Modeling and Optimization of

NO2 Stations in the Smart City of

Barcelona
Raquel Soriano-Gonzalez , Xabier A. Martin , Elena Perez-Bernabeu and Patricia Carracedo

Manuel Revez Candeias Vehicle Routing Problems Class

 Table of Contents

01 Introduction
04 Methodology

02 NO2
Management 05 Results

03 Barcelona as a
Case Study 06 Conclusion
 Introduction to Smart Cities

Smart cities integrate social, human,

and infrastructure resources to
enhance sustainability, livability,
and efficiency.

● Contribute to achieving the Sustainable

Development Goals (SDGs).
● Strive for carbon neutrality and better
quality of life.

Barcelona ranked among Europe’s top 5 smart cities in

2022
 Air Quality Challenges

Current Monitoring
Nitrogen Gaps - Barcelona
Dioxide (NO2)
KPIs ● Only 8 monitoring
stations, unevenly
● Produced by diesel distributed.
engines and industrial ● Challenges in capturing
● Air quality KPIs are processes. spatial variability and
particularly important ● Linked to respiratory assessing policies'
for monitoring urban diseases, climate effectiveness.
pollution. change, and acid rain.
● International frameworks ● NO2 levels in Barcelona
guide environmental frequently exceed EU
assessments. limits.
 Study Objectives

Analyze NO2 Predict NO2 Optimize Monitoring

Patterns Concentrations Infrastructure
Explore existing station Use meteorological ● Maximize spatial
data to identify trends data and machine coverage
learning techniques ● Minimize redundancy

Outcome: Comprehensive framework for air quality management and data-driven urban policies.
 NO2 and Urban Pollution – Concerns
and Impacts
Contributes to global warming,
Environmental greenhouse effect, and acid rain.
Effects
Why A key driver of climate change
Focus affecting ecosystems.
on
NO2? Linked to asthma, lung cancer,
Health and respiratory diseases
Impacts
Some studies indicate gender-based
vulnerabilities - women are more susceptible
Urban Challenges:
● NO2 is primarily emitted by diesel engines and industrial activities.
● Social pressures and the global focus on sustainable cities make this a critical issue.
 Smart City Initiatives for NO2
Management

Mitigation Measures: Monitoring and Evaluation:

● Low Emission Zones ● Comprehensive Key

(LEZs) and Performance Indicators
pedestrian-only areas (KPIs) include NO2 levels
reduce traffic emissions. to measure environmental
performance.
● Shared Mobility Systems
(SMSs), such as car-sharing ● Real-time monitoring and
programs, help lower traffic data-driven evaluations
flow. refine urban interventions.
 Innovative Tools for NO2 Monitoring

AI and Machine Case Studies:

Advanced Techniques
Learning Applications
Barcelona: AI analyzed
● Predict NO2 ● Deep Learning
interactions between NO,
concentrations with Models for
NO2, and O3, revealing
high accuracy. short-term
spikes during public
● Develop pollution predictions.
transport strikes.
control strategies ● Hybrid Techniques
and analyze societal incorporate
Madrid: LEZs reduced
impacts. meteorological data
pollution and noise levels,
to link weather
showing the multifaceted
patterns with NO2
benefits of interventions.
concentrations.
.
 Overview of Barcelona’s NO2
Monitoring System
Air Quality Data:

● Open Data Barcelona Air Quality dataset

(2020–2023).
● Hourly measurements of pollutants, focusing on
NO2 (code 8).
● 8 air quality stations providing microgram per
cubic meter (µg/m³) values.

Meteorological Data:

● From 4 meteorological stations over the same

period.
● Gaps: Station X2 lacks comprehensive records;
excluded from analysis.
Figure 1: Spatial distribution of air quality stations
(red) and meteorological stations (green).
 Spatial and Temporal Challenges in NO2
Monitoring
Limitations of the Current Network:

● Uneven distribution of the 8 air quality stations.

● Current station readings often exceed the WHO
limits.
● Bias in station placement:
- Concentration in high-traffic areas over-represents
pollution in these zones.
- Under-represents residential and peripheral areas.

Role of Meteorological Data:

● Factors like wind speed/direction critical for

understanding NO2 dispersion.
● Proximity to sea and altitude variations influence
Figure 1: Spatial distribution of air quality stations
pollutant behavior.
(red) and meteorological stations (green).
 Methodology

3 - Optimizing
2 - Behavior
1 - Descriptive Monitoring
Prediction of
Analysis Station
NO2
Placement

● Explore NO2 behavior ● Predict NO2 ● Maximize spatial

through temporal concentrations using coverage.
patterns and spatial machine learning
● Minimize overlaps.
variations. models.
● Identify factors ● Identify spatial and ● Improve the
influencing NO2 levels. temporal behaviors representation of NO2
and determine the pollution in Barcelona.
best prediction
method,
 Methodology – 1. Descriptive Analysis
Approaches Used

Meteorological Data
Temporal Analysis Spatial Analysis
Integration

Compared NO2 levels Assessed impacts of wind

Examined daily, weekly, across 8 monitoring speed/direction,
and annual variations in temperature, and
stations that highlighted
NO2 levels that identified humidity. And generated
patterns of emissions. gaps in capturing local wind roses to visualize
pollution variability. pollutant flow

K-means algorithm and Elbow method: To group

Additionally Cluster stations based on similar NO2 trends.
techniques were employed: Hierarchical clustering: To visualize relationships
between stations via a dendrogram.
 Methodology – 2. Behavior Prediction of
NO2
K-Nearest Neighbors Averages closest data points; Evaluation using:
effective for non-linear
(KNN) relationships. Holdout method: Dataset split in 70%
training and 30% testing
Transparent, interpretable
Decision Tree rule-based predictions. Metrics:
● Mean Absolute Error (MAE),
Support Vector Predicts data into higher ● Root Mean Squared Error (RMSE)
Regression (SVR) dimensions for precise predictions. ● Squared error - R²

Combines multiple trees to handle

Random Forest complexity and minimize Prediction steps:
overfitting
1. Data Preparation
2. Model Training
Artificial Neural Models non-linear patterns with
3. Validation
Networks (ANN) high flexibility.
Methodology – 3. Optimizing Monitoring
Station Placement
Optimization Approach: It can be seen as a Capacitated Dispersion Problem (CDP)

Algorithms applied
 Methodology – 3. Optimizing Monitoring
Station Placement

Objective functions:
Maximize percentage of area covered. Minimize percentage of overlap between zones.

Scenarios:
● Maximum Coverage Scenario: More stations with higher overlap and therefore enhanced monitoring.
● Budget-Constrained Scenario: Fewer stations with lower cost, but lower and optimized coverage.
 Results - 1. Descriptive analysis

Spatial analysis Weekly analysis

Figure 2: Mean value of the daily concentration Figure 3: Average hourly distribution of NO2
evolution at each station. concentrations by day of the week.
 Results - 1. Descriptive analysis

Figure 4: Annual mean of each station with the WHO

limit value as the 24 h average (yellow) and limit Figure 5: Daily average of NO2 concentrations in the
annual average 2023 (red). period under study.
Results - 1. Descriptive analysis

Figure 6: Wind rose for each station.

 Results - 1. Descriptive analysis

Table 1: Application of the KNN

algorithm to the stations.

a) b)

Figure 7: Cluster in different stations. (a) Air quality stations. (b) Meteorological station.
 Results - 2. Behavior Prediction of NO2
Table 2: Errors of ML models used to predict NO2 concentrations by
station.
Results - 2. Behavior Prediction of NO2

a) b)

Figure 8: Relationship between average NO2 concentration of stations and MAE error. (a)
With all station. (b) Without station 50.
Results - 3. Behavior Prediction of NO2

Figure 9: Pareto Chart

 Results - 3. Behavior Prediction of NO2

Figure 10: Optimal scenario: Final situation solution Figure 11: Minimum scenario: New locations of the
of the problem, maximum number of stations, minimum stations that should be considered as a
occupying the maximum area without overlapping. result of the problem.
 Results - 3. Behavior Prediction of NO2

a) b) c)
Figure 12: Heat map comparison with annual average NO2 concentrations in 2023. (a) Heat map with the initial
stations. (b) Heat map with minimum scenario. (c) Heat map with the 39 stations.

Table 3: Comparison
of the current NO2
concentration KPI
values (μg/m3) with
two scenarios.
 Conclusion

Key findings

NO2 concentration patterns depend on station

location, traffic density, and meteorological
conditions. Challenge:

Station Placement Impact: Achieving accurate KPI requires

strategic station placement.
● High-traffic zones → Overestimate NO2
levels.
● Residential zones → Underestimate urban
pollution.
 Conclusion

Future Directions:
Proposed Actions: Impact:
Integrate real-time
Improved network
Addition of 2 new traffic, NO2 data, and
reveals KPI values
stations. population density.
exceeding WHO
thresholds
Relocate some existing Enable comprehensive
analysis of traffic impact
stations for better
on air quality and public
distribution. Urgent need for precise
health.
monitoring.
Thank you!
If you have any questions, feel free to ask.

CREDITS: This presentation template was created by

Slidesgo, including icons by Flaticon, and
infographics & images by Freepik