It’s been a week since I released the geosam package bringing Meta’s SAM3 image segmentation model to R. This week, users have: - Segmented trees in Belgium; - Found paths in rural areas; - Identified boats in harbors; All in R! Try out geosam: walker-data.com/geosam

While some have made the claim that "R is dead!" this year on social media... Claude Code is going mainstream, and guess what language it's using. #LongLiveR #rstats

If you work in geospatial, PMTiles is one of the technologies you need to know going into 2026. This past year, I wrote an R package to help you get started. Learn how here:

Interactive energy infrastructure analytics powered by the {geosam} R package. Find solar farms on a satellite image in West Texas, click points to segment the imagery & extract shapes. Save the result then get back an sf object in a single step with area calculated for you.

Testing the SAM3 model via the new R package **{geosam}** (by @kylewalker.bsky.social) for object segmentation. Automated detection of buildings, boats, and swimming pools in Les Goudes (Marseille) using BD Ortho imagery from @ignfrance.bsky.social Code : gitlab.huma-num.fr/aperis/testi...

I know we used to make fun of Google's AI overviews - but it's pretty impressive how much it's improved in a short period of time. I released this R package Wednesday.

Detecting the letter "O" at @uoregon.bsky.social's athletic fields using Meta's Segment Anything Model 3, Mapbox imagery, and the {geosam} #rstats package. SAM3 does quite well - all you need is the text prompt "the letter o"! #GoDucks

Andrej Karpathy‘s comments on agency matter more than ever in a post-Opus 4.5 world It’s not so much a matter of can you or can’t you anymore It’s now a question of will you or won’t you Agency in today’s labor market is the most important skill you can cultivate

West Texas energy infrastructure identified from satellite imagery using zero-shot image segmentation. Shown: Well pads from simple text prompts; Draw a box around a wind turbine, find the others in the view; Click a solar farm, segment the area.

In the age of LLMs, open source software and their developers are more important than ever. When LLMs write code, they will typically do so with open source tools, which are still maintained by humans. If you use LLMs to code, find ways to support the open source you rely on.

Introducing {geosam}: an R toolkit for Meta's Segment Anything Model 3 - Detect objects in satellite imagery with text prompts - Interactive Shiny tools with draw & click-to-identify - Works with Mapbox, Esri, MapTiler - Use regular images too! walker-data.com/geosam/

The "explore" tool in the new {geosam} R package puts Meta's powerful SAM3 directly in the hands of #rstats users. Shown here: we supply the model six prompts to find in an image; it finds all six within an interactive viewer with pan/zoom. Release coming soon!

Oil & gas infrastructure detection across 270 square miles of SE New Mexico with R, {geosam}, & SAM3. Well pads found from a bounding box and a simple text prompt. It's not perfect, as you'll see false positives - but it's a great starting point for filtering to what you need.

Nearly 800 swimming pools identified from Mapbox satellite imagery with Meta's SAM3 in a section of Beverly Hills. In the soon-to-be-released {geosam} R package: review results by confidence level, filter, then export. This is shaping into a tool for serious applied work.

Here's the number one reason why I'm bullish on AI, especially in the post-Opus 4.5 world. If you are in any sort of creative field, you know the rush of dopamine you get when you start a new project. The thrill of discovering new things, of creating something novel.

Draw a polygon around a sand trap in a satellite image, find all of the other sand traps in the view. There are tons of applied use-cases for this tool, but it's also just fun to play around with.

Today's work on bringing Meta's Segment Anything Model 3 to #rstats: An interactive Shiny viewer using @maplibre.org to show and browse regular images. Here we detect: - the basketball - the Mavs player (black uniform) then the Rockets (in red) - the American flag on the backboard

#r5rgui has been on CRAN for a few days now! Your handy helper to quickly visually explore routes built by #rstats #r5r (@urbandemog.bsky.social), powered by #mapgl (@kylewalker.bsky.social). Learn more at www.ekotov.pro/r5rgui/

Today's update on the upcoming R interface to Meta's SAM3: We'll have a Shiny gadget that allows you to interactively explore and segment imagery. Shown here in Positron: finding red cars in a parking lot at TCU via Mapbox imagery, which are returned to your R session as an sf object.

It had to be done. Geospatial tools for @Meta's Segment Anything Model 3 are coming to R. Shown: @TCU's football stadium as an sf polygon. Extracted with SAM3 from the second image with the prompt "football stadium". Keep an eye out here for updates and a package release.

A trick I like to use in my interactive mapgl work: `hash = TRUE` Add that argument to your regular or `_view()` map calls and you'll add the center, zoom, pitch, and bearing to the URL hash. It's very useful for remembering specific views / areas I want to highlight.

The new SAM3 model from Meta is blowing my mind Shown here: detecting putting greens, pools, and cars in Scottsdale from simple text prompts via Mapbox imagery R, Shiny, mapgl for the UI; Python backend via @giswqs.bsky.social's segment-geospatial package (thanks Qiusheng!)

Identifying well pads in the Permian Basin using Meta's new Segment Anything Model 3. Draw a shape on @mapbox.com satellite imagery; SAM3 identifies all similar objects in the view. There's so much going on in AI and geospatial right now beyond just LLMs.

The @Mapbox Standard style has a lot of powerful features for customizing your maps - and it's available to you in #rstats with {mapgl}! Example: how to use `set_config_property()` to change the light preset between "dawn", "day", "dusk", and "night" Try out the code below:

This is the most exciting time ever to be working in data, and I'm not talking about AI. 3 years ago, I wrote a database-centric guide in my book for analyzing the full 92 million record 1910 Census. Now, with #rstats and @duckdb? Analyze those 92 million rows in seconds.

If you work in geospatial, it's time to take a hard look at your value proposition. In a single prompt, GenAI tools can make interactive maps that used to require specialized expertise.

A mapgl trick you can try in your Shiny apps: Use an action button to toggle between a 2D and 3D map view on your @mapbox.com map; Use `set_paint_property()` to make the transition seamless; And use `ease_to()` to tilt your map. Try it out: gist.github.com/walkerke/33d...

Mapped every investor mortgage in the US in 2024 (HMDA data). Draw on any market --> get instant stats. Areas that stand out: - Exurbs (north of Dallas) - Vacation areas (Florida 30A) - Low-income urban (Memphis) Interested in this data? Send me a note.

There is a very broad spectrum between “vibe coding” and “traditional software engineering.” Those who find the sweet spot in the middle stand to profit immensely.

When I say Claude Opus 4.5 is next-level, this is what I mean: - I tell Opus how to complete a task with one of my R packages; - It calls R to inspect a dataset; - It then applies that information correctly (shown: mapping PLSS sections) Opus feels like a true dev partner!

Urban growth in China since 1975, projected to 2050 Explore the newly-updated "World City Populations" by @citygeographics.bsky.social, one of my all-time favorite maps: luminocity3d.org/WorldCity/#5...

Last day of #30DayMapChallenge: Makeover Visualizing daytime Tokyo vs. 2125 cyberpunk Tokyo at night on a 3D @mapbox.com map with a comparison slider. In #rstats mapgl, the comparison map is built with a single line of code - just call `compare()` over two maps.

I’ve been thinking a lot recently about AI and the 40 or 50 year-old entrepreneur. Typically, we associate startups with ambitious people in their 20s, but I think AI changes that calculation.

In R's mapgl package, use `add_image_source()` to add a terra raster to your map, then display it with `add_raster_layer()`. Drape it over Mapbox's Standard style for built-in 3D! #30DayMapChallenge Day 29: Raster

This week I've used the #30DayMapChallenge to showcase the mapgl #rstats package: - H3 hexagonal binning of 70k US wind turbines with "fade effects"; - The US electricity transmission network visualized with "glow" effects; - A spinnable globe of the world's time zones

#30DayMapChallenge Day 28: Black In the @Mapbox Standard Style, set the `lightPreset` in the style's basemap config property to "night" for a stunning dark theme. Try it out with the Standard Satellite style as well!

The world's time zones visualized on a spinnable, interactive globe. Try out the code - it'll default to your current time zone and show you times around the world! #30DayMapChallenge Day 27: Boundaries Code: gist.github.com/walkerke/c4a...

Visualizing US electric power transmission lines for Day 26 of #30DayMapChallenge: Transport A few mapgl tips in this one: - To achieve a "glow effect" for the lines, first add a blurred line using the `line_blur` parameter then add the same layer over it as an opaque line

I've been working with Claude Opus 4.5 for a day now... ...and it's causing me to question a lot of my assumptions about what's possible with AI. If you have an idea but you have been waiting to build it, now is absolutely the time to get moving on it

Video tutorial: mapping wind turbines with H3 hexagons, #rstats, and @maplibre.org riverside.fm/shared/expor... Code / links on the next post

Mapping US wind energy infrastructure with H3 hexagons for #30DayMapChallenge Day 25. Turbines and their capacity are aggregated by H3 level; we apply a "fade effect" as the user zooms to reveal smaller hexagons then individual turbines. Posting a tutorial later today!

Over the past 2.5 years I've been hosting Census data, web mapping, and Shiny workshops. I'm giving my biggest discount ever on all the recordings + tutorials this week - I appreciate you checking them out! This year: walkerdata.gumroad.c... Past years: walkerdata.gumroad.c...

For Day 24 of #30DayMapChallenge: Places and their names, a labeling trick in mapgl you can try. Use the TurfJS integration in the package to dynamically compute centroids for your layer then use them as a symbol layer to label your polygons. Try it: gist.github.com/walkerke/b77...

The story map feature in the #rstats mapgl package is a tremendous teaching tool. Teach your students geographic concepts while visualizing those concepts in real-time. Check out this Census mapping story map tutorial for #30DayMapChallenge Day 23: Process.

My career advice for 2026. You probably have a skill or idea sitting on your computer that solves someone’s exact problem, but no one knows about it. You are one post about this from changing your life.

For Day 22 of #30DayMapChallenge, Natural Earth, a shout-out to Martijn Tennekes and the tmap.mapgl package. tmap.mapgl wraps mapgl, allowing you to make performant interactive maps with the same tmap syntax you know and love! Try it out: r-tmap.github.io/tmap.mapgl/i...

Are you intrigued by the latest innovations in open-source geospatial, but you're committed to your Esri #GIS stack? Not to worry. You can have both. Shown here: 5 million US oil & gas wells, stored as PMTiles, served as a ZXY endpoint and consumed by the ArcGIS JavaScript API.

A new version (0.4.3) of the mapgl #rstats package, your R interface to @mapbox.com and @maplibre.org maps, is now on CRAN! Mostly quality-of-life improvements in this release; the biggest change is that MapLibre maps now default to a globe projection. Install / update today!

Today's mapgl tip: use `add_image()` to add a local or remote image to your map, then pass the image ID to `icon_image` in `add_symbol_layer()` to use custom icons on your maps. You can even cluster your icons too! #30DayMapChallenge Day 21: Icons

If you have maritime mapping needs, consider using the beautiful Ocean map style from @maptiler.bsky.social. It's available in mapgl with `maptiler_style("ocean")`. #30DayMapChallenge Day 20: Water