Compiled by Matt Forrest, this year’s edition features almost 2,500 companies and organisations from 121 countries, all neatly categorised and searchable by organisation size, founding year and funding received. If you’re looking for a job in geospatial, start your search here.
GeoViz is a new JavaScript library by Nicolas Lambert to produce maps on the web. It comes with a wide range of resizable markers and symbols, contour lines, geometry simplifications, map projections, layout marks (yeah, you can have that north arrow), and so much more. Built on D3, the syntax is similar and compatible, making it a flexible tool for serious thematic mapping on the Web.
In the geospatial world, like everywhere else, AI is the hottest topic. Anyone who’s looked at a map in the last five years is now talking about foundational models and geospatial embeddings. Some (probably many)—myself included—still struggle to get their heads around those concepts.
Kiri Carini:
I get that end users (which is where I’d put myself on the expertise spectrum) don’t need to understand all the ins and outs of how embeddings are generated. But if we’re going to use them beyond novelty or experimentation, we need to understand enough to know what level of confidence and trust to place in them when applying them to real work.
Kiri’s list is an excellent starting point to learn about geospatial foundational models and embeddings.
The premise of TerraBit is simple: Click on any area around the globe and return similar locations. We’ve seen solutions to this problem before, NASA’s (currently broken) Similarity Search springs to mind. But TerraBit is different because the computation happens entirely in the browser using modern, cloud-native technologies.
The Clay v1.5 corpus is binarised, packaged as a partitioned GeoParquet and uploaded to a public object storage. The browser queries the data using DuckDB-WASM. The comparison of binary vectors then happens on the GPU.
Isaac Corley and Caleb Robinson have more detail:
The 50M embeddings are partitioned into geohash-aligned Parquet shards and published on Source Cooperative, which serves them cloud-natively out of S3 — public HTTP with byte-range support, no egress fees, no intermediate server. A single manifest file records the path, row count, and spatial extent of every shard.
When you draw an ROI, DuckDB-WASM queries the manifest with a bounding-box predicate — manifest-based shard pruning: the manifest acts as a coarse spatial index so the browser never opens metadata on shards outside the ROI. Once the intersecting shard list is resolved, DuckDB streams those shard files over HTTP (via httpfs range requests) and applies a second filter at the row level — a bbox predicate for rectangles, or
ST_Intersectsfor freehand polygons — to extract only patches within the drawn region. Ranking over the candidate slice is exact brute-force Hamming: binary embeddings arrive as packedUint8Arraycolumns (128 bytes per 1024-dim vector) and are scored in a Web Worker via XOR+popcount, which maps directly to hardware-accelerated popcount instructions and completes in milliseconds for a typical AOI partition.
New results appear instantly, once the data is downloaded for the specified area of interest. Not long ago, this needed a round-trip to a back-end, now it runs entirely in your browser.
Waddah Hago:
Consider a simple example. Flat concrete roofs are common throughout the Caribbean because they are designed to withstand hurricane-force winds. Yet AI models trained largely on North American imagery have sometimes classified these homes as parking lots because they do not resemble the residential structures the model learned to recognize. The model isn’t broken. It simply learned from a dataset where homes generally have pitched roofs, asphalt shingles, and suburban street layouts. When confronted with something different, it makes the best prediction it can based on what it knows.
The same problem appears across much of the world. A neighborhood in Lilongwe doesn’t look like a suburb in Virginia. Traditional homes in Malawi don’t resemble the structures that dominate most Western training datasets. Informal settlements in Abuja don’t follow the same patterns as planned communities in Arizona. The buildings are different. The materials are different. The settlement patterns, densities, vegetation, and infrastructure are different. When AI encounters environments it has never seen before, performance often begins to deteriorate.
Without local knowledge, you cannot produce an accurate map. This is a problem now, as it was in the days of ERDAS Imagine, when we still called it image classification rather than AI. What has changed is how we view AI output today. The marketing of leading AI companies and a litany of one-sentence-per-paragraph influencer stories on LinkedIn suggest that the models get better and better — which they do — and that they never fail; let’s go all in, trust the output and move on.
One of my first jobs was to sit in an office, stare at a screen for eight hours, draw rectangles over satellite imagery of Hungary, and classify each rectangle. We had training and extensive examples to draw from. But in the end, a team still went to Hungary to verify on the ground what we mapped. If you can’t trust a human to make correct assumptions about the state of the environment in a remote location, how can you trust the output of a statistical model when you don’t know what data it has been trained on?
I’m not arguing not to use AI at all. These tools are efficient and useful. But the model outputs include errors, and cognitive surrender is real. You need a human to control and own the output, with all its consequences. The model output is your output. If the model output is wrong, then you are.
Overture has switched the tool they use to create pmtiles from tippecanoe to planetiler. A change that yields impressive results.
Brandon Liu:
As Overture has grown, these tilesets have grown in size and take longer to generate. Prior to the April release, the tilesets for the places and divisions themes were created using the tippecanoe tool.
Tippecanoe is a powerful C++ command line utility for creating tilesets. Tippecanoe’s algorithms adaptively create low-zoom tiles by processing all features. This makes it convenient for visualizing data without configuration, but means that the large themes like Places took over one hour to generate.
From the April release onward, tilesets are created with Java code using the planetiler library. Instead of considering all features at each zoom, features are filtered to zoom levels by type. This requires more explicit feature selection, but means tilesets can be generated in a fraction of the time. […]
The Places theme tileset creation went from over 2 hours to ~5 minutes. The Divisions tileset creation went from 48 minutes to under 20 minutes.
On top of that, they changed how they manage their tile-generation infrastructure and open-sourced it. Using terraform-aws-overture-tiles, you can now create customised tile sets from Overture data and even enhance it with other data.
Åsa Forsell:
Developers can now add detailed indoor airport maps to their application using the new Mapbox airport indoor maps tileset. Designed to seamlessly connect the outdoors and indoors, the tileset helps guide travelers and airport staff through more than 200+ international airports.
The airport indoor maps tileset can be used on the web and mobile SDKs with Mapbox Standard Style, and the tileset can be enabled in Mapbox Studio or via code. Pre-styled indoor layers and UI components are ready to use, reducing the time and resources required to design and maintain indoor maps from scratch. Teams that need more control can customize map styling and interactions to match the specific needs of their product and users. They can also add their own custom layers to show their own data on the indoor maps.
There’s a long list of potential use cases, including retailers allowing people to find their shops and airports managing logistics behind the scenes. Though airlines guiding passengers through busy airports, helping them find gates, lounges or help desks, immediately springs to mind.
Airlines delay flights for lost or late passengers and manage complaints from loyalty members who cannot find amenities.
I don’t think airlines delay many flights because passengers get lost finding the correct gate. Most late passengers get lost dawdling through the duty-free shop buying things they might as well get at their destination and then proudly strolling into the aeroplane with their five bags. Not all problems can be solved with better maps.
From a passenger perspective, the most convenient solution would be to integrate airport maps into Google Maps or Apple Maps (both of which do), which most people already use for day-to-day navigation. The second best place will be airline apps. I don’t want to download a separate app for every airport I pass through.
Cloud-native GIS platform GeoLibre has a new release out.
Qiusheng Wu, on LinkedIn:
What’s new in v1.9.0
- CAD import: Add CAD drawings (DXF/DWG) as a layer, with a picker for choosing which drawing layers to load and a CRS selector for placing them correctly on the map.
- Smarter service discovery: WMS and WFS panels now read the service’s GetCapabilities, so you pick available layers and feature types from a populated dropdown instead of typing names by hand.
- Docked SQL Workspace: The SQL Workspace now docks as a resizable panel beside the map, with editor autocomplete for tables, columns, and SQL keywords.
- Generic Vector to Vector conversion: Convert between any supported vector formats by file extension, alongside the existing targeted converters.
- Richer camera tours: Per-keyframe hold and transition duration controls for finer pacing, plus save and reload of a named tour setup.
- Better story maps: A hide-itinerary toggle, subtitle and byline fields on the printable handout, and dedicated start and closing slides.
- Styling and plugin extras: A transparent (no fill / no outline) option in the color picker, a Legend populated from a paletted raster’s color table, and plugins can now use the maplibre-gl-raster stack and the projection control.
Overture uses data from user interactions with locations to compute a signal indicating whether a business is still operating and update its business data.
Albi Wiedersberg on the Overture Maps blog:
Instead of relying solely on user feedback or web scraping, we leverage real-world data from Overture members who regularly interact with businesses. Meta, TomTom, Tripadvisor, and Uber currently contribute the aggregated, anonymized “heartbeat” signals behind our monthly updates — and because each one sees the world differently, together they catch changes that any individual company would miss.
These signals come from securely combined, anonymized data: foot traffic at business locations, online and offline interactions, and similar activity. A pulse of activity means a business is alive and operating. Collaborating on this mechanism means we can reduce redundant efforts to keep track of this basic but highly volatile data set.
I’d love to see a more technical breakdown of how this works, what data flows in and how they compute the signal.
Two new editions of Satsummit have just been announced: St. Louis will host the conference on 18 and 19 November, followed by Lisbon next year on 16 and 17 March.
As usual, the conferences cover a range of earth-observation-related topics; this time with special emphasis on the effects of AI on the landscape. From the newsletter (unfortunately, there’s no public permalink available):
The satellite and geospatial sectors are entering a period of rapid change. Artificial intelligence is reshaping how Earth observation data is analyzed and applied. Commercial and government priorities are increasingly intertwined. Questions around access, competition, ethics, funding, and public benefit are becoming more urgent than ever.
Topics will include:
- The role of satellite data in humanitarian response, conservation, and climate resilience
- Funding shifts and their impact on open science and mission-driven work
- Sustaining community resources as funding landscapes shift
- The dual-use dilemma and when satellite imagery serves both humanitarian and military ends
- How humanitarian and climate organizations are doing mission-critical work with less
- GeoAI in practice: what works, what’s hype, and who benefits
- Commercial imagery, government pressure and the price of access