Interactive map exploring where glass floats have been found on Block Island during an annual summer scavenger hunt
Every summer since 2012, local glass artists have handmade and hidden roughly 500 clear glass orbs around Block Island, Rhode Island for visitors and residents to find. Each year, the first batch of floats made are brightly colored and decorated. When someone finds a float, they are asked to log the unique float number and the location where it was found on a website. This project is an interactive web map of the locations where these floats have been found on Block Island over the last 11 years to see whether the locations where floats were found have changed between years, and whether the limited number of each years' first run of decorated 'fancy' floats are more commonly found in specific areas of the island.
Maps locations where clear ('regular') and decorated ('fancy') floats have been found on the island over the period the user selects, with dynamic tooltips for more details.
The web map shows where regular and 'fancy' floats have been found on Block Island by using different image icons that are sized in proportion to how many floats were found at that location. Each icon displays a label when clicked that gives more information about the location name and the number of floats found there. You can choose which basemap you want to use - from an artistic background to aerial photos and can also select the year you want to look at, or all years combined.
The map is built using the Leaflet JavaScript library. The underlying data were cleaned and georeferenced using custom Python scripts
To make the interactive map, I used the Leaflet JavaScript library with calls to two different external raster map tile providers to provide the user-selectable background maps. The Leaflet maps are generated from geoJSON files (1 per year) that give the locations where regular and 'fancy' floats were found that year, and also contain the details shown in the tooltips for each float location in the map. The geoJSON files were generated using a custom Python script after applying fuzzy matching to the user-entered location description for each float against a standardized set of location names and coordinates that I tabulated (more detail about this process is below).
I had fun learning about this Block Island tradition, and when I found the list of locations where floats have been found over the years on the Block Island tourism bureau's website, I was super curious to map them. It also provided great experience building an interactive map that contains several different features.
I was intrigued by this project when I first heard about it, because a summer scavenger hunt to get people out and exploring the island seemed like neat idea to me and it also seemed like a mutually helpful link between a local artist who started this project and the local tourism bureau that supports it. I was looking for a way to get experience with Leaflet interactive mapping, and this project provided lots of varied experience with cleaning datasets, outputting geoJSON data, data-driven changes to map icon sizing, dynamic tooltips, and updating the map with user input from sliders, radio buttons, and check boxes.
Ensured the functionality for mapping and fuzzy georeferencing was possible, then built from both ends of the project toward the middle (linked by files in geoJSON format)
I knew I wanted an interactive map so you could choose what data would be mapped, but I also knew I had to break it down into small pieces and build to the final project incrementally. This started with building static maps that allowed custom images and I then added additional layers of functionality on top of that. I knew that I could use geoJSON data reliably to make Leaflet maps, so this allowed me to pursue the data cleaning work simultaneously with proof-of-concept trials to build the interactive functions I wanted for the map.
Parsing the raw location descriptions to get mappable locations for 96% of the recorded floats (this was also one of the main challenges!)
The feature I'm most proud of is the data parsing that enabled the raw input data, with a wide variety of site descriptions where the floats were found, to be standardized and georeferenced to provide the locations rendered in the map. Although this isn't directly visible in the map outcome, it's what allows all the map features to be possible (like showing the different special float icons for year 2020) while confidently showing locations for ~96% of the registered floats based on a description of where they were found.
Parsing the raw location descriptions that were found using an iterative approach with fuzzy matching
The largest hurdle during development was parsing and cleaning the input data to make it ready to map. The input dataset was a list of float identification numbers found by year (each float has a unique number engraved on it, allowing the location where it was found to be documented). However, the description where each float was found is submitted by its finder as a free-text description.
I knew that the map would only be possible if I could reliably parse the majority of these free-text descriptions so I could geocode them to produce a latitude/longitude value for each float, along with a standardized location name and float type (regular or fancy).
Because the glass floats project has been active for 11 years, there are ~3350 records to parse, which is more than I could efficiently do manually. To automate this process, I used fuzzy matching from Python's fuzzywuzzy package to match each location description against a list of curated location names that I created. The fuzzy matching returns a match score for each float location description compared to each location in the curated list.
It was an iterative process to add common locations to the curated location list, run the fuzzy match against all float records, evaluate how well the curated list represented float record descriptions, add more locations to the curated list as needed, and repeating.
When I had added enough locations to the curated list to account for most of the float record descriptions (137 locations including common spelling variations), I looked at the quality of the fuzzy matches produced and their associated scores to set a minimum score threshold for largely correct matches.
Any float with a description that couldn't be reliably matched with at least that score was discarded, but this only applied to 136 records (~4% of total), with many of them being ambiguous descriptions, like 'beach'.
This turned out to be an excellent project for learning a lot of skills related to interactive mapping. It required cleaning and organizing crowd-sourced data, implementing a range of ways users could interact with the mapped data, and managing project development from both the input (data) and output (mapping) ends simultaneously.
Through this project, I learned a lot more about building interactive webmaps, which is what I had wanted. This included having maps be responsive to a range of different user input types, and to access the correct datasets as required to meet the user's specifications.
I also worked with fuzzy string matching and creating geoJSON output files from Python scripts, which were key foundations to providing the data for the map to display.
This also required project management skills to balance the progress of both the mapping and the data cleaning/packaging successfully, building the required proof-of-concepts, and quickly catching and addressing challenges as they arose.
In that sense, I was building from the ends of the project toward the middle, but I knew this would work because geoJSON file standardization ensured that the dataset output would work seamlessly as the required mapping input. I have already leveraged some of these skills through the project mapping crop rotations in Minnesota.