch and Eickelpasch
2004, p. 62). The procedural distinction does not apply without restriction, as indivi-
dual measurement procedures may possibly include components of both measurement
procedures. For example, discontinuous measurements can be automated both in samp-
ling by automatic sampling devices and in laboratory analysis by laboratory automates
(Eickelpasch and Eickelpasch 2004, p. 61).
In order to meet a certain data quality, minimum requirements for data collection
apply with regard to duration and possible acceptable uncertainties in the measurement
methods. For both fixed and indicative measurements, for example, 90% of the time of a
calendar year data per pollutant must be determined. Further parameters for high-quality
data collection are hourly and 8-hour averages, as well as daily (24-hour average) and an-
nual averages (see see ELAW 2002 - BlmSchV, A in Annex 1; Eickelpasch and Eickelpasch
6
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
2004, p. 50). If a method other than the reference method is used, the measuring station
operators must prove the equivalence of the method used (UBA 2021d). The purpose of
these quality requirements is the equivalence measurement of the occurring pollutants
and an equivalence of the measurement results in order to comply with the applicable
limit and target values. Table 1 lists the EU and World Health Organisation (WHO) limit
and target values for PM2.5and PM10 currently in force to protect human health. It is
noticeable that the WHO specifies stricter target values in its air quality guideline. There
are binding limit values and so-called target and guideline values. It should be the aim to
fall below as far as possible and therefore these value are only strong recommendations
(LfU 2021a). “But these limit values are more than 20 years old.” emphasised the UBA
president in spring 2021. “As the WHO is planning to develop new recommendations in
2021, the EU would also have to develop new limit and target values following the WHO
values”, he further stated (ÄrzteZeitung 2021).
Table 1: EU and WHO limit values for PM2.5and PM10
Pollutant Organisation Measuring time Max. value Liability
PM2.5EU Annual mean value 25 µg/m3Limit value
20 µg/m3Target value
WHO Annual mean value 10 µg/m3Target value
Daily average 25 µg/m3Target value
PM10 EU Annual mean value 40 µg/m3Limit value
Daily average 50 µg/m3Limit value*
WHO Annual mean value 20 µg/m3Target value
Daily average 50 µg/m3Target value**
Notes: * May be exceeded on max. 35 days / year; ** Must not be exceeded more
than 3 times according to the recommendation.
Source: own presentation; see EUR-Lex 2008; ELAW 2002; EEA 2017, p. 30; WHO
2005: 175; LfU 2021a; LfU 2021b.
When assessing the long-term development of PM, the respective weather conditi-
ons must be taken into account in addition to changing environmental conditions, as
weather-related fluctuations have a considerable influence on the level of pollution. In
cold weather, for example, the pollutants released increase because more heating takes
place. In wintry high-pressure weather, pollutants tend to accumulate in the lower air
layers due to low wind speeds and limited vertical air exchange. And ground-level ozone
is favoured above all by intense solar radiation and high temperatures. But wind speeds
also have a high influence on the level of pollutant load. High wind speeds and generally
7
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
good mixing conditions reduce the pollutant load. Interannual fluctuations and also an-
nual fluctuations during extreme weather phenomena are thus possible (UBA 2021c). In
Brandenburg, for example, so-called PM episodes occur in late autumn or winter, during
which the PM concentration is elevated for several days in a row. This is caused by wea-
ther conditions in which cold, pollutant-enriched air masses are not exchanged due to a
weak flow (MLUK 2021e).
The current air pollution situation is made available to every citizen by each federal
state within the framework of an area-wide presentation. These actual data, which are
mostly published online, are data that are determined by means of the interpolation me-
thod. With the method of optimal interpolation, hourly measurement and model results
are linked. The measurement results obtained represent the ambient air quality of a mea-
suring station without prior quality assurance of the measured values (see UBA 2021b;
MLUK 2021c). This is because quality assurance can only take place retrospectively by
means of the necessary laboratory tests. Final, validated and quality-assured measure-
ment data are published in the annual reports of the Federal Environment Agency and
in the annual reports of each ministry of environment of the respective German federal
states.
After these final, validated measured values are available, those areas are identified in
which the applicable limit values for the protection of human health and the environment
have been exceeded. Clean air plans must be developed for these areas within 2 years.
The aim and purpose of an air pollution control plan is to ensure, by means of suitable
measures, that air pollution is permanently improved by complying with limit values and
/ or reducing the period during which they are exceeded (see ELAW 2002). The respective
federal and local authorities cooperate with the public in an appropriate manner when
drawing up the clean air plans (MLUK 2021d). The main cause of localised limit value
exceedings is road traffic, especially in urban areas (VM 2021). The primary goal of many
clean air plans is therefore to reduce private motorised transport and to encourage citizens
to choose more sustainable forms of mobility, for example cycling, car sharing, electric
mobility and local public transport. The establishment of an environmental zone and the
introduction of bans on lorries passing through are further measures. But conceivable
measures are also possible in the field of urban development and urban land use planning.
2.3 Citizen science and air pollution measurements
Anyone and everyone can do research - this approach illustrates the idea of the citizen
science movement. The participation of citizens in acquiring knowledge and gaining in-
8
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
sights is itself not a new invention. Passionate amateur researchers who experimentally
investigated natural phenomena and took part e.g. in bird counts existed already centuries
ago (Leibnitz 2021). Nowadays, advancing digitisation and new technical developments
enable interested citizens to participate more actively in all steps of the research process.
Research questions are formulated, observations reported, measurements taken, data ana-
lysed and findings published. This increasingly creates a dialogue between science and
society (ACS 2021), because without the support of active citizens, science would no
longer be able to collect or evaluate large amounts of data itself (Leibnitz 2021). The
term citizen science itself is not clearly definable, as it is dynamic and changes over ti-
me (ACS 2021). The potential of citizen participation has recently been recognised by
science and different European states and, for example, the European Citizen Science
Association (ECSA) has been founded.
Research into air pollution control is one of the areas that the citizen science move-
ment supports and develops through the collection of data. For a few years now, low-cost
air pollution sensors that can be used with technical understanding have made it possible
for citizens to monitor air quality themselves. Citizens collect data, recognise correlations
and can thus influence municipal strategies for cleaner air (Snyder et al. 2013). Various
pollutants are measured by different citizen science projects such as the Air Quality
Egg or the sensor.community, formerly “luftdaten.info”. The latter is the best known in
German-speaking countries in the field of PM measurement. Furthermore, the HackAir
initiative of the German government, which has been in existence since 2018, supports
the citizens’ desire for cleaner air.
The various sensor-based citizen science projects differ mostly in terms of data trans-
mission technologies, measurement accuracy, prices for the components and whether mo-
bile or stationary measurement methods are used. The advantage of all projects is that
large amounts of data can be collected. Critically, however, laypersons collect measure-
ment data that does not adequately meet scientific standards of comparability, accuracy
and reliability. Many of the sensors used in the citizen science projects were tested by
the United States Environmental Protection Agency (EPA) in 2017 to assess the quality
of the measurement data. The SDS011 sensor used here was also tested by LUBW 2017
or Benabbas et al. 2019, with the result of a satisfactory correlation at humidity levels
of 50-70% and response to climatic variations (LUBW 2017, p. 20). As the market for
sensor measurement systems is subject to constant further developments, there are also
constant further developments with regard to measurement accuracy and more up-to-
date comparative studies, such as the one by the LUBW 2017 or from Benabbas et al.
2019, would be desirable.
9
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
3 Materials and methods
3.1 Workflow and software
Data transmission by means of IoT low-cost microprocessors and sensors based on Lo-
RaWAN basically consists of four topics: (1) Building the local gateway infrastructure,
(2) Construction and programming of the sensors, (3) Data storage and visualization of
the transmitted data, and (4) Statistical analysis of the transmitted data.
All these four topics contain different hardware and software components. They are
not structurally based on each other, but can be worked on in parallel. Figure 1 shows
schematically these four topics and the entire hardware and software included. First, a
PM sensor box was built, programmed, tested and then inserted into the created LoRa-
WAN network. Thus, the sensor box could now send it’s collected data, which refers to
PM2.5, PM10, air temperature, air humidity, as well as air pressure, to the databroker
TTN (The Things Network, see TTN 2021) every three minutes. Since TTN does not
offer visualisation and permanent persistence of the data, a way had to be found to ta-
ke care of this. Using the MQTT protocol, it was possible to build up the data stream,
which was provided in .csv/.json format, from TTN to a server of the Eberswalde Univer-
sity for Sustainable Development (Hochschule für nachhaltige Entwicklung Eberswalde -
HNEE), as well as an alternative private server. The LfU was able to retrieve this data
independently from the private server as a .csv file. In order to test measurement ad-
justment approaches, a function was built into the data flow with the help of NodeRed
to make this possible. From this point on, the rest of the process was double-tracked.
The data for the HNEE is provided in .csv and .json format so that the HNEE can
implement it in its PostgreSQL DB. In the private database, the data was transferred
to InfluxDB for subsequent visualisation and evaluation. Every 180 seconds, the sensor
send the measured values in a data packet of approx. 33 bytes to the next reachable ga-
teway. From this gateway, the measurement information is sent to the TTN data broker
via an internet connection. The information is kept there for up to seven days and must
be actively retrieved, otherwise it will be lost. The information is actively retrieved via
machine-to-machine communication using an MQTT network protocol and persisted on
a server. The data packets from the database are then visualised.
10
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Figure 1: Workflow
Source: own presentation.
3.2 Hardware
The used SDS011 sensor box is a further development of a former system, which was
initially operated with an SDS011, DHT11 ESP8266. Its use was limited to a few hours,
as it was powered by a power bank. By porting it to the LoRaWAN technology, it was
possible to use the complete sensor box with a 5V/1A power supply (e.g. smartphone
power supply) at any location within the existing LoRaWAN network in Eberswalde.
In order to connect all components of the sensor box (see table 2) with each other in
a modular way, an extension board was constructed and printed as a circuit board.
This enables a modular construction without soldering. Should it be necessary to replace
components, this is done in a sustainable way. A printed 3D housing made of PLA,
called a Stevenson Screen, enables the electronics to be enclosed and thus allows free air
exchange (see figure 2). At the same time, this housing prevents the penetration of adverse
weather conditions (e.g. heavy rain, heavy hail) and protects the entire electronics. The
air intake is via a 5cm long flexible, transparent hose. The total material costs for an
SDS011 sensor box are approx. 70 - 80 EUR.
11
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Table 2: Description of the PM sensor parts
Hardware used for the sensor box
1x MCU ESP32 SX1276 LoRa 868 MHz
1x Sensor SDS011
1x Sensor BME280
1x DIY Backplane connector for MCU and wires
1x Koaxial Pigtail, U.FL male to N female
1x LoRaWAN Antenna 868 MHz
1x DIY 3D Print Enclosure
1x Set of M2.5 screws and nuts
1x USB microcable
1x Power supply 5V-1A
Source: own presentation.
Figure 2: DIY PM sensor box based on SDS011 and BME280
Source: own presentation.
12
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
3.3 Comparison measurement sites and methods for PM measure-
ments
The study area “Breite Straße” is located in the eastern part of Eberswalde (north-east
of Germany in the federal state of Brandenburg) 20m above sea level. It is an urban
main traffic road, on which the Bundesstraße 167 runs. It is built up on both sides within
the urban boundary, partly the buildings are located directly at the edge of the roadway
with a narrow pavement in between. It has a traffic volume of 14,225 vehicles per day on
average (MLUK 2021c). The LfU measuring station is located in front of house number
22 in the Breite Straße. It is located between the pavement and a narrow bicycle lane
runs to the side of the street. The air inlet of the LfU measuring station for sampling is
at a height of 3.50 m and that of the SDS011 sensor box at a height of 3.15m. In 2020,
an average annual mean value for PM2.5of 10µg/m3and an average annual mean value
for PM10 of 14µg/m3were measured in the year-end report of the LfU. In 2020, the daily
mean value of 50µg/m3for PM10 was exceeded only once (MLUK 2020, p. 13f.).
Three measurement methods are used by the LfU: (1) the continuous measurement
with beta absorption, (2) gravimetric measurement with a low-volume sampler for samp-
ling and (3) a continuous scattered light measurement with the EDM 180 (LfU 2021c, p.
53f.).
According to the annual report 2019, the latter measurement method is used for air
quality in Breite Straße and, according to LUBW 2017 testing, determines equivalent
measurement results as the reference method of DIN EN 12341 (see LUBW 2005, p.
36; Grimm 2011, p. 5). The stationary dust measuring device EDM 180 from Grimm is
used for continuous measurement of dusts in the air and their aerosol distribution. The
particle rate is measured as a function of diameter according to the principle of scattered
light measurement and the mass concentration is calculated via a calculation factor. The
device can measure PM2.5and PM10 simultaneously (LUBW 2005, p. 15).
The actual data of this measurement are made available hourly approx. 20 minu-
tes after the measurement at: https://luftdaten.brandenburg.de/home/-/bereich/aktuell.
However, these are only pre-tested, preliminary measurement data that still have to be
compared with the reference procedure (daily filter sampling and subsequent gravimetric
determination in the laboratory according to DIN EN 12341: 2014). This takes place
retroactively for the complete calendar year, so that final quality-assured measurement
results of the Brandenburg air quality measurement network can only be taken from the
respective annual reports on air quality of the LfU (see LfU 2021d).
However, the measurement method used here by means of the IoT low-cost sensor
13
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
technology uses the laser scattering method, which is a variant of multiscan technology.
It can evaluate surface features with the help of laser triangulation, which is created
by the scattering of laser light when it hits a material surface. Scattering here basically
refers to anything that deflects the incident laser light from its original direction. The
particles detected in this way, which fall within the detection range, are converted into
electrical signals, amplified and further processed into information. The SDS011 sensor
from Nova Fitness sucks in air containing particles (e.g. PM particles) via a 5cm long
flexible, transparent tube. The measurement parameters of the sensor include the values
PM2.5and PM10.
The particle measurement parameters have a range of 0.0-999.9 µg/m3. The tempe-
rature operating range of the sensor is -10 to +50°C. According to the manufacturer,
the relative error of the sensor is ±15%, which corresponds to ±10µg/m3. The power
requirement is 5V (Nova Fitness 2015). The air sampling has a temporal resolution of
180 seconds, which corresponds to approx. 500 measuring points per day. The data is
forwarded to the MCU via a serial interface and transmitted to a gateway via LoRaWAN.
3.4 Data and statistical methods
The statistical data evaluation covers the period from 19.04.2021 to 24.05.2021 (12:00 to
12:00 CET). The possibility of installing a sensor box (SDS011) in the immediate vicinity
of the air inlet of the LfU measuring station in Breite Straße made it possible to generate
measured values under real conditions and to compare them with those of the reference
measuring station of the LfU.
The main objective of this paper is the statistical evaluation of the particulate matter
measurement values in Breite Straße with and without correction algorithm in comparison
with the measurement values of the reference data of the LfU. This allows also to measure
the efficency of the correction algorithm for the low-cost sensor (SDS011). Besides visual
comparisons, the following formulas are used to describe and evaluate the data:
The mean absolute error (MAE) measures the average absolute deviation between
the values of the reference source of the LfU station (x) and the values of the low-cost
sensor (y) while (n) indicates the number of measurements:
MAE =1
n
n
X
i=1
|xiyi|
The BIAS, on the other hand, indicates the systematic error, i.e. it indicates how
strongly the measurement is biased in a certain direction (sign) (Sachs and Hedderich
14
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
2006: 87f.):
BI AS =1
n
n
X
i=1
(xiyi)
The Root Mean Square Error (RMSE), i.e. the root of the mean square error, is a
measure of the spread of the data. It shows the average of the deviance between the
measured values and its mean (expected value). The formula for this is:
RSM E =v
u
u
t
1
n
n
X
i=1
(xiyi)2
Thus, the RSME measures the average strength of an error. In this work, it is a
measure of how much the values of the SDS011 sensor with its uncorrected as well as
corrected measured values deviate from the reference values.
The RE (relative error) shows the percentage of all SDS011 observations with a de-
viation of less than 15% from LfU. Thus, the ratio, which is based on the manufacturer’s
value for the relative error, provides additional information on the distribution and com-
parability of the data.
RE =number of observations with a deviation less then 15%
number of all observations ×100
4 Results
4.1 Results before correction
During the period mentioned, sensor box (SDS011) transmitted 17,551 data packets (one
data packet contains several measured values) with a time resolution of approx. 180
seconds via LoRaWAN. The LfU made its measured values available with a temporal
resolution of 30min and comprise 1,681 data packets. To make the data comparable,
average values were determined for the time span of 30 minutes (30min mean and 30min
median). After adjusting and merging the result is a data set with 1,466 observations.
Figure 3 and 4 presents a visual comparison of the captured data and table 3 shows
some descriptive results using the formula for MAE, BIAS, RSME and RE.
The statistical analysis shows the following results: Both PM2.5and PM10 show de-
viations from the reference values. However, the deviations for PM2.5are comparatively
small, and those for PM10 are about 2 times larger than those for PM2.5. With regard
to the BIAS indicator, it can be seen that on average the LfU values are higher than
15
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
those of the sensor box. The sign of BIAS is positive in each case. For PM10 the expected
value for the deviation is closer to zero. However, the value for RMSE also shows that the
values for PM10 are generally larger, i.e. that the LfU values and the sensor box values
generally show larger differences. Finally, the values for the indicator RE show that for
PM2.5about 80% of all observations the deviation between the LfU and the sensor box
values is bigger than 15%. For PM10 the indicator shows a slightly better rate.
The statistics clearly show that the sensor box covers the PM2.5values comparatively
well. With regard to PM10, however, there are more significant deviations. The next
section deals with the reasons for these and presents approaches for adjustments.
Figure 3: PM2.5LfU reference values from 19.04.-24.05.2021 to uncorrected SDS011 values
Source: own calculation and presentation.
16
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Figure 4: PM10 LfU reference values from 19.04.-24.05.2021 to uncorrected SDS011 values
Source: own calculation and presentation.
Table 3: Descriptive statistics - uncorrected vaules
PM2.5PM10
MAE 2.88 6.24
BIAS 1.40 0.61
RMSE 3.94 9.88
RE 18.35% 22.37%
Source: own calculation
and presentation.
4.2 Error correction method
The statistical evaluation shows that relevant deviations exist especially for PM10. The
values of PM2.5, on the other hand, can rather considered as robust. The visual inspection
(but also the value for MAE and so on) (see Figure 3 and 4) suggests that in the range
of low values the sensor box underestimates the PM exposure. On the other hand, in
the range of high values, the deviation from the reference value often seems to be quite
high. This means that the inaccuracies increases with the PM exposure. This could be a
17
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
starting point for a correction approach.
Two papers (see LUBW 2017, Benabbas et al. 2019) give further starting points.
The papers show that certain weather conditions can influence the results of the sensor
box. Especially humidity seems to have an influence. However, both studies could not
determine a systematic correlation between the measured deviations and air humidity.
LUBW lists further points. Thus, deviations in the measurements of sensors of different
batches are shown. Anyway, the aim of this paper is not to find a way to achieve 100
percent agreement of LfU values and sensor box values. This seems rather unachievable
in view of the results from the LUBW paper. Therefore, the aim is to work out whether
there are generally adaptation approaches and how efficient they are.
For this purpose, linear regression analyses are used in a first step. The above men-
tioned assumption that high deviations are associated with high measured values (and
vice versa) should be shown by a significant positive correlation between the MAE value
and the PM10 (respectively PM2.5) value measured by the sensor box. If this is the case,
the determined regression coefficients can be used to correct the sensor box values.
In addition to the PM10 (PM2.5) value as an influencing factor, the air humidity
must also be included. The study of LUBW suggests a positive influence of humidity
on the MAE values. Since other weather data were determined in addition to humidity
(temperature, air pressure), these can be included as control variables in the regression
analysis.
Table 4 and 5 show the correlation between the addressed factors. The values support
the assumption that the deviations between the values of the sensor box and LfU station
become stronger when the PM values are high (and vice versa). With regard to the
influence of the weather data, a relevant correlation is only shown for the humidity. This
is more pronounced with regard to the PM10 values. It is particularly evident with regard
to the BIAS indicator.
18
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Table 4: Correlation matrix for PM2.5
PM2.5MAE BIAS temp press hum
PM2.5- 0.62 -0.62 -0.07 -0.03 0.29
MAE 0.62 - -0.25 0.07 -0.03 -0.01
BIAS -0.62 -0.25 - 0.15 -0.01 -0.35
temp -0.07 0.07 0.15 - -0.20 -0.35
press -0.03 -0.03 -0.01 -0.20 - -0.12
hum 0.29 -0.00 -0.35 -0.35 -0.12 -
Note: temp = Air temperature, press = Air pressure,
hum = Air humidity; Source: own calculation and pre-
sentation.
Source: own calculation and presentation.
Table 5: Correlation matrix for PM10
PM10 MAE BIAS temp press hum
PM10 - 0.63 -0.73 -0.16 0.04 0.37
MAE 0.63 - -0.28 -0.00 -0.06 0.14
BIAS -0.73 -0.28 - 0.24 -0.11 -0.44
temp -0.16 -0.00 0.24 - -0.20 -0.35
press 0.04 -0.06 -0.11 -0.20 - -0.12
hum 0.37 0.14 -0.44 -0.35 -0.12 -
Note: temp = Air temperature, press = Air pressure,
hum = Air humidity; Source: own calculation and pre-
sentation.
Source: own calculation and presentation.
Considering the results of correlation analysis, the following two regression models
are tested:
(M1) MAEP Mi,n =b0 + P Mi,n b1 + humnb2 + un
and (M2) BI ASP Mi,n =b0 + P Mi,n b1 + humnb2 + un
With P M =measured PM value from sensor box, hum =humidity, i(2.5,10) and
n(1, ..., N )observations. Table 6 and 7 present the results.
19
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Table 6: Regression results for PM2.5
M1M2
n= 1466
Intercept ***2.96 ***5.88
P M2.5***0.28 ***-0.32
hum ***-0.02 ***-0.03
R20.43 0.42
Fstat ***552.6 ***530.7
Source: own calculation and pre-
sentation.
Table 7: Regression results for PM10
M1M2
n= 1466
Intercept **3.45 ***14.83
P M10 ***0.50 ***-0.62
hum ***-0.04 ***-0.10
R20.41 0.57
Fstat ***518.6 ***975.7
Source: own calculation and pre-
sentation.
In order to check the robustness of the results, various tests can be performed. These
are tests related to the efficiency of the calculated estimators (best linear unbiased esti-
mators). These are also tests that ask how much the calculated results depend on how
many data and which data (which time period) are considered. The bootstrap method
is used for this purpose.
Examination of the efficiency of the estimated parameters indicates problems. An
important requirement is that there must be a linear relationship between dependent
and independent variables. There is little empirical evidence for this. A transformation
of data could not eliminate this problem. As a result, the estimated parameters have
limited explanatory power.
Tests that examine structural breaks in the data show that there are some approaches
for adjustment: There is an indication of limited homogeneity with regard to a certain
range of PM-values. Clusters can also be identified with regard to weather data especi-
20
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
ally to the humidity level (as expected with regard to LUBW 2017). But the number of
observations shrinks very quickly when corresponding models are set up and as a con-
sequence, the robustness of the results decreases. It is as LUBW 2017 already shows:
there are influences (especially with regard to humidity), but the correlations are not
sufficiently systematic. Therefore, the results of the first regression models can be used,
taken into account, that the results are of limited reliability.
Beside the efficiency of the regression models, it is of interest how strong the value
of the estimation parameters varies with the amount of data. Two methods were used
to test this. First, bootstrap procedures were performed. Here, a subsample of the data
set is randomly produced. Subsequently, the regression analysis is carried out with this
subsample. This procedure is repeated a sufficient number of times. Here, with a number
of 2.000 repetitions, there is hardly any relevant deviation from the original estimation
result (see table 8).
Table 8: Example for bootstrap statistic for PM10 (M1)
Original Bias Std. error
Intercept 3.45 0.015 0.502
PM10 0.50 -0.001 0.029
hum -0.04 -0.000 0.007
Note: Bias = difference between the original esti-
mate and the bootstrap estimate; Std. error is
the standard error of the bootstrap estimate;
Source: own calculation and presentation.
Source: own calculation and presentation.
Second, the data set is increased by one observation at a time (starting with n= 1).
The idea is to determine how much data is needed to get useful results from the regression.
Usefulness can then be judged by the results of the regression for the entire data set. So,
figure 5 and 6 show how the regression parameters develop. It turns out that reasonably
useful results can be derived only after about the first 500 observations. This is an
important result: It gives an indication of how many observations or how much time is
needed to calculate reasonably robust correction values.
21
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Figure 5: Regression results for M1 (PM10) with stepwise data set enlargement
Source: own calculation and presentation.
Figure 6: Development of the average regression coefficients for M1 (PM10 )
Source: own calculation and presentation.
22
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
In summary, the correction approach is as follows: It is possible to estimate LfU
values by using the regression results and measured value of the sensor box (for P Mi
and hum). Finally the comparison of the real and the estimated LfU ?values show the
efficiency of the method. Therefore, the correction algorithm is (for M1): Because of
MAEP Mi,n =|Lf UP Mi,n P Mi,n|, for P M2.5if
LfUP M2.5,n > P M2.5,n then Lf U?
P M 2.5,n = 2.96 + 1.28P M2.5,n 0.02humn
otherwise LfU ?
P M2.5,n =2.96 + 0.72P M2.5,n + 0.02humn
furthermore for P M10, if
LfUP M10 ,n > P M10,n then Lf U ?
P M 10,n = 3.45 + 1.5P M10,n 0.04humn
and otherwise LfU ?
P M10,n =3.45 + 0.5P M10,n + 0.04humn
4.3 Results after correction
Figure 7 and 8 presents the visual comparison of the corrected sensor box data and the
LfU data. Table 9 shows corresponding descriptive results using the formula for MAE,
BIAS, RSME and RE.
The results show that adjusting the values using the correction approach significantly
reduces the discrepancies between the LfU and sensor box data. This result is stronger
in the case of PM2.5values. While before the adjustment only about 20% of the sensor
box data show small deviations (< 15%), after the correction more then 50% (25%) of
the values for PM2.5(PM19) deviate by less than 15%.
With regard to the key figure MAE, the average absolute deviation between the LfU
and sensor box data could be reduced by 50% (40%) in the case of PM2.5(PM10).
The visualization also illustrates that the adjustment works very well, especially in
the data range that shows low fluctuations or low PM values. This effect is particularly
clear for PM10. The sensor box overemphasis high PM values.
23
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Figure 7: PM2.5LfU reference values and corrected Sensor Box (SDS011) values
Source: own calculation and presentation.
Figure 8: PM10 LfU reference values and corrected Sensor Box (SDS011) values
Source: own calculation and presentation.
24
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Table 9: Descriptive statistics
uncorrected corrected
PM2.5PM10 PM2.5PM10
MAE 2.88 6.24 1.51 3.88
BIAS 1.40 0.61 -0.03 1.00
RMSE 3.94 9.88 2.08 5.86
RE 18.35% 22.37% 50.64% 27.01%
Source: own calculation and presentation.
5 Conclusions
Are the particulate matter data measured using low-cost sensor technology comparable
with those of the LfU? Can data adjustments contribute to comparability? The results
show that the low-cost sensor is quite capable of imaging particulate matter data with
some accuracy. Especially for PM25, useful results are obtained in sum. For PM10 there
are clear deviations. The deviations become stronger when high concentrations of PM
are detected. Here, the sensor shows weaknesses.
But: With statistical methods, this weakness can be dampened. The analysis has
shown that efficient correction approaches can be found. They contribute to the fact
that the data of the sensor box can be adapted to those of the LfU station. It must be
taken into account that the correction approaches require a sufficient number of sample
measurements and reference data at the same time. I.e. an adjustment process is possible,
but needs access to official measuring stations. To what extent the correction approaches
can be transferred could not be determined. The investigation only referred to one specific
sensor. However, looking at other studies (LUBW 2017), it can be assumed that the
adjustment should be checked with regard to its effect if it is applied to other sensors of
type SDS011.
In sum, it can be stated that the low-cost sensor can be used very well as a kind of
warning system. As long as PM values are low, the sensor measures sufficiently accura-
tely. As soon as the PM values increase, the sensor produces significant deviations and
fluctuations. In this case, it is no longer possible to say exactly how the PM values turn
out. However, it can be said that there is a fairly high probability of high PM concen-
tration. This warning can be taken into account by the public. After all, this result is
interesting. Combined with the fact that a favorable sensor is considered here, this results
in a sufficiently good instrument for the public to monitor PM levels at arbitrary locati-
25
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
ons and to detect high levels of pollution. This is the prerequisite for further discussion,
for example in the field of traffic planning or urban greening.
Next studies could develop further correction approaches. They could also investigate
the reliability and transferability of the correction approaches through larger test series.
26
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Bibliography
ACS (Arbeitsgruppe Citizen Science) (2021): Was ist Citizen Science?; URL:
https://www.citizen-science.at/eintauchen/was-ist-citizen-science; [last
visit: 03.08.2021].
ÄrzteZeitung (2021): So sah die Feinstaub-Belastung im Corona-Jahr 2020 aus.
Veröffentlicht am 18.02.2021; URL: https://www.aerztezeitung.de/Panorama/
So-sah-die-Feinstaub-Belastung-im-Corona-Jahr-2020-aus-417179.html; [last
visit: 01.07.2021].
Benabbas, A., Geißelbrecht, M., Nikol, G. M., Mahr, L., Nähr, D., Steuer, S., Wiesemann,
G., Müller, T., Nicklas, D., and Wieland, T. (2019): Measure particulate matter by
yourself: data-quality monitoring in a citizen science project, J. Sens. Sens. Syst., 8,
317–328; URL: https://doi.org/10.5194/jsss-8-317-2019; [last visit: 26.01.2022].
Eickelpasch, D. and Eickelpasch, G. (2004): Feststellung und Bewertung von Immisio-
nen. Leitfaden zur Immissionsüberwachung in Deutschland. Umweltforschungsplan des
Bundesministeriums für Umwelt, Naturschutz und Reaktorsicherheit, Texte 07/04, 3.
Auflage, Verlag Umweltbundsamt (Hrsg.); URL: https://www.umweltbundesamt.de/
publikationen/feststellung-bewertung-von-immissionen; [last visit: 01.07.2021].
EEA (European Environment Agency) (2017): Air Quality in Europe – 2017 re-
port. No 13/2017. Publications Office oft he European Union, Luxembourg;
URL: https://www.eea.europa.eu/publications/air-quality-in-europe-2017;
[last visit: 01.07.2021].
ELAW (2002): Federal Immissions Control Act (BlmSchG); URL: https://www.elaw.
org/system/files/de.air.noise.act.eng.pdf; [last visit: 22.01.2022].
EUR-Lex (2005): Thematische Strategie zur Luftreinhaltung, Rechtsakt, Mitteilung an
den Rat und das Europäische Parlament vom 21.09.2005; URL: https://eur-lex.
europa.eu/legal-content/DE/TXT/?uri=LEGISSUM:l28159; [last visit: 28.06.2021].
EUR-Lex (2008): Luftqualität und saubere Luft für Europa. Richtlinie
2008/50/EG des Europäischen Parlaments und des Rates vom 21.05.2008;
URL: https://eur-lex.europa.eu/legal-content/DE/TXT/HTML/?uri=CELEX:
02008L0050-20150918&from=EN; [last visit: 30.06.2021].
27
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
Grimm (2011): Environmental Dust Monitors. Approved & Unrivaled; URL: https://
mluk.brandenburg.de/i/download/Staubmessger\%c3\%a4t\%20EDM180.pdf; [last
visit: 04.08.2021].
Helbig, A., Baumüller, J. and Kerschgens, M.J. (Hrsg.) (2013): Stadtklima und Luftrein-
haltung, Springer Verlag, Berlin, 2. Auflage.
Kotsev, A., Schade, S., Craglia, M., Gerboles, M., Spinelle, L. and Signorini, M. (2016):
Next generation air quality platform: openess and interoperability for the internet oft
hings, in: Sensors 16 (3): 403; URL: https://www.mdpi.com/1424-8220/16/3/403/
htm; [last visit: 29.06.2021].
LfU (Landesamt für Umwelt Brandenburg) (2021a): Luftqualität. Luftgü-
temessnetz Brandenburg. Ziel-, Richt- und Grenzwerte; URL: https:
//lfu.brandenburg.de/lfu/de/aufgaben/immissionsschutz/luftqualitaet/
luftguetemessnetz-brandenburg/ziel-\%2C-richt-und-grenzwerte/
~mais2redc245804de; [01.07.2021].
LfU (2021b): Luftqualität. Luftgütemessnetz Brandenburg. Ziel-, Richt- und Grenzwer-
te. Inhaltsstoffe im PM10; URL: https://lfu.brandenburg.de/lfu/de/aufgaben/
immissionsschutz/luftqualitaet/luftguetemessnetz-brandenburg/ziel-\
%2C-richt-und-grenzwerte/~mais2redc245825de; [last visit: 04.07.2021].
LfU (2021c): Luftqualität in Brandenburg. Jahresbericht 2019. Ergebnisse der
Emmissionsmessungen; URL https://lfu.brandenburg.de/sixcms/media.php/9/
Luftqualitaet_BB_2019.pdf; [last visit: 04.08.2021].
LfU (2021d): Luftgütedaten Brandenburg. Aktuelle Luftgütedaten. Nutzungshinwei-
se; URL: https://luftdaten.brandenburg.de/home/-/bereich/aktuell; [last visit:
04.08.2021].
LUBW (Landesamt für Umwelt, Messungen und Naturschutz Baden-Würtemberg)
(2005): Prüfung des Schwebstaubmessgerätes Environmental Dust Monitor – Model
180 der Firma Grimm Aerosoltechnik auf gleichwertige Ergebnisse wie mir der Refe-
renzmethode der Europäischen Norm DIN EN 12341 bei der Messung von Schweb-
staub der Fraktion PM10; URL: http://wp12311247.server-he.de/assets/grimm_
aerosol_technik_pruefbericht-pm10-grimm_.pdf; [last visit: 04.08.2021].
28
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
LUBW (2017): Messungen mit dem Feinstaubsensor SDS011, Ein Vergleich mit ei-
nem eignungsgeprüften Feinstaubanalysator. https://pudi.lubw.de/detailseite/
-/publication/90536; [last visit: 03.08.2021].
Leibnitz (Leibnitz Gemeinschaft) (2021): Citizen Science; URL: https:
//www.leibniz-gemeinschaft.de/forschung/citizen-science.html; [last vi-
sit: 03.08.2021].
MacDonnell, M., Raymond, M., Wyker, D., Finster, M., Chang, Y., Raymond, T.,
Temple, B., Scofield, M. and Vallano, D. (2013): Mobile Sensors and Applica-
tions for Air Pollutants, in: US Environmental Protection Agency, Washington
D.C; URL: https://cfpub.epa.gov/si/si_public_record_report.cfm?Lab=NERL&
dirEntryId=273979; [last visit: 29.06.2021].
MLUK (Ministerium für Landwirtschaft, Umwelt und Klimaschutz in Bran-
denburg) (2020): Jahresbericht zur Luftqualität in Brandenburg 2020. Jah-
reskurzbericht. https://luftdaten.brandenburg.de/documents/335843/13113341/
Jahreskurzbericht+LQ+2020/2c85a23f-6539-1b6b-69da-e07004080f52; [last visit:
23.07.2021].
MLUK (2021a): Feinstaub. https://mluk.brandenburg.de/mluk/de/umwelt/
immissionsschutz/luft/luftqualitaet/feinstaub/; [last visit: 28.06.2021].
MLUK (2021b): Luftgütedaten Brandenburg. Überwachung der Luftqualität. Wo und wie
wird gemessen; URL: https://luftdaten.brandenburg.de/wo-wie-wird-gemessen;
[last visit: 30.06.2021].
MLUK (2021c): Luftgütedaten Brandenburg. Luftgütedaten. Messstationen;
URL: https://luftdaten.brandenburg.de/home/-/bereich/messstationen;
[07.07.2021].
MLUK (2021d): Luftqualität; URL: https://lfu.brandenburg.de/lfu/de/aufgaben/
immissionsschutz/luftqualitaet/; [07.07.2021].
MLUK (2021e): Feinstaub/ Partikel; URL: https://lfu.brandenburg.de/lfu/de/
aufgaben/immissionsschutz/luftqualitaet/luftguetemessnetz-brandenburg/
feinstaubpartikel/~mais2redc257180de; [last visit: 07.07.2021].
MLUK (MLUK 2021f): Luftqualität. Luftgütemessnetz Brandenburg. Feinstaub/ Par-
tikel; URL https://lfu.brandenburg.de/lfu/de/aufgaben/immissionsschutz/
29
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
luftqualitaet/luftguetemessnetz-brandenburg/feinstaubpartikel/; [last visit:
30.06.2021].
Snyder, E.G., Watkins, T.H., Solomon, P.A., Thoma, E.D., Williams, R.W., Hagler, G.S.,
Shelow, D., Hindin, D.A., Kilaru, V.J. and Preuss, P.W. (2013): The changing para-
digm of air pollution monitoring, in: Environmental Science & Technology,2013, 47,
11369 – 11377; URL: https://pubs.acs.org/doi/10.1021/es4022602#; [last visit:
30.06.2021].
TTN (The Things Network) (2021): Building a global open LoRaWAN network; URL:
https://www.thethingsnetwork.org/; [last visit: 20.07.2021].
UBA (Umweltbundesamt) (2009): Feinstaubbelastung in Deutschland; URL:
https://www.umweltbundesamt.de/sites/default/files/medien/publikation/
long/3565.pdf; [last visit: 30.04.2021].
UBA (2013): Das Luftmessnetz des Umweltbundesamtes; URL: https://www.
umweltbundesamt.de/sites/default/files/medien/378/publikationen/das_
luftmess/netz_des_umweltbundesamtes_bf_0.pdf; [last visit: 29.06.2021].
UBA (2017): Abfallwirtschaft. Abfallwirtschaft in Deutschland; URL: https://www.
umweltbundesamt.de/themen/abfall-ressourcen/abfallwirtschaft; [05.08.2021].
UBA (2018): Fragen und Antworten: Ultrafeine Partikel; URL: https://www.
umweltbundesamt.de/themen/luft/luftschadstoffe-im-ueberblick/feinstaub/
fragen-antworten-ultrafeine-partikel; [last visit: 30.04.2021].
UBA (2019a): Beurteilung und Kontrolle der Luftreinhal-
tung in der EU; URL: https://www.umweltbundesamt.de/
themen/luft/regelungen-strategien/luftreinhaltung-in-der-eu#
beurteilung-und-kontrolle-der-luftreinhaltung-in-der-eu; [last visit:
04.08.2021].
UBA (2019b): Luftmessnetz: Wo und wie wird gemessen?; URL: https:
//www.umweltbundesamt.de/themen/luft/messenbeobachtenueberwachen/
luftmessnetz-wo-wie-wird-gemessen#gesetzliche-grundlagen; [29.06.2021].
UBA (2020): Feinstaub; URL: https://www.umweltbundesamt.de/themen/luft/
luftschadstoffe-im-ueberblick/feinstaub; [last visit: 03.08.2021].
30
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
UBA (2021a): Aktuelle Luftdaten. Stationen; URL: https://www.umweltbundesamt.de/
daten/luft/luftdaten/stationen/; [last visit: 29.06.2021].
UBA (2021b): Entwicklung der Luftqualität; URL: https://www.umweltbundesamt.de/
themen/luft/luftqualitaet#luftdaten; [last visit: 30.06.2021].
UBA (2021c): Luftqualität 2020. Vorläufige Auswertung; URL: https://www.
umweltbundesamt.de/publikationen/luftqualitaet-2020; [last visit: 30.06.2021].
UBA (2021d): Anerkannte Messgeräte und Messverfahren; URL: https://https://
www.umweltbundesamt.de/themen/luft/messenbeobachtenueberwachen; [last visit:
01.07.2021].
VM (Ministerium für Verkehr Baden-Württemberg) (2021): Luftreinhaltepläne.
Maßnahmen; URL: https://vm.baden-wuerttemberg.de/de/mensch-umwelt/
luftreinhaltung/luftreinhalteplanung/luftreinhalteplaene/; [last visit:
07.07.2021].
WHO (World Health Organization) (2005): Air Quality Guidelines. Global Upda-
te 2005; URL: https://www.euro.who.int/_data/assets/pdf_file/0005/78638/
E90038.pdf; [last visit: 01.07.2021].
WHO (2018): Household air pollution and health; URL: https://www.who.int/
news-room/fact-sheets/detail/household-air-pollution-and-health_2; [last
visit: 30.04.2021].
WHO (2021a): Air pollution Overview; URL: https://www.who.int/health-topics/
air-pollution#tab=tab_1; [last visit: 30.04.2021].
WHO (2021b): Air pollution Ambient air pollution; URL: https://www.who.int/
health-topics/air-pollution#tab=tab_2; [last visit: 30.04.2021].
Williams, R., Kilaru, V., Synder, E., Kaufman, A., Dye, T., Rutter, A., Russel, A. and
Hafner, H. (2014): Air sensor guidebook, in: National Exposure Research Laborato-
ry, Office of Research and Development, US Environmental Protection Agency, NC,
USA; URL: https://cfpub.epa.gov/si/si_public_record_report.cfm?Lab=NERL&
dirEntryId=277996&simpleSearch=1&searchAll=air+sensor+guidebook; [last vi-
sit: 29.06.2021].
Wu, X., Rachel, C., Nethery, M., Sabath, B., Braun, D. and Dominici, F. (2020): Ex-
posure to air pollution and COVID-19 mortality in the United States: A nationwide
31
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
cross-sectional study, in: Sciences Advances, Vol. 6 no. 45, eabd4049; URL: https:
//advances.sciencemag.org/content/6/45/eabd4049; [last visit: 16.07.2021].
32
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 22 February 2022 doi:10.20944/preprints202202.0283.v1
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.