|
Hero Image:
|
Our MotivationNoise pollution can have negative health effects such as stress, sleep disturbance, hearing loss, cardiovascular issues, and even cognitive impairments in children.
Many previous studies have measured noise pollution levels in London over a period of months-years, and have examined its correlation with other factors. We wanted to take the perspective of a pedestrian in London and survey decibel levels in multiple areas across the Bloomsbury and Central London area. |
Timeline
May-June 2024
Research Questions |
How are noise pollution levels distributed across London, and how do our predictive models of this compare to previously established work?
|
Background
Health Effects of Noise Pollution
Noise pollution has been shown to negatively affect both mental and physical wellbeing. Alves et al.’s (2015) paper on low-frequency noise pollution in Portugal emphasizes how these inaudible frequencies caused discomfort in many individuals. This ties in to Coco Khan from The Guardian’s opinion piece (2023) on noise pollution; it’s extremely disruptive to everyday life. While initially it may seem to be a simple annoyance, it can potentially cause long term health effects. Smith et al.’s (2017) published paper in the British Medical Journal on correlations between air/noise pollution and birth weight. While there wasn’t enough evidence to strongly suggest that noise pollution affects birth weight, air pollution did negatively affect it. Additionally, noise pollution and air pollution were shown to correlate strongly with each other. This suggests further research should be done on potential effects of noise pollution. The StressSense (2012) team’s analysis on audial stress detection gave insight into how audial perturbations can cause predictable levels of discomfort. These references raise the potential question of what frequencies are the most harmful, and do different frequencies cause different health effects?
Data Collection Methodology
We need to be considerate of how we collect our data influences the meaning of our results. With any sort of data collection there is always context that is important to properly understanding how to analyze the data. Barrigón Morillas et al. (2014) provides important insight into understanding how the time frame of our sound data sampling can affect the certainty of our results or analysis. Specifically they mention how their long time frame of data directly simulates random sampling of noise levels at a given day. Since we are unable to replicate a whole year of data points we have decided to expand our data collection by taking samples at a far greater variety of locations than possible with immobile collection stations. The hope is that a larger spread of location values will help give our data set more validity. Alberola et al. (2005) discusses how time of day will affect traffic based noise pollution and its general trends over time. Fecht et al. (2016) is a large-scale analysis of noise pollution in London including a detailed methods section which discusses exactly how they collected sound data. Their methods of standardizing every location to being 1 meter from the facade of the building gave us good reference to how particular the location of our sampling ought to be.
Case Studies in Noise Pollution
Quality of life can be impacted by noise pollution in a multitude of ways. Tonne et al. (2018) explores the socio-economic gaps between different income levels and races exposure to pollution. In many cases groups, like lower income households, are more susceptible to living in areas with higher levels of pollution. These case studies show how this may influence certain health conditions caused by pollution to disproportionately affect socio-economic groups. A potential question that these references raise is how do other intersectional factors influence pollution exposure levels?
Noise Pollution Policy
At the end of the day noise pollution in a city is often determined by the government policies that either allow for or reduce noise pollution. The GOV.UK webpage on noise nuisances provides important insight into what laws have already been implemented to deter large amounts of noise pollution. This webpage shows the policy for reporting noise violations and the punishments for breaking them. Without these existing laws and punishments it would be impossible to reduce noise pollution. Doygun et al. (2007) shows us how a city or country may need improvement to policy in order to make significant change. Their analysis of Kahramanmaras, Turkey leads them to the conclusion that there needs to be more adequate laws in order to restrict the most harmful sources of noise pollution. What ways might there be to reduce noise pollution effectively without significant changes to existing laws?
Data Analysis Methodology
Having a wide array of regression models to choose from will lead to a much more informed decision when choosing a regression for our noise pollution model in London. Jovic et al.’s (2015) review on feature selection gives us a wide selection of feature choosing methods which will help us accurately find important features in our data set that have strong correlation to our noise pollution samples. The following list of articles provide us with options or reference to specific regression models and their optimal uses; Goudreau et al. (2014), Rey Gozalo et al. (2016), Barrigón Morillas et al. (2015), Al-Shargabi et al, (2023), and Roth et al. (2004). Any specific regression we end up selecting will be created in JupyterLab (n.d.) using the scikit-learn (n.d.) library for python. Will our methods of data collection or sample size remove some of these models as valid options?
Noise Pollution Analysis Tools
A variety of tools and methods for analyzing noise pollution are available. SoundPLAN is a 3D modeling tool that can be utilized to visually graph, predict, and analyze noise level data in a variety of different environments. It could be useful for representing a city map with noise pollution data. The software Sonic Visualiser is able to analyze the amount of specific frequencies in an audio recording. This could then connect to the SoundPLAN software, with potentially a common frequency map. Morillas et al. (2015) gives a model to predict levels of urban traffic noise using Discrete Fourier Analysis and Fast Fourier Transform, relating to Sonic Visualiser’s Fast Fourier Transform spectrography. These methods and software are able to combine to make clear visual analyses of noise pollution. Could Moriallas et al.’s prediction methodology be utilized within SoundPLAN for consistent noise pollution prediction standards?
Noise pollution has been shown to negatively affect both mental and physical wellbeing. Alves et al.’s (2015) paper on low-frequency noise pollution in Portugal emphasizes how these inaudible frequencies caused discomfort in many individuals. This ties in to Coco Khan from The Guardian’s opinion piece (2023) on noise pollution; it’s extremely disruptive to everyday life. While initially it may seem to be a simple annoyance, it can potentially cause long term health effects. Smith et al.’s (2017) published paper in the British Medical Journal on correlations between air/noise pollution and birth weight. While there wasn’t enough evidence to strongly suggest that noise pollution affects birth weight, air pollution did negatively affect it. Additionally, noise pollution and air pollution were shown to correlate strongly with each other. This suggests further research should be done on potential effects of noise pollution. The StressSense (2012) team’s analysis on audial stress detection gave insight into how audial perturbations can cause predictable levels of discomfort. These references raise the potential question of what frequencies are the most harmful, and do different frequencies cause different health effects?
Data Collection Methodology
We need to be considerate of how we collect our data influences the meaning of our results. With any sort of data collection there is always context that is important to properly understanding how to analyze the data. Barrigón Morillas et al. (2014) provides important insight into understanding how the time frame of our sound data sampling can affect the certainty of our results or analysis. Specifically they mention how their long time frame of data directly simulates random sampling of noise levels at a given day. Since we are unable to replicate a whole year of data points we have decided to expand our data collection by taking samples at a far greater variety of locations than possible with immobile collection stations. The hope is that a larger spread of location values will help give our data set more validity. Alberola et al. (2005) discusses how time of day will affect traffic based noise pollution and its general trends over time. Fecht et al. (2016) is a large-scale analysis of noise pollution in London including a detailed methods section which discusses exactly how they collected sound data. Their methods of standardizing every location to being 1 meter from the facade of the building gave us good reference to how particular the location of our sampling ought to be.
Case Studies in Noise Pollution
Quality of life can be impacted by noise pollution in a multitude of ways. Tonne et al. (2018) explores the socio-economic gaps between different income levels and races exposure to pollution. In many cases groups, like lower income households, are more susceptible to living in areas with higher levels of pollution. These case studies show how this may influence certain health conditions caused by pollution to disproportionately affect socio-economic groups. A potential question that these references raise is how do other intersectional factors influence pollution exposure levels?
Noise Pollution Policy
At the end of the day noise pollution in a city is often determined by the government policies that either allow for or reduce noise pollution. The GOV.UK webpage on noise nuisances provides important insight into what laws have already been implemented to deter large amounts of noise pollution. This webpage shows the policy for reporting noise violations and the punishments for breaking them. Without these existing laws and punishments it would be impossible to reduce noise pollution. Doygun et al. (2007) shows us how a city or country may need improvement to policy in order to make significant change. Their analysis of Kahramanmaras, Turkey leads them to the conclusion that there needs to be more adequate laws in order to restrict the most harmful sources of noise pollution. What ways might there be to reduce noise pollution effectively without significant changes to existing laws?
Data Analysis Methodology
Having a wide array of regression models to choose from will lead to a much more informed decision when choosing a regression for our noise pollution model in London. Jovic et al.’s (2015) review on feature selection gives us a wide selection of feature choosing methods which will help us accurately find important features in our data set that have strong correlation to our noise pollution samples. The following list of articles provide us with options or reference to specific regression models and their optimal uses; Goudreau et al. (2014), Rey Gozalo et al. (2016), Barrigón Morillas et al. (2015), Al-Shargabi et al, (2023), and Roth et al. (2004). Any specific regression we end up selecting will be created in JupyterLab (n.d.) using the scikit-learn (n.d.) library for python. Will our methods of data collection or sample size remove some of these models as valid options?
Noise Pollution Analysis Tools
A variety of tools and methods for analyzing noise pollution are available. SoundPLAN is a 3D modeling tool that can be utilized to visually graph, predict, and analyze noise level data in a variety of different environments. It could be useful for representing a city map with noise pollution data. The software Sonic Visualiser is able to analyze the amount of specific frequencies in an audio recording. This could then connect to the SoundPLAN software, with potentially a common frequency map. Morillas et al. (2015) gives a model to predict levels of urban traffic noise using Discrete Fourier Analysis and Fast Fourier Transform, relating to Sonic Visualiser’s Fast Fourier Transform spectrography. These methods and software are able to combine to make clear visual analyses of noise pollution. Could Moriallas et al.’s prediction methodology be utilized within SoundPLAN for consistent noise pollution prediction standards?
FieldworkMany previous literature works that have researched noise pollution in London utilized static stations that would track noise pollution over a period of multiple years. Due to our limited time-constraint with only being in London for 3 weeks, we decided to take a more mobile tracking approached. A hand-held industrial decibel meter allowed us to be mobile about the city to include many locations in our dataset, while also being an affordable option. This method also more closely simulated the experience of a pedestrian walking through the city.
We set the meter to sample at a rate of once/second, to eliminate chances of sudden noises such as car doors slamming. After 5 seconds we would take a photo of the meter with an iPhone 15 camera. These iPhone photos automatically saved metadata of location (in latitude/longitude), time, and date for each data entry. We saved these photos into a folder to export the metadata automatically into a CSV file with a script; the decibel-level data for each image is extracted manually to avoid any image-analysis errors of what the numbers say. We have not yet fully fleshed-out our data analysis. However, looking at initial data points, they do seem to line up with previous assumptions that busy intersections are some of the loudest points in the city, with some data points pushing over 80dB. One of the biggest realizations we had from conducting this type of fieldwork is how spread-out London actually is. At first we wanted to do many different districts, but realized this might be a difficult feat. To try to mitigate bias in our data, we decided to focus on a more localized area near Bloomsbury and a few touristy parts of Central London to get data points within a closer localized area. |
|
Design Sketches
|
Since we became aware of issues with our data collection methods while doing fieldwork and in the lit review prior we considered alternative methods in our design sketches. One of the proposed ideas was attaching noise level sampling devices to delivery bikes in order to expand our time span of our data set as well as our location spread.
|
|
One of the most helpful tools in hindsight would have been this sort of measurement device that we designed in our sketches. This decibel meter would be capable of taking several measurements throughout a set continuous time span and averaging out the set of measurements to produce one averaged sound level for the given time span. Since our actual collection methods involved taking an instantaneous snapshot of the decibel meter at a random given time it would have been more comprehensive if our data points were instead the averages of larger spans of time.
One of our design sketches is a map that shows visually the relationship between air pollution and noise pollution. We believe that we will find a strong positive correlation between the two variables since generally noise pollutants similarly pollute the air (i.e. traffic, industrial buildings). Similarly in greener spaces with less noise pollution we expect to find less air pollutants. Comparing a mapped visualization of both can demonstrate which areas we might expect to find similarly caused spikes in air and noise pollution.
https://data.london.gov.uk/air-quality/ |
|
One of the most likely negative correlations that we are looking for is between our sampled noise levels and local normalized difference vegetation index (NDVI) levels. We have found a large data set on NDVI throughout the world and now have to do some EDA to determine if there is enough correlation to include the relationship in our regression model. This negative correlation assumption would fit our experience of greener spaces being quieter that we had during our fieldwork.
https://modis-land.gsfc.nasa.gov/vi.html |
|
One of the more likely positive correlations that we are looking for is between our sampled noise levels and local traffic density levels. We have found a large data set on traffic density levels in the UK and now have to do some EDA to determine if there is enough correlation to include the relationship in our regression model. Based off our fieldwork we believe this to be a very likely candidate for positive correlation since we saw in real time how cars passing would affect our samples. https://roadtraffic.dft.gov.uk/downloads |
Conclusions
|
We are still in the earlier stages of our EDA. We are beginning to write a paper that will focus more-so on regression models that are looking to find correlations between variables from different publicly available datasets and our data. This project made us learn that even though we are very passionate about this subject and want to do more in-depth research, it turned out to be extremely hard to execute! We didn't anticipate how much time we needed to properly working models based off of our data.
If time wasn't a constraint, we definitely would've liked to gather more data points in London and potentially interview healthcare professionals on noise-induced hearing loss. We're going to continue refining our research and hope to get our paper to a point where it could potentially be published. References
|
