In the summer of 2021, a group of Northeastern University students applied their data-science skills to help the community in Portland Maine shed light on a nasty environmental problem. As of fall 2021, the project is ongoing.

• The Greater Portland Council of Governments (GPCOG)
• The Sierra Club
• The Roux Institute

Problem: Residents of Portland and South Portland in Maine have been reporting disruptive smells in the air on certain days of the year. South Portland has over 25,500 residents, and is the host of oil, gasoline, and asphalt tanks. These tanks do not have emissions controls and the emissions, in the form of odors, are reported in both South Portland and Portland. Many of the emissions are odorless and toxic, and may or may not correlate with odor complaints. It's been unclear what the source of the smells might be and difficult to understand the extent of the issue on both sides of the Fore River, which separates Portland and South Portland.

The City of South Portland relies on Smell My City for odor complaints, while Portland relies on See Click Fix. The Department of Environmental Protection has 9 VOC Canister Monitors (volatile organic compounds) in Portland and South Portland. The tanks are considered minor sources of air emissions, and regulated accordingly.

• Coordinate the various data sources
  • Smell My City odor complaints (South Portland)
  • See Click Fix odor complaints (Portland)
  • DEP VOC sampling data
  • Civilian grab canister results
  • Tom Mikulka monitoring data (for comparison only)
  • Weather patterns
  • Tank fills and transfers (dependent on access to data)
  • Individual tank inventory and temperature status
• Clean the data and make it easily accessible for the general public
• Create some simple visualizations
• Add additional data from ME DEP (e.g., wind speed, direction, temperature, etc) Initial mapping will use compiled data, but the intent is to harvest data in real time to account for established correlations, additional sampling, and continued monitoring. Per LD163, additional sampling will include:
• Fenceline monitoring (continuous)
• Self reported emissions testing on heated tanks (2x yearly)

• Maria Guerra (Environmental Health Fellow, Resilience Corps at GPCOG)
• Andrew Butcher (Director of Innovation & Resilience, GPCOG)
• Luke Truman (Sierra Club's Maine Chapter)
  • Luke has experience in data mapping, and working on projects with college students.
  • He has been invested in this problem of Portland/South Portland air quality for some time.

• Excel spreadsheet -- for City of Portland, Maine
  • SeeClickFix app for Portland residents
  • residents add odor reports with the app
  • these reports are collected by the city and sent (to us) monthly in an excel file
• https://smellmycity.org/data
  • This site has directions for accessing South Portland odor complaints
  • Enter ZIP codes of interest (South Portland have several ZIP codes)
  • This produces a spreadsheet for download
• https://rainwise.net/weather/SMRO3
  • This site has DEP weather reports.
  • The top drop-down menu has reports of interest to the project.
  • There are many stations, only a portion of them are online
  • SMRO 3-7 (the units that are online) -- note the location at the bottom of the page.
  • These were installed sometime in 2020 (that's when ME DEP started monitoring South Portland in particular)
  • To download spreadsheets, hit the button at the top of the bar that is a downward arrow.
  • We use data measured by the hour, so future updates require the user to download the weather data in the same time frame. To do this, after clickin te "download data" icon, select "60 minutes" from the interval dropdown from the Ennhanced Download section.
• Oil terminals -- all of this data is in paper form
  • When barges come into the Portland terminal to deliver or pick up oil, they have to give DEP notice
  • There are written notifications from boats to ME DEP -- these are faxed paper copies
  • We hope to be be getting this data...

Summer 2021 term - links to Jupyter notebooks. (Links at the top of these notebooks will open them automatically in Colab).

1. Files for updating data (these notebooks clean and merge raw data, as needed)

• stinky_dataset.ipynb -- Use to update df_stinky, table that combines data from SmellMyCity and SeeClickFix
• oil_vessel_dataset.ipynb -- Use to update df_vessels, table that contains merged, annual data tracking oil vessel passages through Portland

2. Files with students' analyses

• text_analysis.ipynb -- Analyzes the text of smell descriptions and builds visualizations based on these analyses. Includes attempts at building a classifier, which would allow us to tag each row by smell type. Methods and results of each attempt are described in order to help build a better classifier in the future.
  • huggingface_results.ipynb -- Includes results from one of the classifier attempts in text_analysis.ipynb. The model takes ~30 minutes to run; results were downloaded and loaded into huggingface_results.ipynb for easy reference. See "Approach 4: Use Huggingface" in text_analysis.ipynb.
• distributions_allvars_analysis.ipynb -- Includes visualizations regarding the pattern of complaints on a monthly basis and relates it to different factors such as arrival of vessels, type of oil carried by these vessels, average temperature, average wind speed and also distributes the complaints according to the city of Portland and South Portland. Convenience functions are made to make the analysis user friendly as the distributions can be changed for different months and year.
• stinky_maps.ipynb -- Merge the df_Stinky data with weather data and visualise the geographical data on the basis of temperature, wind direction and season. Plotting pods and terminals as well for better understanding. The data is visualised region wise (Portland and South Portland) and weather data is considered from the nearest pod only.

• All raw pod measurements are uploaded to the "stinky/weather_data/pod_raw_data" directory in this repo. Those files are in the form of .csv. To update this repo simply replace the pod files. The replacement file needs to keep the original file naming scheme, every new replacement file needs to to have the pod name as the first 5 characters of the file name.

• All weather intermediary csv/files are present in the stinky/weather_data/weather_intermediary_files. If the raw pod data is updated, please run the weather_datasets.ipynb notebook to regenerate these files. Then update this directory with the new files generated.

1. Files for download (this notebook will autmatically download the weather group's data sets/ all the weather intermediary files - just run to download). This notebook can be used to generate the intermediary files present in the weather_data/weather_intermediary_files directory. Please run this notebook if the any updates are made to the weather_data/pod_raw_data directory. Then use the generated files to replace the files in weather_data/weather_intermediary_files

• weather_datasets.ipynb

2. This notebook contains the visualizations and analyses performed by the weather team - weather feature plots over time, weather and complaint quantity analysis and visualization, logistic modeling, etc. Text descriptions are included. This notebook may take 2-3 minutes to completely run as the visualizations are working with a large amount of data points. Every time this notebook is run any updates to the raw data files will be reflected in the notebook. Be aware that updating the raw files may change the conclusions of the analysis as new data may make our current conclusions obsolete.

• Weather_Team_Analysis_Notebook.ipynb

1. Parent data folders

• smell_data: contains raw data (csv files) of smell data (from SmellMyCity and SeeClickFix), and tidied csv files merging raw smell data and oil vessels data
• vessels_data: contains raw data (csv files) for data tracking oil vessel passages through Portland
• weather_data: contains raw data (csv files) for weather data

2. Raw data folders and files

• smell_data/smell_raw_data
  • scf.csv: it is the original file with odor complaints from SmellMyCity
  • smc.csv: it is the original file with odor complaints from SeeClickFix
• vessels_data/vessels_raw_data
  • 2020 SMRO VESSEL ARRIVALS.csv: it is the original file containing all of oil vessels data for year 2020
  • 2021 SMRO VESSEL ARRIVALS.csv: it is the original file containing all of oil vessels data for year 2021
• weather_data/pod_raw_data
  • SMRO3_2019-07-24_2021-07-24.csv : Raw data from pod SMRO3, replace this file when updating raw files
  • SMRO4_2020-11-14_2021-07-24.csv : Raw data from pod SMRO4, replace this file when updating raw files
  • SMRO5_2021-01-20_2021-07-24.csv : Raw data from pod SMRO5, replace this file when updating raw files
  • SMRO6_2020-11-04_2021-07-24.csv : Raw data from pod SMRO6, replace this file when updating raw files
  • SMRO7_2020-11-04_2021-07-24.csv : Raw data from pod SMRO7, replace this file when updating raw files

3. Intermediary data folders and files

• smell_data/smell_intermediary_files
  • df_stinky.csv: this file is the tidied and merged file containing odor complaints from both SmellMyCity(smc.csv) and SeeCLickFix(scf.csv). This file is used for all other analytical notebooks and further analysis.
  • huggingface_results.csv: results from a model run in text_analysis.ipynb. The model takes ~30 minutes to run; results were downloaded and loaded into huggingface_results.ipynb for easy reference. To update with new data in df_stinky and download: see instructions in text_analysis.ipynb.
• vessels_data/vessels_intermediary_files
  • df_vessels.csv: this file is the tidied and merged file containing oil vessels data for year 2020 (2020 SMRO VESSEL ARRIVALS.csv) and year 2021 (2021 SMRO VESSEL ARRIVALS.csv). This file is used for all other analytical notebooks and further analysis.
• weather_data/weather_intermediary_files
  • df_master_weather : The complilation of raw pod measurement, just concatenating them.
  • df_weather_master_outliers_removed: The same as df_weather_master but SMRO4 data is removed.
  • df_complaints: A dataframe that groups the total complaints by hour and aggregates the complaints by sum.
  • df_merged : A dataframe that groups measurements by the hour and merges the dataframe with the complaints. Weather values are aggregated by average and complaint totals are aggregated by sum.
  • df_model_analysis: a dataframe used for logistic regression analysis