Skip to content

6. Rendering Engines, 3D, and Time

2:00 - 3:00 p.m.

Lecture 15 min · Hands-on 35 min · Break 2:50 - 3:00

Recording of section 6: Rendering Engines, 3D, and Time

Learning goals

By the end of this section you will be able to:

  • Explain why one GeoLibre project can be drawn by four different rendering engines, and pick the right one for a task.
  • Stream a 20-billion-point USGS 3DEP LiDAR survey of Reno in the browser and read it by elevation and by ASPRS classification.
  • Turn on 3D terrain, drive it from a Cloud-Optimized GeoTIFF, and choose an honest vertical exaggeration.
  • Open the same map on the Cesium 3D globe and come back without losing anything.
  • Animate ten years of earthquakes with the Time Slider, and tell it apart from the globe's scene clock.
  • Save views as bookmarks and measure true distance, heading, and terrain profile.

Lecture (15 min)

There is no separate 3D view

Desktop GIS usually keeps 3D in a second window: the QGIS 3D map view, an ArcGIS Pro scene beside a map. GeoLibre has no such split. Tilt the map and the 3D content already in the layer stack is there: terrain, extruded buildings, point clouds, 3D Tiles. The camera has five numbers (center, zoom, bearing, and pitch, up to 85°), and the status bar shows all of them.

GeoLibre goes one step further and separates the project from the thing that paints it. The project store holds plain layer records, styles, and a camera, not MapLibre sources or Cesium primitives. Each rendering engine subscribes to the same store, so switching engines is not a conversion: layers, visibility, opacity, most styling, and the camera simply carry over.

Engine Best suited to Credentials
MapLibre GL JS (default) The widest tool and plugin support: vector styles, rasters, terrain, lightweight 3D None
CesiumJS A true 3D globe: curvature at every scale, terrain as a surface, 3D Tiles, CZML, streamed point clouds None required; a Cesium ion token adds Cesium World Terrain and ion assets
Mapbox GL JS Mapbox styles and services, Mapbox Standard A Mapbox access token
ArcGIS Maps SDK Native ArcGIS services, Esri basemaps, projections Web Mercator cannot draw None; an ArcGIS API key adds Esri basemap styles

Switch with View → Rendering engine. Rules of thumb:

  • Start on MapLibre. Every exercise in this workshop runs there.
  • Use Cesium when the third dimension or time is the point: terrain-heavy scenes, photogrammetry, orbits and trajectories, continental extents where Mercator distortion misleads.
  • Use ArcGIS when your data lives in ArcGIS services or the projection is the message (Equal Earth, Spilhaus).
  • Use Mapbox when a Mapbox style is the deliverable.

Capability gating keeps this honest. A command the active engine cannot run is greyed out with a reason, a layer it cannot draw stays in the project with a tag such as 2D only, and a plugin that does not support the engine is suspended rather than half-mounted. If a tool is enabled, it works there.

Coming from QGIS or ArcGIS Pro

Think of the engine as the canvas, not the project. Switching to Cesium is like opening the QGIS 3D view, except the globe is a full workspace with its own tools, and the switch is saved with the project.

Point clouds: the COG idea, in 3D

A LiDAR survey of a metro area holds billions of points. The formats differ mainly in whether they can be streamed:

Format What it is How GeoLibre reads it
LAS ASPRS interchange format, uncompressed Downloaded whole
LAZ LASzip-compressed LAS, about a quarter of the size Downloaded whole
COPC (.copc.laz) One LAZ file with points in an octree and an index at the front Streamed by view and level of detail
EPT (ept.json) An octree of many small files described by a manifest Streamed by view and level of detail

The distinction is the same one that separates a plain GeoTIFF from a COG. A COPC or EPT survey opens at full extent with a coarse sample and fetches finer octree nodes only where you zoom, up to a point budget. The USGS publishes its whole 3DEP LiDAR program as public EPT on Amazon S3. If you publish point clouds yourself, publish COPC (PDAL's writers.copc converts LAS and LAZ).

Classified LiDAR assigns every point an ASPRS class code:

Code Class Code Class
0 Never classified 6 Building
1 Unclassified 7 Low point (noise)
2 Ground 9 Water
3, 4, 5 Low, medium, high vegetation 17 Bridge deck
18 High noise

Ground points (class 2) build bare-earth DEMs like the Reno-Tahoe DEM you styled in Section 3. How far a survey goes beyond ground depends on its contract, as Exercise 6.2 shows.

Terrain, and the honesty rule

Controls → Terrain drapes the basemap and your layers over keyless global elevation tiles. Double-click the terrain button to set the vertical exaggeration or to swap the source for a single-band COG DEM (EPSG:3857 or EPSG:4326) read in small HTTP range requests, so a 1 m LiDAR DEM can shape the scene instead of a global 30 m model.

Exaggeration is a presentation decision with no correct value. At 1× most landscapes look flatter than they feel; at 1.5-2× subtle relief becomes legible; past 3× you are making a graphic, not a map. State the exaggeration whenever you publish a 3D image. A reader cannot infer it.

Two models of time

Time Slider Cesium scene clock
Animates Layers: shows features whose date falls in the current step, or a cube's time slice Entities: positions and appearance are functions of time (CZML)
Use it for Tables of dated observations: earthquakes, permits, survey epochs, Zarr or NetCDF time axes Moving things: satellites, flights, vehicle tracks
Engines MapLibre and Mapbox (suspended when Cesium is the primary engine) Cesium

You do not import data into a special time-aware layer: you add layers normally and bind them to the shared timeline. The time window is a filter like any other, so it combines with quick filters and expression filters.

Small tools that carry context

  • Bookmarks (Controls → Bookmark) save the camera and the visible layers, in folders, inside the project, so they travel with a shared link.
  • Measure (Controls → Measure) reports distance, area, or a circle, plus the true great-circle heading and, when elevation is available, slope-following Terrain (3D) distance with gain and loss. It measures on a sphere, which is fine for a quick look. For numbers in a report use the field calculator or the processing tools, which measure on the WGS 84 ellipsoid: Measure to look, the field calculator to record.

Hands-on (35 min)

Start a new project: Project → New..., name it NGIS 2026 - 6 3D and time, pick the Liberty 3D basemap, and click Create. Liberty 3D is the Liberty style with the camera tilted into a 3D view; its buildings rise in 3D when you zoom in to a city.

Exercise 6.1: Stream the USGS Reno-Carson survey (5 min)

  1. Choose Add data → LiDAR layer. The Add LiDAR layer panel opens on the left of the map.
  2. Under Load COPC or EPT from URL, paste the USGS 3DEP Reno-Carson 2017 (QL1) survey and click Load:

    https://s3-us-west-2.amazonaws.com/usgs-lidar-public/USGS_LPC_NV_Reno_Carson_QL1_2017_LAS_2018/ept.json
    
  3. The map flies to the survey (zoom about 9). It runs from north of Reno past Carson City to the east shore of Lake Tahoe, plus a block east of Carson City. Points are colored by Elevation with Viridis; the ramp reads 1260 to 3669 (meters).

  4. Scroll the panel to Loaded Point Clouds: the survey reports 20,286,530,383 points. The browser is drawing a sample of a few million and refining it as you zoom.
  5. Notice that Z Offset is ticked with an offset of -1260 m, the survey's lowest elevation. That lowers the cloud so it sits on the flat map instead of floating a kilometer above it.

Discussion

A 20-billion-point survey appeared in a few seconds, and the Layers panel has one new row tagged LIDAR. What did the browser actually download to draw this view, and what would a plain LAZ download of the same survey cost?

Exercise 6.2: Read the classification (6 min)

  1. In the panel, under Styling, set Color By to Classification. A class list appears with Show All and Hide All buttons.
  2. Read the list. This survey lists Never Classified, Unclassified, Ground, Low Point (Noise), Water, Bridge Deck, High Noise and two extra codes. There is no Building or Vegetation class: the QL1 contract classified ground and water, and left everything above ground as Unclassified.
  3. At this regional view much of the map is magenta. Click Hide All, then tick High Noise alone: those magenta points are class 18, high noise (birds, haze, multipath returns) that a coarse sample over-represents.
  4. Untick High Noise and tick Ground alone. The brown ground points show holes wherever buildings and trees stood, because those returns are in class 1, not class 2.

    Only the ground class of the Reno-Carson survey, with gaps where buildings and trees stood

  5. Click Show All and set Color By back to Elevation.

Click the label, not the box

If a class checkbox does not respond, click the class name next to it.

Exercise 6.3: Tilt the camera and add terrain from a COG (6 min)

  1. Choose View → Set view... and enter Longitude -119.80, Latitude 39.25, Zoom 10.2, Pitch 55, Bearing -15. Click Go. The status bar reads Pitch: 55.0° and Bearing: -15.0°, and the survey now recedes toward Reno.
  2. Close the LiDAR panel (×). Choose Controls → Terrain. A mountain button joins the map controls on the right, already switched on, and the basemap drapes over global elevation tiles.
  3. Double-click the mountain button. The Terrain exaggeration dialog opens.
  4. Set Vertical exaggeration to 2. In Terrain source, paste the workshop DEM, click Use COG DEM, then Done:

    https://data.source.coop/opengeos/data/workshops/ngis-2026/reno_tahoe_dem.tif
    
  5. The terrain is now built from the 30 m Reno-Tahoe COG, fetched tile by tile in range requests, at 2×. Drag with the right mouse button (or Ctrl+drag) to change pitch and bearing and look along the Carson Range. Click the compass to reset north.

Terrain travels with the project

Saving the project records that terrain is on, the exaggeration, and the COG source URL, so the finished project reopens at 2× on the workshop DEM. A DEM opened from a local file has no URL to save, so it falls back to global terrain. If you publish a 3D screenshot, write the exaggeration in the caption.

Exercise 6.4: The same map on the Cesium globe (5 min)

  1. Choose View → Rendering engine → Cesium. The first switch loads the Cesium engine, which takes a few seconds.
  2. The camera and the LiDAR layer carry over. The satellite imagery can take 20 to 30 seconds to appear on the first switch, so the cloud may float over a starry background at first. The survey is streamed natively by Cesium onto a satellite-imagery globe with real terrain, so the Carson Range and the Lake Tahoe basin rise without any exaggeration setting.

    The Reno-Carson point cloud streamed onto the Cesium globe

  3. Look at the globe's own controls at the top right: Reset view (home), a scene-mode picker (3D globe, 2D map, Columbus view), a Basemap picker, and fullscreen. Try Columbus view, then go back to 3D globe. Each scene-mode switch zooms out to the whole Earth; click Zoom to layer on the LiDAR row to fly back to Reno rather than scrolling in from space.

  4. A note at the bottom right suggests adding a Cesium Ion token. It is optional: the globe works without one.
  5. Choose View → Rendering engine → MapLibre. Everything is where you left it, except that the LiDAR panel's Color By is back to Elevation.

Expected Diagnostics warnings

The Diagnostics counter in the status bar climbs by about ten on the first Cesium switch. They are deprecation warnings from the Cesium library ("... has been moved from cesium/engine to cesium/core") and are harmless.

Exercise 6.5: Ten years of earthquakes in time (8 min)

  1. Add data → Vector layer, paste the GeoParquet earthquake file, and click Load:

    https://data.source.coop/opengeos/data/workshops/ngis-2026/nv_earthquakes.parquet
    
  2. Close the panel and rename the layer Nevada earthquakes (double-click its name). Hide the LiDAR layer (eye icon), and use View → Set view... with Longitude -117.6, Latitude 38.9, Zoom 6.2, Pitch 0, Bearing 0 to see all of Nevada.

  3. On the Nevada earthquakes row, open Layer actions (⋯) → Bind to time slider.... Time property is already time; leave Show features at In the current step and click Bind.
  4. A timeline docks below the map, spanning 2016 to 2025 with the Y (year) step selected. Click 2020. Two clusters dominate: the Monte Cristo Range sequence in west-central Nevada and the tail of the 2019 Ridgecrest sequence in California.

    The Time Slider bound to the earthquake layer, showing 2020

  5. Click play (▶) and watch the decade year by year. Use these counts to check what you see:

    Year Events Largest Year Events Largest
    2016 349 M5.6 2021 756 M6.0
    2017 286 M4.1 2022 381 M4.8
    2018 158 M4.5 2023 292 M4.5
    2019 3,070 M7.1 2024 444 M5.7
    2020 1,871 M6.5 2025 424 M4.8
  6. Pause, click 2020 again, then switch the step to M (month). The marker reads Jan 2020, and Next (⏭) now moves one month at a time. The filter does not shrink with it: In the current step means the step that was active when you bound the layer (a year), so each click slides a 12-month window. Click Next four times to reach May 2020. The map shows May 2020 through April 2021 (1,851 events), not May alone (811 events: the Monte Cristo Range M6.5 of May 15 and its aftershocks). Then click Previous (⏮) ten times to reach Jul 2019: the window holds 4,274 events, and 2,486 of them fall in July 2019 alone, the busiest month of the decade (Ridgecrest, M7.1 on July 6).

Discussion

2019 has nearly twenty times the events of 2018 (3,070 against 158), yet most of them are not in Nevada. What would you change, with what you learned in Section 4, so the animation shows only Nevada? And which Show features option would turn this into a map of where earthquakes have accumulated?

Exercise 6.6: Bookmark a view and measure it (5 min)

  1. Choose View → Set view... with Longitude -119.85, Latitude 39.2, Zoom 8.8, Pitch 0, Bearing 0.
  2. Choose Controls → Bookmark. Type Reno to Mount Rose in Bookmark name..., keep Include complete layer state ticked, and click +.
  3. Zoom far out, then click the bookmark name. The map returns to zoom 8.80 with the same layers visible. Close the Bookmarks panel.
  4. Choose Controls → Measure; Distance is already selected. Click the University of Nevada, Reno (the venue, just north of downtown), then double-click the summit of Mount Rose to finish. The peak is not labelled at this zoom: it sits about two-thirds of the way from Reno toward the north shore of Lake Tahoe, northeast of Incline Village.
  5. Read the panel. You should see about 24 km, a Heading of about 202° SSW, and a Terrain (3D) block: a surface distance slightly longer than the straight one (about 24.7 km), Gain / loss, and Min / max elevation of roughly 1,380 m / 3,230 m. Your exact numbers depend on where you clicked.
  6. Click Clear All and close the panel.

Heading is not screen angle

The heading is the true great-circle initial bearing, not the angle the line makes on screen. Draw a line from Washington, D.C. to London at low zoom: it looks due east, but the heading reads about 50°.

Save the project (Project → Save).

Bonus 6.7: Downtown Reno in 3D, and the winter sun (if you finish early)

  1. View → Set view...: Longitude -119.8138, Latitude 39.5296, Zoom 16, Pitch 60, Bearing -30. Hide the earthquake and LiDAR layers. The Liberty basemap's buildings stand as 3D blocks along Virginia Street and the Truckee River.
  2. Choose Controls → Sun. The panel opens on 06/21/2024, the summer solstice. Click the month in Date and type 12212024 to make it December 21, the winter solstice; the panel header updates to Dec 21, 2024. Click the hour in Time and type 0330P for 03:30 PM. The buildings take on warm, low-angle winter light. Drag the time-of-day slider under the clock, or press play, to watch the light move. The sun position is computed on your machine; nothing is sent anywhere.
  3. Close the Sun panel with its × button.

Bonus 6.8: Full detail on a small classified cloud (if you finish early)

Regional surveys give the big picture; a small COPC shows full detail and a richer classification. Open this link in a new tab:

https://web.geolibre.app/?data=https://s3.amazonaws.com/hobu-lidar/autzen-classified.copc.laz
  1. The Autzen Stadium cloud (Eugene, Oregon) streams in at full extent: 10,653,336 points.
  2. Open Add data → LiDAR layer and set Color By to Classification. This time the list includes Ground, High Vegetation, Building, Water, Transmission Tower, Bridge Deck and several custom codes.
  3. Use View → Set view...: Longitude -123.0688, Latitude 44.0590, Zoom 16.3, Pitch 60, Bearing 30. Buildings are orange, trees green, water blue, and the stadium seating gray: it was left as Never Classified, which is exactly the kind of error the Point Cloud Annotation plugin is built to fix.

    The Autzen COPC colored by ASPRS classification

  4. Tick Enable point picking and click a few points to read their attributes.

Check your work

Open the finished project

The finished project (06-3d-and-time.geolibre.json) restores the Reno-Carson survey, the earthquake layer bound to the Time Slider at 2020, and two bookmarks. It opens on the tilted camera (pitch 55°, bearing -15°) with terrain on at 2× from the workshop DEM.

Going further

  • Plugins → Web services → USGS LiDAR maps where 3DEP point clouds exist, streams them into the view, and clips one to an area you draw as COPC. USGS 3DEP is the raster counterpart: the 1 m, 1/3 arc-second, and 1 arc-second DEMs derived from that LiDAR.
  • Add data → ICESat-2 / GEDI reads NASA spaceborne LiDAR granules in the browser; Layer actions → Along-track profile charts ground and canopy along a beam.
  • Plugins → Point cloud annotation labels points with box, lasso, and brush tools, and saves the labels with the project.
  • View → Split view puts a MapLibre pane beside a Cesium pane that follows it at the same ground resolution.
  • Add data → CZML dynamic 3D scene (Cesium only) loads time-dynamic scenes such as satellite orbits that animate on the globe clock.
  • Controls → Record map tour... turns keyframes into a video. Hold each stop for three or four seconds; slower reads better.
  • Plugins → Timelapse animates annual Sentinel-2, Landsat, and land-cover basemaps with no scene search at all.
  • Read Rendering Engines, Cesium Renderer, Map Controls, Indoor Maps, and Point Cloud Annotation.

Break: 2:50 - 3:00. Next: 7. Communicating and Collaborating