Analyzing customer usage to prioritize global imagery updates
By: Pratik Yadav
Few datasets are as visually appealing and capable of delivering specific context as our imagery layer. Any single image offers a view of a place at a point in time, that can be used to find new development sites, aid in disaster response, serve as a source of ground truth, and monitor changes.
Making sure that view accurately reflects what’s going on is a challenge and opportunity that our Satellite team relishes. In this post, we share how we set priorities for updating a 4 petabyte (PB — or 4M gigabytes) global imagery layer made up of almost a dozen sources.
Imagery Coverage 101
We optimize our basemap for a balance of recency and aesthetic quality depending on data availability. While the layer is called mapbox.satellite, it’s actually made up of imagery sourced from multiple satellites and airplanes, from both commercial and open sources. Imagery sources and resolutions are tied to specific zoom levels to provide consistency and quality of data. They break down as:
- Zoom levels 0-8 use declouded data from NASA MODIS satellites
- Zoom levels 9-12 use NASA/USGS Landsat imagery
- Zoom levels 13+ use a combination of open and commercial sources
All our imagery is checked for color fidelity, optimized, and composited together into a single raster tileset by our Satellite team. The end result is an imagery layer that provides varying levels of detail based on the needs of users looking at larger, less detailed areas or smaller, more detailed areas.
To determine where to update imagery, we take into account industry use cases and needs, customer feedback, and internal data on tile requests and zoom levels.
Imagery Use Cases
When looking to understand our thousands of customers and millions of users, we focus on the broader application of imagery and how to categorize the use case, be it business, scientific, humanitarian, and government or something else.
We segment the uses into two top-level categories — using imagery as visual context, and using imagery as a source for extraction/analysis:
Context — Use imagery as a reference to deliver additional contextual information. This comprises:
- Low-resolution users — Customers who visualize global/country datasets
- Outdoor users — Rural (recreation, non-precision agriculture), non-urban business intelligence, visualization/simulation
- Basemap users — Most app developers, real estate, business intelligence, fleet and logistics management
Extraction/Analysis — Perform work against the imagery; use imagery to derive information. This comprises:
- Analytical users — Hand-tracing image features or running machine learning algorithms for feature extraction
- Precision users — Precision agriculture, construction, auto, extractive industries — users where precision and accuracy are vital to their work
Drilling down on customer usage
Over the past few years, we have been improving our approach to update our imagery basemap based on using anonymized telemetry to identify where people are looking at our maps, and systematically assessing imagery quality at scale.
We combine this with an assessment of aggregated data from our largest satellite users to get a deeper understanding. Looking at the data in the chart below, we see the vast majority of views across our top 100 customers come from zoom levels 17+, or imagery greater than 1-meter resolution (peaks on the right of the charts).
This approach lets us segment our customers based on their usage patterns.
With this information, we can make data-driven decisions on our imagery allocation and forecast which customers will benefit from an update.
Looking ahead
As a result of these analyses, we are focusing the majority of our attention on higher resolution data sets. We now have high-resolution aerial imagery in the contiguous US, covering about 80% of the country’s population. As we move through 2020, we are continually working with our imagery partners to keep the highest demand areas up-to-date with high resolution, high-quality imagery.
We’re also working on some new offerings to give customers greater control of the imagery they can see, which we’re excited to share in the coming months. In the meantime, don’t hesitate to reach out to us here with any questions on our imagery layer.
Pratik Yadav - Developer - Mapbox | LinkedIn
How we update Mapbox Satellite was originally published in Points of interest on Medium, where people are continuing the conversation by highlighting and responding to this story.