An Approach to Improve Distribution Efficiency for the Deployment of Air Quality Monitoring Devices

Abstract

Effective air quality monitoring is crucial for understanding and mitigating the adverse impacts of pollution. This research confronts the challenges of obtaining precise data and identifying sources of pollution. It achieves this by introducing an effective algorithm for the deployment of air quality monitoring devices (AQMDs), thereby enhancing the efficiency of their distribution. Our study initiates by verifying current setups of AQMD, which have a limited number of pre-installed devices. This algorithm considers the spatial distribution of pollution sources to minimize the capital and operational costs associated with AQMD installation. It utilizes a dataset that spans 11 months and covers 759.13 km² of Delhi and 75 km² of Durgapur. The proposed algorithm demonstrates its efficacy by achieving an accuracy rate of 90% − 95% in predicting air quality. By strategically selecting monitoring locations based on the distribution of pre- installed AQMDs and existing pollution sources, the algorithm significantly reduces unnecessary costs while maximizing data coverage for comprehensive testing and analysis. This research contributes to optimizing air quality monitoring networks, facilitating better decision-making for pollution control and resource allocation. The outcomes bear significance for urban planners, policymakers, and environmental researchers who are in search of cost-effective solutions to address air pollution challenges in affected areas.

Full text

An Approach to Improve Distribution Eciency for
the Deployment of Air Quality Monitoring Devices
Pritisha Sarkar
National Institute of Technology Durgapur
Munsi Yusuf Alam
Budge Budge Institute of Technology
Mousumi Saha
BBIT: Budge Budge Institute of Technology
Arup Roy
BBIT: Budge Budge Institute of Technology
Saurav Mallik
Harvard University https://orcid.org/0000-0003-4107-6784
Ujjwal Maulik
Jadavpur University
Research Article
Keywords: Outdoor air pollution, Air quality monitoring device, Ideal placement, Spatial-temporal factor
Posted Date: August 29th, 2024
DOI: https://doi.org/10.21203/rs.3.rs-3897355/v1
License:   This work is licensed under a Creative Commons Attribution 4.0 International License. 
Read Full License

An Approach to Improve Distribution Efficiency for the
Deployment of Air Quality Monitoring Devices
Pritisha Sarkara,Munsi Yusuf Alamb,Mousumi Sahaa,Arup Royc,Saurav Mallikd,∗and
Ujjwal Maulike
aNational Institute of Technology, Durgapur, India,
bBudge Budge Institute of Technology, Kolkata, India,
aNational Institute of Technology, Durgapur, India,
bBudge Budge Institute of Technology, Kolkata, India,
cDepartment of Environmental Health, Harvard T H Chan School of Public Health, Boston, MA 02115, USA,
cDepartment of Computer Science & Engineering, Jadavpur University, Kolkata, West Bengal, India,
A R T I C L E I N F O
Keywords:
Outdoor air pollution
Air quality monitoring device
Ideal placement
Spatial-temporal factor
A B S T R A C T
Effective air quality monitoring is crucial for understanding and mitigating the adverse impacts
of pollution. This research confronts the challenges of obtaining precise data and identifying
sources of pollution. It achieves this by introducing an effective algorithm for the deployment of
air quality monitoring devices (AQMDs), thereby enhancing the efficiency of their distribution.
Our study initiates by verifying current setups of AQMD, which have a limited number of
pre-installed devices. This algorithm considers the spatial distribution of pollution sources to
minimize the capital and operational costs associated with AQMD installation. It utilizes a
dataset that spans 11 months and covers 759.13 km²of Delhi and 75 km²of Durgapur. The
proposed algorithm demonstrates its efficacy by achieving an accuracy rate of 90% −95% in
predicting air quality. By strategically selecting monitoring locations based on the distribution
of pre- installed AQMDs and existing pollution sources, the algorithm significantly reduces
unnecessary costs while maximizing data coverage for comprehensive testing and analysis. This
research contributes to optimizing air quality monitoring networks, facilitating better decision-
making for pollution control and resource allocation. The outcomes bear significance for urban
planners, policymakers, and environmental researchers who are in search of cost-effective
solutions to address air pollution challenges in affected areas.
1. Introduction
Continuous air quality monitoring helps to identify and control poor air quality, leading to evidence-based policies
for a healthier future. Real-time information on pollutant concentrations, particularly PM2.5and PM10, is vital for
protecting human health from the harmful effects of air pollution Giannadaki (2016) et al., Fann (2011) et al., Geng
(2021) et al. Our study focuses solely on these two pollutants due to their significant health risks, including heart
disease, respiratory distress, and cerebrovascular mortality 1due to their heightened risk, particularly to respiratory
diseases. In India, for instance2, the coverage of air monitoring stations is estimated to be only 3% according to the
Central Pollution Control Board (CPCB), with a mere 133 stations present out of the required 4000. Even in New Delhi,
often referred to as the "Pollution Capital" of the world, there are only 33 monitoring stations to cover the expansive
area of 1484 km2. These statements illustrate the significant gaps in air quality monitoring infrastructure and the need
for further expansion and improvement.
However, the high costs associated with establishing and maintaining air quality monitoring infrastructure present a
persistent challenge for many developing countries. Setting up regulatory-grade air quality monitoring stations requires
substantial installation and maintenance costs. The construction cost alone amounts to approximately 200,000 USD,
∗Corresponding author
ps.19cs1110@phd.nitdgp.ac.in (P. Sarkar); munsiyusufalam@bbit.edu.in (M.Y. Alam); mousumi.saha@cse.nitdgp.ac.in
(M. Saha); aruproy.cse@gmail.com (A. Roy); sauravmtech2@gmail.com, smallik@hsph.harvard.edu (S. Mallik);
ujjwal.maulik@jadavpuruniversity.in (U. Maulik)
ORCID (s):
1https://www.epa.gov/pm-pollution/health-and-environmental-effects-particulate-matter-pm
2https://shorturl.at/iyEFM
Page 1 of 23

Optimizing Air Quality Monitoring Device Deployment
with an annual maintenance cost of 30,000 USD Zheng, Liu and Hsieh (2013). Optimal placement of air quality
monitoring devices (AQMD) at appropriate locations poses a considerable challenge as air pollution exhibits a direct
relationship with spatial and temporal factors.
The importance of considering the placement of new devices is not merely a geometric problem. When determining
the locations for new stations 𝑥in an area with existing ones 𝑦, it is not sufficient to find equidistant positions for the
𝑥in relation to the 𝑦.This is because the air quality in a metropolis is affected by numerous factors, such as weather
conditions, demographic profile, and traffic patterns, leading to non-linear spatial variations Zheng et al. (2013). For
instance, consider two positions, A and B, located 6 kilometers apart. If their demographic profiles and traffic patterns
are similar, they may experience similar air quality conditions. However, another position, C, located 2 kilometers
away from A, might exhibit distinctly different air quality due to its proximity to an industrial hub. Therefore, when
determining the placement of new devices, it is crucial to take into account the complex interplay of various factors that
influence air quality variations rather than relying solely on geometric considerations or equidistant positioning. In our
study area, Durgapur, four AQMDs were placed. Despite their close proximity of approximately 2.7 km, a significant
disparity in reported Air Quality Indices was observed between Device-1 and Device-4 locations on 22/12/2019,
highlighting location-dependent variations in urban air quality as depicted in Figure 1.
Figure 1: Air quality reports from 4 stations in our targeted city on 27/10/2019 and 22/12/2019
However, another interesting insight revealed by Sharma (2021) et al., Pramanik (2023) et al., in their innovative
framework, called AQuaMoHo obtain up-to-date air quality information by utilizing only temperature and humidity
measurements. Furthermore, in another research Sarkar (2022) et al., we focused on understanding the spatial and
temporal nature of air pollution by employing multiple linear regression and Pearson correlation models to analyze
its correlation with the Air Quality Index (AQI). The study specifically investigated air pollution levels in four cities:
Chennai, Delhi, Hyderabad, and Durgapur. In research, Lu (2019) et al., Leung (2015) et al., Zanobetti (2002) et al., Liu
(2020) et al., the experiments verified that air quality significantly correlated with temperature and humidity. According
to the survey, it has been established that particulate matter (PM) affects temperature and humidity. Additionally, PM
significantly influences the overall AQI, indicating a direct connection between AQI and temperature humidity.
The high cost of air quality monitoring devices necessitates a strategic approach to ensure their placement where
they are most needed, rather than randomly. In this particular paper, we sought to validate the effectiveness of the
heuristic problem-solving methodology by placing affordable lab-scale prototype devices.
In general, the device placement problem has been conducted in the context of other domains like Baume, Gebhardt,
Gebhardt, Heuvelink and Pilz (2011); Wu and Bocquet (2011). However, to the best of our knowledge, a few works
deal with the placement of air quality monitoring devices. In the papers by Dhillon and Chakrabarty (2003),Joshi
and Boyd (2008), the authors discuss sensor placement for environmental monitoring. In addressing the needs of
native populations, Sun, Li, Lam and Leslie (2019) proposed an optimal sensor placement problem for air pollution
monitoring but lacked field information specific to Cambridge city, limiting its applicability in developing countries.
Hsieh et al. Hsieh, Lin and Zheng (2015) developed a parameterized air quality prediction model using graph theory,
addressing correlation challenges among spatio-temporal nodes. They used a semi-supervised inference model for
prediction and entropy minimization for optimal station placement. Coverage is a well-researched issue in sensor
networks, as explored in Kumar, Lai and Balogh (2004),Chen, Li, He, He, Gu and Sun (2013). The deployment
approach suggested in Alexander and Imberger (2009), Shaban, Kadri and Rezk (2016) directs the development of
monitoring systems.
A review of the above research work reveals that, firstly, existing methods rely on various weather-related information
Page 2 of 23

Optimizing Air Quality Monitoring Device Deployment
(temperature, humidity, cloud rainfall, wind speed, and pressure, etc.) and demographic profiles that are often difficult
to obtain. A thorough observation shows that out of all, population information may partially be available in some of
the developing regions. In contrast, our work utilizes only two temporal (temperature and humidity) and spatial data
variables (like natural land, and human-made, roadways). However, traffic-related information is almost not present in
most of the developing sub-urban cities. Therefore, some important parameters used in existing algorithms may not be
applicable in many researched regions. Second, it is important to note that environmental factors, urban dynamics, and
data availability may vary significantly across locations, potentially affecting the performance of the proposed models.
Therefore, it is crucial to validate and adapt the framework to suit the specific characteristics and context of each city
or region. The authors did not validate their work in different types of regions, whereas in our study, we apply our
method to two different demographic regions.
Problem statement
Understanding these challenges from existing research work, we define the primary goal of our work as follows. Given
an area with symmetric grid A of a place with the percentage of spatial features (like natural land, human-made,
road networks, water bodies, parks) x𝑖and meteorological features (temperature, humidity) y𝑖and the number of the
pre-installed device of that place is k, (i) Is it possible to check the ideal location of preinstalled device, which can
fulfill the maximum area coverage in that particular place? (ii) If we have a new device, where should we place so that
it can fulfill the maximum area coverage criteria?
The major contributions of the proposed work are summarized as follows:
•Finding the correlation between AQI and meteorological features: We target a fast-growing industrialized
suburban city for our study. After installing our AQMD at four distinct locations, we evaluate population, climate
parameters, and pollutants. Using geospatial techniques, we derive spatial distributions and Pearson correlations
that help find relationships between AQI and meteorological properties.
•Deploying devices to meet data sets required for our work: We made two different setups to collect the
required data. Firstly, a sensor-based portable air quality monitoring device (PAQMD) has been developed by us
to monitor air pollutant concentration and meteorological parameters along the corresponding latitude-longitude
in a granular way. Secondly, we also installed air quality monitoring devices at four distinct locations in our
study area. AQMDs capture logs from sensors such as nitrogen dioxide gas sensors, laser dust modules, electro-
chemical carbon monoxide sensors, and low voltage temperature humidity sensors. The collected time series
data is embedded with the back-end server for online processing of the data.
•Determining easily accessible feature sets for optimal device deployment: Our proposed algorithm adopts
nominal features that are easily accessible with accepted effects on the AQI of any location. Then we go through
the data analysis and find the most responsible features directly related to the increase or decrease in pollution
levels. This efficient selection of features helps us build a simple model that can sense the local state of the
features with a minimum number of devices.
•Optimal device placement solution: We formulate two objectives for our optimal device placement problem
with minimal field information. For the needs of the local population, the first objective is to check whether the
best of the existing devices can meet the maximum area coverage. The second objective is to set up a new station
at the target location. It seems reasonable and practical. After all, the selection of monitoring stations can be
more effective, and we want to accurately estimate the AQI value of new locations with very few features. We
have conducted a case study in the cities of Delhi and Durgapur, India, to assess the feasibility of the formulation
and evaluate the effectiveness of the proposed approach.
The paper is structured as follows; Section 2outlines the workflow of our proposed research, which involves
developing a model to optimize device placement. Section 3showcases the architecture of our static AQMD and
Portable AQMD devices and validates the datasets used in our study. We analyze internal and publicly available datasets
for our two targeted cities in India, Durgapur, and Delhi, highlighting the impact of spatio-temporal features on AQI.
Section 4delves into the methodology, while Section 5 focuses on evaluating our 1st targeted city Durgapur,India and
Section 6depicts our proposed approach in our 2nd targeted city, Delhi,India. Section 7presents a thorough discussion
of related works, focusing on the development and placement of environmental monitoring devices, the significance of
spatio-temporal features in device placement, and different techniques used in device placement for various research
purposes. Finally, in Section 8, we conclude the paper.
Page 3 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 2: System work-flow of Device placement in an ideal way
2. System overview
This section provides a broad overview of the framework proposed to answer our problem statement, as shown
in Figure 2. We first collect spatial, meteorological, and pollutant data over the past three years to find an algorithm
for optimal AQMD placement. Using the Google Maps API, we extract the percentage of spatial features such as
natural land, human-made roads, and other POIs (parks, water bodies, medical stores, and educational institutions).
Meteorological(temporal) features such as temperature and humidity are collected using our Portable Air Quality
Monitoring Device (PAQMD). We obtain pollutant data PM2.5, PM10 from our installed AQMD and CPCB. During
the data pre-processing step, we take various methods to prepare the pollutant, meteorological and spatial data for
further analysis. It includes several necessary steps, such as removing null values, normalizing data, removing negative
values and performing other operations. Removing negative values allows us to focus only on relevant and positive
data points which increases the accuracy of subsequent analysis. Overall, the data pre-processing step ensures that the
spatial, meteorological, and pollutant data are clean, standardized and suitable for further investigation, enabling us to
draw accurate insights and make decisions based on the processed data. The evaluation process involves several steps.
Firstly, a clustering method is used on the spatial data to determine whether the targeted grids belong to a similar cluster
or not. If there is a similar spatial structure among all targeted grids, a statistical approach is employed to optimize the
similar clustered grid using pollutant data. Based on our AQI variation, it can be determined whether the pre-installed
grid is optimally placed or not. Consequently, redundant grids can be removed. Additionally, variance computation
using temperature and humidity attributes is performed to determine precise geolocation for installing new AQMD in
unoccupied grids. This constitutes the outcome of the evaluation process.
3. Data acquisition
Data is the most evident and first thing we need in order to comprehend, conduct analysis, and interpret it according
to our research purpose. Initially, we installed air quality monitoring evice (AQMD) in four distinct locations within
Durgapur to collect pollutant data. Subsequently, we designed a Portable Air Quality Monitoring Device (PAQMD),
as illustrated in Figure 3, to capture meteorological data such as temperature and humidity. To gather pollutant data for
Delhi, on our second targeted city, we utilized a publicly accessible website. We obtained spatial data for both cities
using the Google Maps API.
•Lab-scale portable air quality monitoring device (PAQMD) -
We developed a Portable Air Quality Monitoring Device (PAQMD), as depicted in Figure 3) that utilizes sensors
to measure the concentration of pollutants in the surrounding environment. The PAQMD incorporates various
sensor types, such as the Grove multichannel gas sensor for detecting pollutants like CO, CO2, NO2as well
as the Dust sensor (SKU: SEN 0177) for PM10, and PM2.5measurements. Additionally, it includes sensors for
measuring meteorological parameters such as temperature and relative humidity using the Grove temperature
and humidity sensor, as well as GPS sensors for capturing latitude-longitude data of the corresponding location.
Power banks and Arduino are utilized to power the device and run the system, respectively, ensuring seamless
operation while the device is in motion. With our sensor-based PAQMD, we were able to collect meteorological
Page 4 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 3: System architecture of our portable device(PAQMD), right upper corner image of our portable
environmental monitoring Device.
Figure 4: Meteorological Data(temperature, humidity) collection route map, yellow line defines the route of
meteorological data collection for only selected grid our portable device
and GPS data, as illustrated in Figure 4, which denotes the temporal data acquisition path, and Figure 3(left-side
image) shows the layered design of our system, while the right-upper image is the glimpse of the Portable Air
Quality Monitoring Device, which was also used for collecting temperature and humidity data.
•Regulatory grade for pollutant data -
We previously implemented a 4x4 grid system where our four AQMDs were symmetrically placed, as shown
in Figure 5A. Data plots generated from these devices revealed significant hourly variations in pollutant
concentrations. Since October 2019, we have collected approximately 8.5 million data samples from each
grid station. The four deployed stations can simultaneously measure pollutants such as PM2.5, PM10, CO, and
NO2. We focused on PM2.5and PM10 due to their adverse effects on heart disease, respiratory distress, and
cerebrovascular death 3. We collected data from our installed device in Durgapur for the period from October
2019 to January 2020. The overall sample size was approximately over 10,000 data samples, which after pre-
processing, was reduced to approximately 2,000 data samples at any given time for our experimental study. We
utilize the Delhi Central Pollution Control Board (CPCB) to obtain pollutant data for Delhi 4.
•Google map API for spatial data -
We utilized the Google Map API and the ’mapImaging’ approach to extract Geographic Information (spatial
data) for Durgapur and Delhi. We extracted several features, such as human made, natural land, local ways,
highways, parks, education, medical, and water bodies. However, we noticed that the mobility patterns of
education and parks were nearly identical, as people tend to visit these places at specific times of the day.
Similarly, water bodies and medical facilities had almost the same mobility pattern. The pollution levels in
these places may not always be high, which is why we represented these four features as percentages of category
as ’other POIs’ (Points of Interest).
3https://shorturl.at/hszI1
4https://shorturl.at/uzHZ0
Page 5 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 5: A. Map view of our targeted place, black marks indicate the pre-installed device’s location. B. Our
pre-installed static Air quality monitoring device. C. Deployment of that device in rooftop
Figure 6: Comparative result of the government-provided CPCB device and our Portable Air quality monitoring device data
Validation is required before pre-processing to ensure that data flowing from a network is correct and reliable.
We used two separate seasonal data sets for validation to validate the static air quality monitoring device: July and
December, 2021. To verify, our experimental static device is placed near a device given by the CPCB. There are
no significant discrepancies in the data from both devices. When the CPCB shows moderate air quality, our AQMD
similarly shows moderate pollution. When we tested, both device data concentrations ranged from 15 to 65 µg/m3.
When the AQI of the AQMS data was moderate, the CPCB data was also moderate. We also validate our Portable Air
quality monitoring device data with CPCB. Figure 6shows that 90% times a week, the pollutant concentration is an
almost similar range, but 10% time it has maximum pollutant variation is 50-80 µg/m3
3.1. Choice of features
Our research has highlighted the importance of the relationship between Spatio-meteorological features and Air
Quality, as evidenced by our previous work Sarkar (2022). During the data pre-processing stage, we thoroughly
examined the association between Air Quality Index (AQI) and Spatio-meteorological features. Our analysis revealed
that meteorological data exhibited a stronger correlation with air quality compared to spatial data. However, due
to practical constraints such as the high cost and maintenance associated with densely installing sensors to capture
temperature and humidity in fine-grained spatial-temporal scales, it becomes impractical to rely solely on these features.
Additionally, temperature and humidity exhibit significant variations on a seasonal basis, further complicating the
sensor installation process. To overcome these challenges, we initially incorporated spatial features like natural land,
human-made structures, and roadways into our heuristic algorithm, which is far better than the existing random method.
Acknowledging that while meteorological features have a higher correlation, spatial features also contribute valuable
insights. We observed correlations between spatial data and air quality through the Pearson correlation in Figure 7,
further supporting our decision to include spatial features in our work. Using one month of AQI data, as well as
Page 6 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 7: Pearson Correlation (r-value) coefficient of features and Air quality
meteorological and spatial data, we targeted a specific segment in Delhi. Through correlation coefficient analysis, we
established direct and indirect relationships between the features and air quality.
•Moderate correlation with AQI and meteorological features : The temperature-AQI coefficient of 0.412452
suggests that an increase in temperature leads to higher pollution levels. Conversely, the humidity-AQI coefficient
of -0.32378 indicates that an increase in humidity is associated with lower pollution levels.
•Direct relation with AQI vs. human made, roadways, and others spatial features : The coefficient of human-
made features (0.24242) indicates that as their value increases, pollution levels also increase. This correlation
can be attributed to increased human activity and the use of pollutant sources, which leads to higher pollution
levels.
In line with the expected impact of roadways on AQI, our findings for Delhi exhibit a weak correlation coefficient
of 0.244309. The chosen observation area is characterized by a diverse roadway system, where an abundance of
roadways typically corresponds to increased vehicle activity. Though the weak correlation can be attributed to
the specific nature of the selected area. It is possible that the segment experienced low traffic due to congestion
in other areas with fewer roads, influencing the observed correlation. However, we see that for the ’Other spatial’
parameter, the coefficient is positive. The value of other features is negligible.
•Indirect natural land relation: An indirect relationship has been observed between Natural Land and AQI, with
a correlation coefficient (r-value) of -0.0339. An increase in natural land is associated with lower pollution levels,
indicating that it may have a mitigating effect on air pollution.
To summarize, although temperature and humidity correlate with AQI, spatial features also display significant
correlations. Therefore, we gave priority to spatial features in our analysis, alongside temperature and humidity.
4. Methodology
The study focuses on optimizing the placement of air pollution monitoring devices in the rapidly growing city,
Durgapur and the heavily polluted city, Delhi. The methodology involves three modules: Module 1 utilizes Principal
Component Analysis (PCA) and K-means Clustering for spatial optimality checks, Module 2 evaluates Air Quality
Index (AQI) similarity in clustered grids, and Module 3 shows an effective placement of AQMDs by incorporating
meteorological features. This process is outlined in Figure 8and detailed in Algorithm 1and Algorithm 2.
Module 1- Ideal placement check based on spatial data: Air pollution is influenced by two broad factors -
meteorological and spatial. In our study, we utilized geospatial data obtained through ‘mapImaging’ techniques since
grids represent distinct geospatial locations of the devices in a well-separated clustered form. Meteorological data,
which changes with time (temporal feature), cannot be employed to select grids. Therefore, the clustering of grids will
not be static, and seasonal variations in meteorological data can differ, sometimes being similar in different grids (The
Page 7 of 23

Optimizing Air Quality Monitoring Device Deployment
feature factor we have explained in Section 3.1).
When AQMD devices are placed, they serve as representatives of the entire grid. Clustering might change at one time
compared to another, making geospatial data more appropriate for our purposes since it does not typically vary with
time. Although several clustering algorithms are available, we opted to use K-Mean Clustering to cluster the dataset
in a manner where there is only one cluster for each data point. This was necessary for our study as we assumed that
the number of clusters formed would be equivalent to the number of devices available. Subsequently, we selected one
grid from each cluster for further analysis. There can be major outcomes at this step:
•The effective placement of AQMD can be confirmed by observing whether two or more devices converge in the
same cluster or not.
•If so, the device placement can be reassessed, and we can conclude that there might be a more ideal or effective
way of selecting the grid for device placement.
•Once the optimal locations for the already placed AQMDs are confirmed, we can then evaluate the grid selection
for placing new devices.
In steps 1-2 of our Algorithm 1, to perform K-Means clustering, we initially reduce the dimensionality of the
dataset. Specifically, we aim to reduce the number of spatial data points, denoted as ’n,’ to 2 or 3 components using a
specialized technique known as Principal Component Analysis (PCA). It is an unsupervised statistical approach that
involves finding the linear combination of a dataset that has maximum variance and removing its effect over the whole
dataset. It is a method to reduce the dimensionality of a large data set into smaller ones. Then we perform K-Mean
clustering on the data set to get K-clusters. After the clustering method, if we get some similar clustered grid, then we
have to check if the AQI label is similar or not in the same labeled clusters. If they are in the same AQI label, we can
eliminate one; in Algorithm 1, step 3 to 4, we depict this. In this way, we can check if an already placed device is in an
ideal way or not.
Module 2- Assessing AQI similarity within clustered grids : After clustering the grids depending on spatial
features, there might be different grids with almost the similar data (in the same cluster), which proves that the pre-
installed device is not in ideal way. In that case, we can reduce the device for that grid and look for another place.
Otherwise, if the AQI does not match, we can move forward with the placement of the device following the required
procedure.
Module 3- Enhancing AQMD placement efficiency with meteorological features: We begin by assessing the
effectiveness of the pre-installed devices’ placement. This evaluation involves following the steps outlined in Module
1 and 2. These steps are also utilized for the installation of a new AQMD, as depicted in Figure 8. Our second
algorithm outlines the methodology for placing new AQMDs. The process commences with conducting clustering on
the designated grids. Subsequently, we exclusively consider the unoccupied grids based on the clustering outcomes.
Following this, two conditions are applied:
1. Initially, assess whether there are any similar clustered grids within the unoccupied set. If multiple comparable
clustered grids are found, we give priority to targeting those grids. This approach enhances our area coverage by
selecting multiple similar clustered grids, thereby increasing the likelihood of representing other grids within the same
cluster. Opting to target a single grid from the similar clustered group enables us to effectively cover the remaining
grids.
2. After selecting a grid from the similar clustered grids, we proceed to check for any adjacent unoccupied grids
near the chosen one. If such nearby vacant grids are identified, they are included in the target selection process. It’s
important to note that we intentionally avoid selecting a grid that is immediately adjacent to the already installed
devices.
This two-step approach ensures that we focus on unoccupied grids while maximizing the area coverage and taking
advantage of cluster similarity and proximity. Once a suitable grid is selected, we must determine the optimal location
to install the new device. To achieve this, we utilize meteorological data, precisely temperature and humidity. Since,
a grid is relatively small compared to the area of observation, spatial data in this context is not particularly critical.
Instead of, meteorological (temporal) data, which is highly variable, can assist us in mapping pollutant gradients. The
steps for temporal mapping can be described as follows:
•We collect meteorological readings, specifically temporal features, from various locations within the chosen
grid.
Page 8 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 8: Modular structure of device placement algorithm
•Subsequently, we calculate the means of all the temporal features within that selected grid. We then determine
the mean deviation for all the data points in our dataset.
•The next step involves identifying the data point with the mean deviation closest to zero. The new device is
placed at this chosen data point.
Algorithm 1 explains the concepts behind Module 1 and Module 2, falling within the spatial data clustering of
pre-installed grids. It evaluates the presence of analog clustered grids with their respective Air Quality Index (AQI)
values, thereby measuring the suitability of previously installed Air Quality Monitoring Device (AQMD) locations.
Subsequently, Module 3 involves the deployment of the new AQMD, as detailed in Algorithm 2. Table 1is a notation
table that supports the clear delineation of these algorithms and their associated steps.
5. Results
Our research focuses on two regions: Durgapur and Delhi. While Durgapur serves as our initial implementation
site, Delhi, known for being one of the world’s most polluted areas, follows. Our algorithm’s efficacy and observations
are thoroughly discussed in both areas.
Page 9 of 23

Optimizing Air Quality Monitoring Device Deployment
Notation Definition
GiGrid of i-th location where, i={0,1,2,. . . ,x}, Gi∈G, x ∈W
kNumber of available devices(AQMD) already placed/number of clusters to be formed
𝐺𝑆
𝑖Spatial data for the grid in i th location where 𝐺𝑆
𝑖⊆ 𝑆 , S = {Natural land,human-made,
road-ways,other POI’s} extracted from each grid, Gi
𝐺𝐴𝑦
𝑖yindexed AQMD Ayplaced in corresponding iindexed grid Gi
𝐺𝑋After applying clustering algorithm, a set of grids having similar spatial distribution
which belonging in same cluster Cm
GAGrid with contain pre-installed AQMD, where GA∈G
GRGrid which is selected for new device placement from the set of grid Gi,
but not included in GA,GR∈G,GR∉GA
MMeteorological data (temperature and humidity)
𝐌Mean value of Meteorological data of whole GR
MtValue of Meteorological data of each lat-long of the traversed path in GR
△MtVariance of Meteorological data of each lat-long of the traversed path in GR
G(x,y) Latitude(x), longitude(y) of new AQMD in GR
Table 1
Notation table of our Algorithm
Algorithm 1 Optimality check (Effectiveness pre-placed AQMD ):
Require:
•Spatial Dataset, S
•Number of devices available = cluster to be formed = k
Ensure: Optimal placement of pre-installed Air Quality Monitoring Devices (AQMDs)
Step 1: Perform K-means Clustering
Perform K-means clustering on a set of grids 𝐺𝑖with 𝑘, considering 𝑘clusters based on the number of pre-installed
devices. Obtain clusters 𝐶= {𝐶1, 𝐶2, ..., 𝐶𝑘}, where the 𝑚-th cluster 𝐶𝑚is defined as 𝐶𝑚= {𝐺𝑚1, 𝐺𝑚2, ..., 𝐺𝑚𝑛},
with 𝑛∈ℕ, and 𝐺𝑚𝑖 is the grid belonging to the 𝑚-th cluster 𝐶𝑚.
Step 2: Check for Multiple AQMDs in a Cluster
If any cluster 𝐶𝑚contains more than one pre-installed AQMD, define 𝐺𝑥= {𝐺𝐴𝑦′
𝑖∈𝐶𝑚| |𝐺𝐴𝑦𝑖∩𝐺𝐴𝑦′
𝑖|>1}.
Step 3: Compare AQI Levels
Compare the Air Quality Index (AQI) levels for all 𝐺𝐴𝑦′
𝑖in 𝐺𝑥.
Step 4: Check AQI Dissimilarity
If AQI dissimilarity is observed, then 𝐺𝐴𝑦𝑖are in an optimal state; otherwise, 𝐺𝐴𝑦𝑖are not in an optimal state. For
all 𝐺𝑋,(AQI(𝐺𝑋)≠Constant ⇒𝐺𝐴𝑦𝑖) ∧ (AQI(𝐺𝑋) = Constant ⇒¬𝐺𝐴𝑦𝑖).
Step 5: Remove Redundant AQMDs
If 𝐺𝐴𝑦𝑖are not in an optimal state, remove the redundant ones. Calculate the Euclidean distance 𝑑for all 𝐺𝐴𝑦′
𝑖, and
define 𝐴𝑄𝑀𝐷𝑠_𝑡𝑜_𝑘𝑒𝑒𝑝 =Farthest(𝐺𝐴𝑦′
𝑖)for argmax(𝑑(𝐺𝐴𝑦′
𝑖)).
5.1. Evaluation in our study area Durgapur -
In our work, we have pre-installed devices located in four distinct grids: G0, G7, G9and G14, represented as L1,
L2, L3, and L4 respectively. Figure 9(a) shows the grid arrangement in our Durgapur city.
5.1.1. Ideal placement check based on spatial data: Tested in Durgapur City
In this section, an ideal placement method has been conducted based on spatial data and gain insights into the spatial
distribution of air pollution. As our contribution, decision-making and effective air pollution management strategies
have been adopted based on placement of devices. We use Machine Learning models to determine the possible relation
between AQI and spatial features. For this purpose, we took the pollutant concentration data of Durgapur from October
2019 to January 2020. To classify the spatial data of these locations, we employed a clustering method with k=4. As
a result, we obtained a cluster with four distinct labels, and L1 and L4 which are appeared to the same labeled cluster,
are represented by two green zones in Figure 9(b). We have implemented different clustering algorithms, including
Page 10 of 23

Optimizing Air Quality Monitoring Device Deployment
Algorithm 2 New AQMD placement :
Require:
•Grid with contain pre-installed AQMD 𝐺𝐴𝑦
𝑖
•Meteorological features: M
Ensure: Optimal placement of new Air Quality Monitoring Device (AQMD)
Step 1: Call ‘Optimality check’, After getting optimal pre-placed (O)
Step 2: Randomly select a grid, 𝐺𝑅, where, 𝐺𝑅∈G∧𝐺𝑅∉𝐺𝐴
Step 3: Compute the mean deviation of the meteorological features in 𝐺𝑅. for each lat,long of traverse path
△Mt←|𝑀𝑡-𝐌|, where t is the index traverse path.
Step 4: Select the lat, long where variance in minimum. Geo-position (lat, long) ←arg min (△Mt)
This is achieved by making an n-d graph for features and selecting the point closest to the origin
Step 5: Accuracy ←acc =([point]/AQI) x 100
Step 6: Return geo-location (latitude, longitude), 𝐺(𝑙𝑎𝑡,𝑙𝑜𝑛)
Figure 9: Algorithm 1 Steps: effective grid slection of optimal placement,here our pre-installed four AQMD grid denote as
L1, L2,L3 and L4
Grid K-means K-means K-med Mean Affinity K-means
++ (random) -oids shift
G0C3C2C4C2C1C4
G7C1C3C1C1C4C3
G9C2C1C3C4C2C2
G14 C3C2C4C2C1C4
Table 2
Comparative results of different clustering algorithms applied on our dataset. All algorithms show almost similar results for
two grids G0& G14.
’K-means++’, ’K-means (random)’, ’Mean Shift’, ’Affinity’, and ’K-Medoids’ in our experiment to achieve a decent
result. However, regardless of the algorithm used, all of them yielded almost similar results, as summarized in Table
2. We have preferred K-means clustering to be the most popular and easy to use as a solution for categorizing our
pre-installed devices.
5.1.2. AQI similarity check for two similar clustered grid
We have discovered that two pre-installed devices namely L1 and L4, are in the same labeled cluster (Figure 9b).
We analyze pollutant data grid-wise to check the similarity of AQI levels for two grids during the same time duration.
Our analysis revealed a 95% similarity between these two grids, as depicted in Figure 10. By segregating different time
segments of a day, we observed that they consistently displayed almost identical AQI levels. In contrast, the AQI level
among other grids (as shown in Figure 11) has been observed quite dissimilar. Based on the analysis of both spatial
and pollutant data, we can conclude that these two grids, L1 and L4, are nearly identical. Figure 9c) shows that the
unnecessary device is removed by using Euclidean distance formula due to pre-installed devices are belonged in same
Page 11 of 23

Optimizing Air Quality Monitoring Device Deployment
cluster based on spatial data. The steps we followed is depicted here;
We aim to choose a grid contain AQMD, either L1 or L4, from similar spatial structured grid set, 𝑋𝑖, (L1, L4) that
exhibits the greatest average Euclidean distance from the dissimilar spatial structured grid set 𝑌𝑗, (L2, L3). To achieve
this, we initialize max_distance as 0 and compute the Euclidean distances between each element of 𝑋𝑖and each element
of 𝑌𝑗. If a calculated distance exceeds the current max_distance, we update max_distance accordingly. Ultimately, the
selected grid will be the one associated with the highest average distance.
The euclidean distance between L1 to L2 & L3 are 0.037422 & 0.0190212, and L4 to L2 & L3 are 0.012225 &
0.013543 respectively. Now, after calculating distance and taking average the final result is max_distance = L1 to L2
& L3. As a result, the unwanted devices can be eliminated in this redundant grid configuration (as illustrated in Figure
9c). Our findings regarding the AQI dis-similarity of devices in other locations further support the strength of our
results. Consequently, we can assert that the placement of our pre-installed device in the targeted area was not ideally
located. As a ground truth, according to the cross-checking result of pre-installed devices in our targeted place, no
similarities of AQI level is found in the other locations represented in Figure 11.
Figure 10: AQI label similarity between two similar clustered places. Here L1 and L3 are belong to similar cluster
In our result, we achieved that our pre-installed devices in our targeted place are not in optimal way. For cross-
checking, we also found dissimilarities result of the other location devices’ as represent in Figure 11.
Figure 11: AQI similarity check for other device locations, which clearly signifies the dissimilarity of AQI label
5.1.3. Determining an ideal location for additional requirement of AQMD
In a particular situation, a redundant grid has been encountered in our study where a device can be easily removed
because of only three devices are sufficient for maximum area coverage. Our objective is to deploy a new device in
a different grid in an ideal way to maximize data coverage. To accomplish this, we follow a specific process which
has been described as Algorithm 2, is illustrated in Figure 12. We select a grid as folowed the step of Module-3 of
Methodology (see Section 4). The selected grid must have a distinct spatial distribution compared to the pre-placed
device, and it should not be an immediate neighbor of the pre-placed grid. As per our algorithm we choose grid G12,
also referred to as L5 in Figure 12, as our target grid. The next step is to determine the precise position within that grid.
We collect readings of meteorological features such as temperature and humidity from all across the grid and calculate
their mean. Then, calculate the variance of these two features. The latitude and longitude of the selected point are
chosen based on the minimum meteorological data variance. This selection ensures the accuracy in determining the
ideal position. In our case, the latitude and longitude of the chosen position are (23.535750, 87.274269), as depicted in
Page 12 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 12: Accurate location selection to place new AQMD in our targeted grid
Figure 13: Variance point of temporal data: The red point represents the position (latitude, longitude) of the new device,
denoting the tuple of temperature-humidity variance closest to zero(left image). The right image illustrates the data
collection path based on this variance consideration.
Figure 13. By following this process, we can strategically deploy additional devices in grids that enhance the overall
data coverage, considering factors such as spatial distribution and temporal data variance.
5.1.4. Comparative study of different device positions in Durgapur
To assess the accuracy of our approach, we compared pollutant data of different months from the Durgapur CPCB
5as a reference. Initially, we evaluated accuracy of the pre-installed Air Quality Monitoring Device (AQMD) using
Equation 1.
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (𝑚∕𝑛) ∗ 100 (1)
Where, actual AQI of pre-installed location = m, Actual AQI of CPCB = n
Here, L1, L2, L3 and L4 are our targeted locations for our study. We have considered different time periods of AQI
to check the possible highest accuracy. Our analysis based on AQI data shows an accuracy range of 75% − 78%. Our
objective is to determine an ideal location for placing the AQMD in Durgapur. Based on our analysis, we recommend
placing the device in L1, L2, and L3 to achieve maximum coverage. We obtained an accuracy of 93.4% by using
Equation 1, where we assumed the average AQI of these three grids as ’m’ and the AQI of CPCB as ’n’. This means
that our algorithm is able to predict the ideal location for the device with a high degree of accuracy, Table 3represents
the results. Next, we tested the accuracy of a newly placed AQMD using Equation 2. We compared the Average AQI of
the new grid with that of a specific point selected for an AQMD placement within the grid. Our findings revealed that
the selected point provided AQI values with an approximate accuracy of 97.2% compared to the grid average. Figure
14 is the visualization of the accuracy of different grid position.
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (𝑖∕𝐼) ∗ 100 (2)
5https://www.wbpcb.gov.in/
Page 13 of 23

Optimizing Air Quality Monitoring Device Deployment
Where AQI of the new position of chosen grid = i, AQI of the total chosen grid = I
Table 3
Comparison of device data with CPCB for Durgapur
AQI of Average AQI of pre- Actual AQI of optimal grid AQI of old L1,L2,L3 and
Devices installed devices/CPCB AQI L1,L2,L3/CPCB AQI new (grid-12)L5/CPCB AQI
Accuracy 77% 93.4% 97.2%
Figure 14: Distribution of AQMD to measure the accuracy of AQI
Our results support the validity of our methodology. By randomly placing the devices in four different locations,
we observed an accuracy rate of approximately 75% − 77%. However, it is important to note that two of these locations
had similarities in terms of spatial and pollutant characteristics, which affected the average accuracy and resulted in
a lower value. On the other hand, when we identified the appropriate locations for device placement, we achieved a
significantly higher accuracy rate of around 95%. Furthermore, when we selected an entirely new locations for device
placement, the accuracy reached its peak, indicating the strength of our methodology. These findings strongly confirm
the correctness of our selected positions and provide strong evidence in support of our approach.
5.2. Evaluation in our next study area, Delhi
We have encompassed a total area of 759.13 square kilometers within Delhi, virtually partitioned into 256 grids.
We have conducted our targeted observations across 31 AQMD stations of CPCB Delhi6as shown in Figure 15.
5.2.1. Assessment of spatial and AQI data similarities
After applying the K-means clustering algorithm across Delhi’s 256 grids, several Air Quality Monitoring Districts
(AQMDs), including Rohini, Sonya Vihar, Ashok Vihar, Punjabi Bagh, North Campus Du, Ito, Nehru Nagar, Okhla
Phase-2, and Major Dhyan Chand National Stadium, did not align with any similarly-clustered location. These nine grid
points exhibit distinct demographic characteristics and remain ungrouped due to differences in human-made features,
industrial presence, population distribution, and road networks, all of which our algorithm captures. Consequently, we
uphold their original positioning since they do not belong to any cluster. However, for evaluating the optimal placement
of preinstalled grids, we focus on clusters with more than two grids. Next, our attention turns towards evaluating those
grids that fall within analogous clusters, designated as Cluster 1, Cluster 2, Cluster 3, Cluster 4, Cluster 5, and Cluster
6. Let’s review the key findings for each cluster:
Cluster 1 (Pink): This cluster contains five stations: Jahangirpuri, IHBAS-Dilshad Garden, Shadipur, Patparganj,
and Mandir Marg. Among these, IHBAS-Dilshad Garden stands out with distinct pollutant levels. Patparganj and
6https://cpcb.nic.in/
Page 14 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 15: Exploring Delhi’s Grid Distribution: A total of 256 grids were clustered using k=31. The resultant clusters span
the entire city. This analysis centers on Cluster 1(Pink), 2(Red), 3(Yellow), 4(Green), 5(Orange) and 6(Blue), highlighted
with colored cells
Mandir Marg exhibit similarity in AQI for 31 days, prompting us to retain Mandir Marg, considering its proximity to
Patparganj and existing monitoring stations. Since Jahangirpuri and Shadipur share approximately 60% similar AQI
over a month, we opt to retain Jahangirpuri, given its distinctiveness from other clusters.
Cluster 2 (Red): This cluster encompasses Burari Crossing, Lodhi Road, and Sri Aurobindo Marg. Burari
Crossing’s AQI profile differs significantly from other stations in the cluster. Lodhi Road has two stations; however,
they share considerable AQI similarities. Sri Aurobindo Marg shows AQI correlation with Lodhi Road for around 21
days, suggesting redundancy. Thus, Sri Aurobindo Marg could be removed, while Burari Crossing and one Lodhi Road
station remain for their unique characteristics.
Cluster 3 (Yellow): In this cluster, Mundka, Vivek Vihar, Anand Vihar, NSIT Dwarka, and Dwarka Sector-8 are
present. Anand Vihar and Mundka exhibit distinct AQI patterns, justifying their retention. NSIT Dwarka, Dwarka
Sector-8, and Vivek Vihar share over 20 days of AQI similarity. Given this, we could eliminate one station; therefore,
retaining NSIT Dwarka due to its lack of nearby stations.
Cluster 4 (Green): Wazirpur, IGI Airport (T3), and RK Puram comprise this cluster. Wazirpur stands out with
dissimilar AQI trends. IGI Airport and RK Puram have notably similar AQI profiles. Considering RK Puram’s
proximity to other fixed AQMD stations, retaining IGI Airport appears logical due to its relatively isolated location.
Cluster 5 (Orange): This cluster contains Pusa and Sirifort stations, with no distinct AQI similarities. Since all
stations offer unique insights, we should retain three stations as removing any could result in data gaps.
Cluster 6 (Blue): This final cluster includes Chadni Chowk and CRRI Mathura Road. The two stations exhibit
substantial AQI correlation, shown in Figure 16, warranting the removal of one. Therefore, we can retain either station
based on practical considerations.
Page 15 of 23

Optimizing Air Quality Monitoring Device Deployment
Table 4
The final selection of AQMD stations in Delhi, accounting for negligible data loss in relation to the previous arrangement
Cluster no. - color Previous station After applying algorthm Data loss
1 - Pink 5; Jahangirpuri, IHBAS- Dilshad Garden,
Shadipur, Patparganj, Mandir Marg
3; IHBAS- Dilshad Garden,
Jahangirpuri, Mandir Marg 7%
2 - Red 4; Burari Crossing, Lodhi Road(2),
Sri Aurobindo Marg 2; Burari Crossing, Lodhi road 14%
3 - Yellow 5; Mundka, Vivek Vihar, Anand Vihar,
NSIT Dwarka, Dwarka Sector-8
3; Anand Vihar, Mundka,
NSIT Dwarka 2%
4 - Green 3; Wazirpur, IGI airport(T3), RK Puram 2; Wazirpur, IGI airport(T3) 0.04%
5 - Orange 3; Pusa(2), Sirifort No change No change
6 - Blue 2; Chadni Chowk, CRRI Mathura Road 1; CRRI Mathura Road 0.7%
Regarding Cluster 1, we may prioritize retaining Mandir Marg over Patparganj due to the presence of Anand Vihar
station near Patparganj. This choice ensures a better distribution of monitoring stations. Additionally, Shadipur from
the same cluster may be removed since the nearby Pusa station adequately covers its location. In the case of Cluster
3, NSIT Dwarka is the preferred station to keep, given the presence of an IGI Airport station near Dwarka Sector 8
and an IHBAS station near Vivek Vihar. This decision optimizes the coverage of different areas without redundancy.
Considering Cluster 4, IGI Airport station may be retained over RK Puram due to the existence of established AQMD
stations in the vicinity of RK Puram. Choosing IGI Airport ensures better coverage of areas with minimal overlap.
Figure 16: AQI Similarity Comparison of Chandni Chowk and CPRI Mathura Road. One-month data visualization reveals
1 instance of dissimilar AQI over 2 days, and 28 days of AQI similarity, indicating resembling demography and environment
5.2.2. Conclusive determination and key insights from the results
By optimizing the distribution of air quality monitoring stations, we can make significant improvements in data
collection efficiency. In cluster 1 regions, transitioning from 5 AQMD to 3 AQMD leads to a data retention of 93%,
resulting in a mere 7% data loss. Similarly, in cluster 2, shifting from 4 stations to 2 AQMD yields an 86% data capture
rate, with a corresponding 14% data loss.Moving on to cluster 3, replacing 5 AQMD with 3 AQMD culminates in an
impressive 98% data retention and a minimal 2% data loss. For cluster 4, a shift from 3 AQMD to 2 AQMD results in a
negligible data loss of 0.04%. Lastly, within cluster 6, utilizing 1 AQMD in place of 2 only incurs a marginal 0.7% data
loss. The summarization of our results is shown in Table 4. Among the 22 stations, only 14 are necessary, allowing
for the removal of 8 stations and resulting in cost savings. After completing all the analyses, the ultimate outcome is
visualized in Figure 17. Our decision is supported by minimal data loss, serving as evidence of the effectiveness of our
approach.
6. Related work
The optimal sensor placement problem has been studied to monitor spatial and temporal phenomena, like
temperature sensing and field soil moisture estimation. Given limited budget constraints, our main motive is what
strategy we should derive from achieving an integrated goal. For that purpose, we should know the relationship between
Page 16 of 23

Optimizing Air Quality Monitoring Device Deployment
Figure 17: The figure illustrates the final AQMD placements; reduction of 8 AQMDs in Delhi after our Algorithm
implementation. Black cells depict the revised AQMDs, while white cells represent unchanged AQMDs. Consequently, the
pollutant data collection in Delhi can be streamlined to 23 AQMDs out of the initial 31
spatio-temporal features and air quality. It is proven that if we want to do research work in a place, we need to know the
targeted place’s nature (spatial or temporal nature), then we can provide some microservices. Variations in temporal
characteristics (e.g., temperature, humidity) can affect mortality Zanobetti (2002). Not only temporal characteristics
but also spatial characteristics can affect the efficiency of green growth, which has also been proven in Chinese cities
Ma, Long, Chen, Tu, Zhang and Liao (2019). On the other hand, if we try to reduce severe pollution levels in some
areas, it is, therefore, necessary to design and construct a pipeline system of traffic or roadways or personal exposure
responsible for bad air quality. In my earlier research, I used multiple linear regression and Pearson correlation models
to assess the spatial and temporal correlation with the Air Quality Index (AQI) Sarkar (2022). The results show that
the three most polluted cities in India—Chennai, Delhi, and Hyderabad—as well as Durgapur, our current research
site, have significantly different levels of air pollution. It also proved that the relationship varies region-wise. Spatio-
temporal features (Like human-made, natural land, temperature, humidity) also depend upon human activities (like
stoves, bakeries, fish burning, etc.).
6.1. Development and placement of environment monitoring device : indoor and outdoor
In research works, platform integration and qualitative validation are the key subjects. Common Sense Broich,
Gerharz and Klemm (2012), and CitiSense develops sensor nodes with cellular connectivity and interfaces for
crowd-sourced gas pollution monitoring. AirSense Zhuang, Lin, Yoo and Xu (2015) conducts feasibility analysis in
diverse circumstances and combines dust sensors for personal PM2.5monitoring. According to Krause et al., Krause,
Rajagopal, Gupta and Guestrin (2011), a selection of areas should be chosen so that wireless sensors can accurately
forecast specific upcoming scenarios, such as the traffic speed on a highway. A.V. Donkelaar et al.van Donkelaar,
Martin and Park (2006) propose to determine PM2.5using a contemporary resolution image spectroradiometer, one
of the most noteworthy ones. The energy independence of the nodes, the ability to change the network topology with
flexibility, the improvement of the spatial-temporal detail of observations, and lower operating expenses are all benefits
of a wireless sensor network Schurgers, Tsiatsis and Srivastava (2002). According to a technique for monitoring and
forecasting air pollution, Shaban et al.Shaban et al. (2016) use a system that is inexpensive air quality monitoring motes
that are fitted with various gaseous and meteorological sensors.
Utilizing indoor air quality monitoring equipment is the subject of some Sharma, Poddar, Dey, Nandi, De, Saha, Mondal
and Saha (2017) research. By applying the clustering approach of the Fuzzy C-means (FCM) algorithm, it has been
Page 17 of 23

Optimizing Air Quality Monitoring Device Deployment
demonstrated Liu, Sun, Yu, Yue and Zhang (2016) in certain studies that it is necessary to increase environmental
safety, accelerate planning, and build new urbanization to monitor the indoor environment. Kim et al. Kim, Chu and
Shin (2014) proposed a real-time system for monitoring indoor pollutants using MOS (Metal Oxide Semiconductor)
sensors. They applied several calibration approaches and aggregation algorithms to lower network traffic and electricity
usage. A method to monitor indoor air quality on several building floors was proposed by Chen et al. Chen, Zheng,
Chen, Jin, Sun, Chang and Ma (2014) utilizing a Purification Time Inference (PTI) model based on ANN.
Statistics on air quality are gathered using drone technology Yu, Sheppard, Lumley, Koenig and Shapiro (2000) in
3D space. The appropriate sensors, including methane (CH4), carbon monoxide (CO), sulfur dioxide (SO2), nitrogen
dioxide (NO2), ozone (O3), smoke, and particulate matter (PM10) sensors, are installed on a drone that is used to
achieve this. A microcontroller called an Arduino UNO is used to analyze real-time sensor readings, and the resulting
data is then preserved on a Blynk (cloud) server. The ability to show the data in a mobile device or base station unit
results from storage in the cloud. However, temperature and humidity observations must also be included to calculate
an appropriate air quality index.
6.2. Optimal placement techniques for environment monitoring devices
Existing theoretical studies, however, are deterministic; duty-cycled sensor networks’ coverage and connection are
examined Wang, Xing, Zhang, Lu, Pless and Gill (2003); Yan, He and Stankovic (2003). Then, the most informative
places are determined to be the best sensor placement using information theory metrics like entropy Cressie (2015)
or mutual information Guestrin, Krause and Singh (2005). Among these, several applications for monitoring spatial
phenomena like temperature Krause, Guestrin, Gupta and Kleinberg (2006) and soil moisture Wu, Liu and Wu
(2012) have adopted GP-based techniques. In Lim, Miao, Chew, Lee and Raju (2012), High precision geospatial risk
assessment of wind channels in an urban environment has been achieved using CFD. CFD modeling has also been
utilized in data center contexts to solve sensor placement issues Wang, Wang, Xing, Chen, Lin and Chen (2011) and
anticipate temperatureChen, Tan, Wang, Xing, Wang, Wang, Punch and Colbry (2012). A comparable system was
implemented and trialed in several European locations like Cambridge Wu et al. (2012). In order to locate RWIS
stations strategically over a vast regional transportation network, this research suggests a spatial optimization strategy.
Pourali et al. Mosleh and Pourali (2013) found a set of functional sites using a Bayesian network so that sensors can be
placed there to monitor intricate power systems. Du et al. Du, Xing, Li, He, Chua and Miao (2014) to better quantify
the surface wind distribution over a sizable urban reservoir, seek to identify a set of sensor deployment locations.
They discover the spots with the most mutual knowledge with others in order to address this challenge. Due to the fact
that they do not consider incremental deployment, their solutions cannot be immediately applied to our issue. Some
publications have proposed methods like spatial statistics Cressie (2015) and subset selection Das and Kempe (2008).
Due to wind’s temporal and spatial variability, they cannot be used directly to measure wind distribution. Based on the
laws of thermodynamics and fluid mechanics, a thermal forecasting model is suggested Li, Liang, Liu, Nath, Terzis
and Faloutsos (2011) to simulate and forecast temperatures surrounding servers in the data center. Karamshuk et al.
Karamshuk, Noulas, Scellato, Nicosia and Mascolo (2013) seek to identify areas where new retail installations can
attract the greatest number of clients. Erdos et al. Erdös, Ishakian, Lapets, Terzi and Bestavros (2012) plan to place
sensors in a network that delivers information to improve the detection of duplicate data contents. The FHWA ESS siting
recommendations do not include precise placement instructions for monitoring places where weather-related collisions
occur. However, they offer some general instructions on choosing sites appropriately based on the site-specific issue to
be addressed Manfredi, Walters, Wilke, Osborne, Hart, Incrocci, Schmitt, Garrett, Boyce and Krechmer (2008). Slick
pavement, poor visibility, strong gusts, and water levels are some of these circumstances. For the management of road
weather Goodwin et al. (2003), the FHWA created a list of best practices in 2003. 30 RWIS stations implemented
in 21 states are the subject of case studies in this paper. Pindado, Cubas and Sorribes-Palmer (2014) LoRaWAN as
an optimal IoT-based surveillance system with frequencies below 1GHz, some employing GSM and Thingspeak. The
theoretical interpolation inaccuracy averaged over the area of interest, i.e., the mean kriging variance is reduced(e.g.,
Baume et al. (2011); Wu and Bocquet (2011)) in another widely used criterion derived with the use of a geostatistical
estimation technique known as kriging. The chemical transport model (CTM), which simulates physical and chemical
methods such as emission, advection, photochemical reactions, and deposition, has been widely used at various spatial
and temporal scales in the study of air quality, not only to obtain a spatial distribution but also to create an efficient
approach for the control of the levels of air pollutants (e.g., Emmons, Walters, Hess, Lamarque, Pfister, D, Granier,
Guenther, D, Laepple, J, Tie, G, Wiedinmyer, Baughcum and Kloster (2009); Chatani, Morino, Shimadera, Hayami,
Mori, Sasaki, Kajino, Yokoi, Morikawa and Ohara (2014)). However, despite its ability to find locally optimal solutions,
Page 18 of 23

Optimizing Air Quality Monitoring Device Deployment
SA is frequently confined to areas that are distant from the global optimum (Ruiz-Cardenas, Ferreira and Schmidt
(2009)). In an attempt to optimize a PM2.5monitoring network using simulated values acquired from CTM to estimate
the mean kriging variance, Araki, Iwahashi, Shimadera, Yamamoto and Kondo (2015) developed a hybrid of GA
and SA (HGS). Research work Jin, Walker, Cebelak and Walton (2014) focuses on determining strategic locations
for environmental sensor stations based on weather-related crash data. The study aims to identify the key factors that
influence the probability of weather-related crashes and identify locations for sensor stations that can provide useful
information for mitigating such crashes. The researchers used a combination of data sources, including traffic crash
data, weather data, and GIS data, to develop a methodology for identifying optimal locations for environmental sensor
stations. They used a statistical model to analyze the data and identify key factors that affect the probability of weather-
related crashes. The results of the study showed that locations with high traffic volume and high exposure to weather
hazards, such as high winds, snow, and ice, have a higher probability of weather-related crashes. The study provides
valuable insights into the strategic placement of environmental sensor stations for monitoring weather-related risks
and improving road safety.
6.3. Optimal placement of air quality monitoring device
The research works maximum area covering sensor-ed device placement concerning air quality monitoring data is
quite challenging due to data unavailability, user feasibility, etc. The study Rahman, Usama, Tahir and Uppal (2022)
focused on the development of a data-driven framework for the analysis of air quality landscape in the city of Lahore,
Pakistan. The authors used various techniques such as geographic information systems (GIS), remote sensing, and
machine learning algorithms to collect and analyze data related to air quality in Lahore. The study found that Lahore’s
air quality is generally poor, with high levels of particulate matter (PM2.5and PM10), nitrogen dioxide (NO2), and
sulfur dioxide (SO2). The authors suggested that this could be due to a combination of factors, including vehicular
emissions, industrial activities, and biomass burning. The paper concludes by recommending the implementation of
air quality monitoring stations and the use of data-driven approaches to develop effective strategies for mitigating air
pollution in Lahore. Overall, the research provides valuable insights into the complex relationship between air quality
and urban development in developing countries like Pakistan. The paper Yu, Chang, Yu, Guo and Shi (2021) proposes
a location selection approach for air quality monitoring that takes into account the limited budget and estimation error.
The authors developed a mathematical model to optimize the location selection process by considering the monitoring
cost, estimation error, and the importance of the monitoring sites. The proposed model was evaluated using real-world
data from the city of Beijing, China. The results of the study showed that the proposed approach outperformed existing
methods in terms of estimation accuracy while keeping the monitoring cost within a limited budget. The study also
identified the most important monitoring sites in Beijing based on their contribution to improving the overall estimation
accuracy. The paper’s findings can be useful for policymakers and city planners in developing effective air quality
monitoring strategies with limited resources. The proposed approach can also be extended to other areas, such as water
quality monitoring and noise pollution monitoring, where limited budgets and estimation errors are critical factors in
the location selection process. Overall, the research provides valuable insights into the importance of location selection
in air quality monitoring and the need to consider various factors in optimizing the monitoring process. Another work
Robinson, Franklin and Roberts (2021) discusses the importance of optimizing sensor coverage and network design
in creating a responsive city that prioritizes equity. The authors argue that the deployment of sensor networks can
enable cities to collect data on various environmental factors, such as air and water quality, noise pollution, and traffic
congestion. However, they note that sensor deployment strategies are often biased towards more affluent areas, resulting
in an inequitable distribution of data. To address this issue, the paper proposes a method for optimizing sensor coverage
that considers the demographic characteristics of different neighborhoods, including income levels and race/ethnicity.
The proposed approach involves using machine learning algorithms to predict where sensors should be deployed to
ensure equitable data collection. The study also discusses the importance of network design in creating a responsive
city. The authors argue that sensor networks must be designed to be scalable and adaptable to changing conditions to
enable effective data collection and analysis. Overall, the research highlights the need for cities to prioritize equity in
sensor deployment and network design to ensure that all communities have access to vital environmental data. The
paper’s proposed approach can be useful for policymakers and urban planners in developing effective strategies for
deploying sensor networks that prioritize equity and promote a responsive city. The work Sun et al. (2019) presents a
citizen-centric approach for air quality monitoring through optimal sensor placement in the city of Cambridge, UK. The
objective of the study was to enhance the coverage of air quality monitoring in the city by placing a limited number of
low-cost sensors at optimal locations. The study utilized data from the UK’s Automatic Urban and Rural Monitoring
Page 19 of 23

Optimizing Air Quality Monitoring Device Deployment
Table 5
Comparison of Research article 1 Sun et al. (2019), 2 Hsieh et al. (2015) and 3 Yu et al. (2021) and our algorithm for
air quality monitoring device placement in strategic ways, here Features are represented as some code; F-1 -Spatial, F-2 -
Temporal, F-3 - Population, F-4 - Others features
Article Location Objective Features Outcome Limitation
F-1 F-2 F-3 F-4
1 Cambridge Optimize placement ✓×✓ ✓ Improved spatial Focused on one city,
(U.K) of low-cost coverage and accuracy weather patterns
AQMD for citizen-led effects were not considered
2 Beijing Infer air quality for ✓ ✓ ×✓Inferred air quality Limited applicability
station location from urban big data can absence of consideration
recommendation guide recommended for weather effects
locations to improve monitoring
3 China Develop location ✓ ✓ ×✓85% accuracy Lack of device information
selection method for air in estimation absence of consideration
quality monitoring for weather effects on AQI
Our Durgapur, Develop a universal algorithm ✓ ✓ × × Significantly reduces the Meteorological data
approach Delhi for optimal device placement, number of stations while Insufficient for
(India) maximizing coverage while maintaining 80-90% AQMD Placement
minimizing input data accuracy in estimation in Delhi
Network (AURN) and used an optimization algorithm to identify the optimal locations for sensor placement. The
results of the study showed that the proposed approach significantly improved the coverage of air quality monitoring in
the city while minimizing the cost of sensor deployment. The accuracy of the results was verified through a comparison
of the air quality data collected from the optimal sensor placement locations and AURN data. Overall, this research
work presents an innovative approach for enhancing air quality monitoring in urban areas using citizen-centric sensor
placement. Hsieh et al. (2015) presented a method for inferring air quality levels in a city based on urban big data, with
the objective of recommending optimal locations for air quality monitoring stations. The authors utilized various data
sources, including meteorological data, land-use data, and social media data, to build a machine-learning model for
predicting air quality levels at unmonitored locations. The proposed approach was applied to the city of Taipei, Taiwan,
and the results demonstrated the effectiveness of the method in identifying optimal locations for air quality monitoring
stations. This paper highlights the importance of leveraging big data to address environmental challenges and provides
insights into the potential of data-driven approaches for optimizing air quality monitoring networks. Table 5refers to
the comparison with existing research work and our approach.
All the studies mentioned above aim to place sensors or stations at the most informative locations; sometimes, it
is impractical due to complex procedures and high costs. However, most of the research focused on the placement
quality of the approaches, which will depend on the accuracy of the air quality model and require existing air quality
measurements in the region as inputs to the air quality estimation model. In our research, we put forward the following
questions regarding air quality monitoring. Firstly, in the absence of prior knowledge of the field, like population and
traffic pattern, which procedure should we follow? Secondly, if some place already has some AQMD, how did we
know it is in an optimal way that can cover up almost all land area of the targeted place? Thirdly, if we want to deploy
a new device in one place that already has some pre-installed devices and is in an already ideal way, what would be
the suitable sensor placement strategy given a fixed budget constraint.
7. Conclusion and future work
In the face of escalating environmental concerns, air quality monitoring has gained paramount importance to
combat pollution. Our study has endeavoured to contribute to this ongoing effort by focusing on optimizing the
placement of air quality monitoring devices to maximize data collection efficacy. Our primary objective was to
assess the adequacy of existing device placements and propose an algorithmic solution for their optimal positioning.
The results obtained from our study, specifically in the context of Durgapur, underscored the potential of our
approach. Through spatial clustering, we identified the need for better coverage, leading us to employ temporal
mapping to pinpoint optimal device placements within specific grids. The preliminary implementation of our algorithm
demonstrated promising outcomes, achieving an accuracy rate of 80-90% when compared to grid-average air quality
data. Looking ahead, our research holds several avenues for further exploration and refinement. One prominent
direction involves enhancing the temporal mapping technique to reduce errors and improve the accuracy of suggested
Page 20 of 23

Optimizing Air Quality Monitoring Device Deployment
placements. Moreover, extending our algorithm’s applicability to other urban settings remains a priority, as each locale
presents unique challenges and opportunities for air quality monitoring.
While we successfully evaluated the optimality of pre-installed devices in Durgapur and assessed their similarity
through AQI levels, we recognize the limitations faced in applying our algorithm to Delhi due to the unavailability of
temporal data. As part of our future scope, we aspire to overcome this limitation and expand the scope of our algorithm
to include Delhi and other cities, thereby contributing to a more comprehensive and adaptable air quality monitoring
solution. Our study’s initial achievements highlight the potential of our algorithmic approach in optimizing air quality
monitoring device placements. By addressing existing limitations and advancing our techniques, we anticipate making
substantial contributions to the global endeavour of combating air pollution and promoting healthier environments for
present and future generations.
Declarations
•Conflict of interest/Competing interests: The authors affirm that there are no identifiable conflicting financial
interests or personal affiliations that could have potentially influenced the findings presented in this paper.
•Availability of data and materials: The dataset will be provided upon reasonable request.
Author contributions
All authors contributed to the study conception and design. Design the concept & analysis were performed by
[Pritisha Sarkar], [Munsi Yusuf Alam] and [Mousumi Saha]. The first draft of the manuscript was written by [Pritisha
Sarkar] and all authors commented on previous versions of the manuscript. The manuscript was edited by [Saurav
Mallik], [Arup Roy] and [Ujjwal Maulik]. All authors read and approved the final manuscript.
Funding
The authors did not receive support from any organization for the submitted work
References
Alexander, R., Imberger, J., 2009. Spatial distribution of motile phytoplankton in a stratified reservoir: the physical controls on patch formation.
Journal of plankton research 31, 101–118.
Araki, S., Iwahashi, K., Shimadera, H., Yamamoto, K., Kondo, A., 2015. Optimization of air monitoring networks using chemical transport model
and search algorithm. Atmospheric Environment 122, 22–30.
Baume, O., Gebhardt, A., Gebhardt, C., Heuvelink, G., Pilz, J., 2011. Network optimization algorithms and scenarios in the context of automatic
mapping. Computers & Geosciences 37, 289 – 294. doi:10.1016/j.cageo.2010.04.014.
Broich, A., Gerharz, L., Klemm, O., 2012. Personal monitoring of exposure to particulate matter with a high temporal resolution. Environmental
science and pollution research international 19, 2959–72. doi:10.1007/s11356- 012-0806-3.
Chatani, S., Morino, Y., Shimadera, H., Hayami, H., Mori, Y., Sasaki, K., Kajino, M., Yokoi, T., Morikawa, T., Ohara, T., 2014. Multi-model
analyses of dominant factors influencing elemental carbon in tokyo metropolitan area of japan. Aerosol and Air Quality Research 14, 396–405.
doi:10.4209/aaqr.2013.02.0035.
Chen, J., Li, J., He, S., He, T., Gu, Y., Sun, Y., 2013. On energy-efficient trap coverage in wireless sensor networks 10. doi:10.1145/2529973.
Chen, J., Tan, R., Wang, Y., Xing, G., Wang, X., Wang, X., Punch, B., Colbry, D., 2012. A high-fidelity temperature distribution forecasting system
for data centers , 215–224.
Chen, X., Zheng, Y., Chen, Y., Jin, Q., Sun, W., Chang, E., Ma, W.Y., 2014. Indoor air quality monitoring system for smart buildings , 471–475.
Cressie, N., 2015. Statistics for spatial data. John Wiley & Sons.
Das, A., Kempe, D., 2008. Sensor selection for minimizing worst-case prediction error , 97–108.
Dhillon, S., Chakrabarty, K., 2003. Sensor placement for effective coverage and surveillance in distributed sensor networks 3, 1609–1614 vol.3.
doi:10.1109/WCNC.2003.1200627.
van Donkelaar, A., Martin, R.V., Park, R.J., 2006. Estimating ground-level pm2.5 using aerosol optical depth determined from satellite remote
sensing. Journal of Geophysical Research: Atmospheres 111. doi:https://doi.org/10.1029/2005JD006996.
Du, W., Xing, Z., Li, M., He, B., Chua, L.H.C., Miao, H., 2014. Optimal sensor placement and measurement of wind for water quality studies in
urban reservoirs .
Emmons, L., Walters, S., Hess, P., Lamarque, J.F., Pfister, G., D, F., Granier, C., Guenther, A., D, K., Laepple, T., J, O., Tie, X., G, T., Wiedinmyer,
C., Baughcum, S., Kloster, S., 2009. Description and evaluation of the model for ozone and related chemical tracers, version 4 (mozart-4).
Geoscientific Model Development Discussions 3. doi:10.5194/gmd- 3-43- 2010.
Erdös, D., Ishakian, V., Lapets, A., Terzi, E., Bestavros, A., 2012. The filter-placement problem and its application to minimizing information
multiplicity 5. doi:10.14778/2140436.2140439.
Page 21 of 23

Optimizing Air Quality Monitoring Device Deployment
Fann, 2011. The public health context for pm 2.5 and ozone air quality trends. Air Quality, Atmosphere & Health 6. doi:10.1007/
s11869-010- 0125-0.
Geng, 2021. Drivers of pm2.5 air pollution deaths in china 2002–2017. Nature Geoscience 14, 1–6. doi:10.1038/s41561- 021-00792- 3.
Giannadaki, 2016. Implementing the us air quality standard for pm2.5 worldwide can prevent millions of premature deaths per year. Environmental
Health 15, 88. doi:10.1186/s12940-016- 0170-8.
Goodwin, L.C., et al., 2003. Best practices for road weather management: Version 2.0. Technical Report. Road Weather Management Program
(US).
Guestrin, C., Krause, A., Singh, A., 2005. Near-optimal sensor placements in gaussian processes. ICML 2005 - Proceedings of the 22nd International
Conference on Machine Learning , 265–272doi:10.1145/1102351.1102385.
Hsieh, H.P., Lin, S.D., Zheng, Y., 2015. Inferring air quality for station location recommendation based on urban big data, in: Proceedings of the
21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Association for Computing Machinery, New York,
NY, USA. p. 437–446. URL: https://doi.org/10.1145/2783258.2783344, doi:10.1145/2783258.2783344.
Jin, P.J., Walker, A., Cebelak, M., Walton, C.M., 2014. Determining strategic locations for environmental sensor stations with weather-
related crash data. Transportation Research Record 2440, 34–42. URL: https://doi.org/10.3141/2440-05, doi:10.3141/2440-05,
arXiv:https://doi.org/10.3141/2440-05.
Joshi, S., Boyd, S., 2008. Sensor selection via convex optimization. IEEE Transactions on Signal Processing 57, 451–462.
Karamshuk, D., Noulas, A., Scellato, S., Nicosia, V., Mascolo, C., 2013. Geo-spotting: Mining online location-based services for optimal retail
store placement doi:10.1145/2487575.2487616.
Kim, J., Chu, C., Shin, S., 2014. Issaq: An integrated sensing systems for real-time indoor air quality monitoring. IEEE Sensors Journal 14,
4230–4244. doi:10.1109/JSEN.2014.2359832. publisher Copyright: ©2014 IEEE.
Krause, A., Guestrin, C., Gupta, A., Kleinberg, J., 2006. Near-optimal sensor placements: Maximizing information while minimizing communication
cost doi:10.1145/1127777.1127782.
Krause, A., Rajagopal, R., Gupta, A., Guestrin, C., 2011. Simultaneous optimization of sensor placements and balanced schedules. IEEE
Transactions on Automatic Control 56, 2390–2405. doi:10.1109/TAC.2011.2164010.
Kumar, S., Lai, T.H., Balogh, J., 2004. On k-coverage in a mostly sleeping sensor network doi:10.1145/1023720.1023735.
Leung, D.Y.C., 2015. Outdoor-indoor air pollution in urban environment: challenges and opportunity. Frontiers in Environmental Science 2. URL:
http://journal.frontiersin.org/Journal/10.3389/fenvs.2014.00069/pdf, doi:10.3389/fenvs.2014.00069.
Li, L., Liang, C.J.M., Liu, J., Nath, S., Terzis, A., Faloutsos, C., 2011. Thermocast: a cyber-physical forecasting model for datacenters , 1370–1378.
Lim, T.K., Miao, H., Chew, C., Lee, K.K., Raju, D.K., 2012. Environmental modeling for geospatial risk assessment of wind channels in singapore.
GSDI 13, 1–11.
Liu, 2020. Exploring the relationship between air pollution and meteorological conditions in china under environmental governance. Scientific
Reports 10, 14518. doi:10.1038/s41598-020- 71338-7.
Liu, L., Sun, S., Yu, H., Yue, X., Zhang, D., 2016. A modified fuzzy c-means (fcm) clustering algorithm and its application on carbonate fluid
identification. Journal of Applied Geophysics 129, 28–35. doi:10.1016/j.jappgeo.2016.03.027.
Lu, 2019. A temporal approach for air quality forecast , 144–151.
Ma, L., Long, H., Chen, K., Tu, S., Zhang, Y., Liao, L., 2019. Green growth efficiency of chinese cities and its spatio-temporal pattern. Resources,
Conservation and Recycling 146, 441–451. doi:10.1016/j.resconrec.2019.03.049.
Manfredi, J., Walters, T., Wilke, G., Osborne, L., Hart, R., Incrocci, T., Schmitt, T., Garrett, V.K., Boyce, B., Krechmer, D., 2008. Road weather
information system environmental sensor station siting guidelines, version 2.0. Technical Report.
Mosleh, A., Pourali, M., 2013. A functional sensor placement optimization method for power systems health monitoring. IEEE Transactions on
Industry Applications 49. doi:10.1109/TIA.2013.2257978.
Pindado, S., Cubas, J., Sorribes-Palmer, F., 2014. The cup anemometer, a fundamental meteorological instrument for the wind energy industry.
research at the idr/upm institute. Sensors 14, 21418–21452.
Pramanik, 2023. Aquamoho: Localized low-cost outdoor air quality sensing over a thermo-hygrometer. ACM Trans. Sen. Netw. 19. URL:
https://doi.org/10.1145/3580279, doi:10.1145/3580279.
Rahman, A., Usama, M., Tahir, M., Uppal, M., 2022. Data driven framework for analysis of air quality landscape for the city of lahore. The
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XLVIII-4/W5-2022, 167–173. doi:10.5194/
isprs-archives- XLVIII-4- W5-2022-167- 2022.
Robinson, C., Franklin, R., Roberts, J., 2021. Optimising for equity: Sensor coverage, networks and the responsive city. doi:10.21203/rs.3.
rs-902765/v1.
Ruiz-Cardenas, R., Ferreira, M., Schmidt, A., 2009. Stochastic search algorithms for optimal design of monitoring networks. Environmetrics 21,
102 – 112. doi:10.1002/env.989.
Sarkar, 2022. City-wide spatio-temporal effect on aqi, pp. 13–18. doi:10.1109/ICCCMLA56841.2022.9989098.
Schurgers, C., Tsiatsis, V., Srivastava, M., 2002. Stem: Topology management for energy efficient sensor networks, pp. 3–1099. doi:10.1109/
AERO.2002.1035239.
Shaban, K.B., Kadri, A., Rezk, E., 2016. Urban air pollution monitoring system with forecasting models. IEEE Sensors Journal 16, 2598–2606.
Sharma, 2021. Can i go for a roof walk today? know your housing’s air quality from a thermo-hygrometer, in: Proceedings of the 8th ACM
International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, Association for Computing Machinery, New
York, NY, USA. p. 91–100. URL: https://doi.org/10.1145/3486611.3486663, doi:10.1145/3486611.3486663.
Sharma, P., Poddar, B., Dey, S., Nandi, S., De, T., Saha, M., Mondal, S., Saha, S., 2017. On detecting acceptable air contamination in classrooms
using low cost sensors, pp. 484–487. doi:10.1109/COMSNETS.2017.7945439.
Sun, C., Li, V.O.K., Lam, J.C.K., Leslie, I., 2019. Optimal citizen-centric sensor placement for air quality monitoring: A case study of city of
cambridge, the united kingdom. IEEE Access 7, 47390–47400. doi:10.1109/ACCESS.2019.2909111.
Page 22 of 23

Optimizing Air Quality Monitoring Device Deployment
Wang, X., Wang, X., Xing, G., Chen, J., Lin, C.X., Chen, Y., 2011. Towards optimal sensor placement for hot server detection in data centers ,
899–908.
Wang, X., Xing, G., Zhang, Y., Lu, C., Pless, R., Gill, C., 2003. Integrated coverage and connectivity configuration in wireless sensor networks.
SenSys’03: Proceedings of the First International Conference on Embedded Networked Sensor Systems 1, 28–39. doi:10.1145/958491.
958496.
Wu, L., Bocquet, M., 2011. Optimal redistribution of the background ozone monitoring stations over france. Atmospheric Environment 45, 772–783.
doi:10.1016/j.atmosenv.2010.08.038.
Wu, X., Liu, M., Wu, Y., 2012. In-situ soil moisture sensing: Optimal sensor placement and field estimation. ACM Transactions on Sensor Networks
(TOSN) 8, 1–30.
Yan, T., He, T., Stankovic, J.A., 2003. Differentiated surveillance for sensor networks doi:10.1145/958491.958498.
Yu, O., Sheppard, L., Lumley, T., Koenig, J.Q., Shapiro, G.G., 2000. Effects of ambient air pollution on symptoms of asthma in seattle-area children
enrolled in the camp study. Environmental Health Perspectives 108, 1209–1214. doi:10.1289/ehp.001081209.
Yu, Z., Chang, H., Yu, Z., Guo, B., Shi, R., 2021. Location selection for air quality monitoring with consideration of limited budget and estimation
error. IEEE Transactions on Mobile Computing PP, 1–1. doi:10.1109/TMC.2021.3065656.
Zanobetti, A., 2002. The temporal pattern of mortality responses to air pollution: A multicity assessment of mortality displacement. Epidemiology
13, 87–93. doi:10.1097/00001648- 200201000-00014.
Zheng, Y.,Liu, F., Hsieh, H.P., 2013. U-air: When urban air quality inference meets big data, in: Proceedings of the 19th ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, Association for Computing Machinery, New York, NY, USA. p. 1436–1444. URL:
https://doi.org/10.1145/2487575.2488188, doi:10.1145/2487575.2488188.
Zhuang, Y., Lin, F., Yoo, E.H., Xu, W., 2015. Airsense: A portable context-sensing device for personal air quality monitoring. doi:10.1145/
2757290.2757293.
Page 23 of 23