Jared Pemberton

← All work

Mapbase

On-demand, high-resolution LiDAR-based terrain.

Role
Founder
Years
2024 to present
Focus
Geospatial data, procedural terrain
Stack
See stack

Problem

Real-world terrain for games, film and visualization starts with GIS work: finding LiDAR coverage, converting point clouds and cleaning rasters. It is slow, manual, and has to happen before any 3D work begins.

How it works

Pick an area on a satellite map and submit. The backend finds the newest USGS 3DEP LiDAR that covers it, streams the points through PDAL, and builds two 1m rasters: a surface model with trees and buildings, and a bare-earth terrain model. A hillshade preview appears in the portal when the job finishes.

Diagram of the working prototype: web portal, FastAPI backend that finds LiDAR, streams points and builds surface and terrain rasters, then the portal preview and Houdini.
The working prototype, end to end.

The portal

The Mapbase portal over the Grand Canyon: a blue selection square on satellite imagery, the job ID, a hillshade preview and the selected region's coordinates.
A Grand Canyon run: selection, job status and preview in one view.

Cleaning the data

Noise returns are dropped, water returns are kept, and any remaining holes are filled, either smoothly or flat at the waterline.

Greyscale hillshade of canyon terrain, with the river filled flat.
Greyscale hillshade of a residential area, showing streets and building pads.
Greyscale hillshade of canyon terrain along a river, with the water filled flat.
Hillshade previews, written for every job.

Into Houdini

The rasters load into Houdini as a heightfield at true scale. Today that handoff is manual.

Three views of a Grand Canyon heightfield in Houdini.
Three views of a Grand Canyon heightfield in Houdini.

Concepts

Brand and portal direction, explored with AI image tools before the build. These are concepts, not product screens. The working prototype portal is shown above.

AI-generated Mapbase brand concept board showing a logo mark, colour palette and type direction.
Brand concept, AI-generated: mark, palette and type direction.
AI-generated concept image of the Mapbase portal interface.
Portal concept, AI-generated. Compare it with the working prototype portal above.

Where it’s going

The plan adds instant pricing, checkout, cloud processing, and add-ons like canopy models and Unreal-ready terrain.

Diagram of the planned MVP: draw an area, instant quote, checkout, cloud processing, then an email with a download link.
Planned MVP: priced by area, processed in the cloud, delivered by email.
Diagram of the planned full pipeline, from web portal through cloud processing and Houdini to packaged delivery.
Planned full pipeline. Not built yet.

Result

  • Working prototype, end to end: map selection to surface and terrain rasters with a preview.
  • 1m rasters over a 1 km² area per job, in under 30 seconds in testing.
  • Tested on Kansas lakeshore and Grand Canyon terrain.
  • Backed by 31 automated tests.

Stack

  • Python
  • FastAPI
  • PDAL
  • GDAL
  • rasterio
  • GeoPandas
  • React
  • Vite
  • MapLibre GL
  • Houdini
  • USGS 3DEP LiDAR

Next

Honor the drawn area exactly, hand finished runs to Houdini automatically, then move processing to a hosted backend.