In 2018, Bloosurf, a premier broadband internet service provider in the mid-Atlantic region, was awarded subsidies from the Connect America Fund (CAF-II) to build out its broadband service in parts of Delaware and Maryland. The CAF-II award requires service providers to make periodic regulatory filings until they are 100% compliant. However, Bloosurf wanted to be 100% compliant as soon as possible to reduce its regulatory obligations and improve operational efficiency. To meet this 100% compliance goal, Bloosurf had to meet Universal Service Administrative Company (USAC)’s regulatory requirements, which included: (a) determining which addresses were tied to which buildings within a parcel and reporting associated lat/longs (“buildingbased geocoding”), and (b) for each address, reporting how many broadband serviceable locations (BSLs) Bloosurf served.
In March 2020, Bloosurf was considering two approaches to obtain this data: (a) sending field staff to survey the locations within its CAF blocks, or (b) conducting a virtual assessment by reviewing aerial imagery to determine BSLs remotely. However, due to the COVID-19 pandemic, Bloosurf decided against option (a) to keep its field staff safe. While option (b) was possible, Bloosurf realized that the task of reviewing the imagery and making determinations of BSLs remotely was significantly more complex than they anticipated. As a result, Bloosurf reached out to geocoding data providers to assist with making this broadband location structure selection.
However, Bloosurf found that most geocoding providers could not reliably achieve building-based geocoding because they did not have a comprehensive and up-to-date map of buildings across the US to use as reference points. Most geocoding providers offer street segment-based or parcel-based geocoding. Figure 1 below shows that the returned set of coordinates (the “geocodes”) can vary significantly based on the underlying data source for the same area. The geocodes are estimated based on street-segment on the far left, and on the far right, geocodes are parcel-based. Additionally, the bottom of the image shows a significant number of missing new construction addresses.
Assigning their addresses to the right buildings was found to be very challenging, especially in cases where multiple structures were broadband serviceable. In addition to the challenge caused by geocoding, Bloosurf also faced a challenge in determining which structures were broadband serviceable, and how many broadband serviceable locations existed within each parcel remotely. For example, what features should they extract from aerial imagery to help determine whether a structure is broadband serviceable or not? How to build a workflow to enable automated feature extraction and determination of BSLs?