Skip to content

GridLook

A WebGL viewer for Earth system model output, with first-class support for the unstructured and reduced grids that climate models actually produce. Maintained by DKRZ and MPI-M under the MIT license.

At a glance

  • Repo — dkrz-icon/gridlook
  • Hosted — gridlook.pages.dev
  • Shape — Vue 3 single-page app, also embeddable as a Vue component
  • Render API — three.js with custom GLSL shaders
  • Zarr versions — v2 and v3 via zarrita
  • Conventions — CF (grid_mapping, _FillValue, missing_value, lat/lon coordinate variables)

What it does

GridLook's defining feature is grid-topology awareness. Where the other viewers in this section assume a regular lat-lon raster (or accept that as input after preprocessing), GridLook detects and renders eight grid types directly from the data:

  1. Regular lat-lon
  2. Rotated regular (CORDEX-style, detected via grid_mapping_name="rotated_latitude_longitude")
  3. HEALPix (hierarchical, levels 0 to 11+)
  4. Triangular meshes (ICON, detected via vertex_of_cell)
  5. Gaussian reduced (ERA5-style, longitude tapers toward the poles)
  6. Curvilinear (2D lat/lon coordinate arrays)
  7. Irregular (CMIP6 unstructured)
  8. Irregular Delaunay (on-the-fly Delaunay fallback)

The README is explicit that GridLook is "not a competitor to high-end visualization suites" but a tool for intuitive, interactive exploration of cloud-hosted ESM output. Operational use includes the WCRP Global Hackathon 2025 HEALPix datasets and DKRZ-hosted ICON, IFS DYAMOND3, and EERIE collections.

How it renders

The primary view is a 3D nearside-perspective globe with orbit controls. Eight 2D cartographic projections are also available: Mercator, Robinson, Mollweide, Cylindrical Equal Area, Equirectangular, Azimuthal Equidistant, and a custom Azimuthal Hybrid for seamless polar plus global views. Projection math is computed once on the CPU (via d3-geo and d3-geo-projection) and the per-pixel reprojection runs in GLSL, so panning between projections stays smooth.

Coastlines and graticules use Natural Earth GeoJSON at 50m resolution and are rendered via WebGL after a recent refactor; both are toggleable. Land/sea masking is generated CPU-side from GeoJSON onto a canvas, then applied as a GPU texture.

The data pipeline is load chunk, normalize via offset and scale-factor, optionally posterize, look up against one of 58 built-in colormaps (viridis, inferno, coolwarm, RdBu, BrBG, diff, curl, phase, ...), then render. Histogram-based auto-scaling and per-variable default colormaps and ranges are baked into the dataset catalog.

Zarr handling

Reads use zarrita 0.7.x with bitround codec support. Chunk loading targets 1 to 5 MB requests, with byte-range coalescing and a 512-entry LRU byte cache. CF metadata (grid_mapping, _FillValue, missing_value, NaN) drives both grid detection and nodata masking. Catalog files (JSON) list datasets for institutional deployments and are how DKRZ exposes its hosted collections.

Where it fits

Choose GridLook when the data lives on a non-rectilinear grid that the climate community actually uses (HEALPix, ICON triangular, Gaussian-reduced, rotated lat-lon) and you want it visualized on that native grid rather than after a regridding step. The map-hosted libraries above either require regular gridded input or assume web-map tiles, so they are not substitutes for the unstructured-grid case. The trade-off is that GridLook is its own scene, not a layer on top of a basemap; if you need a vector basemap underneath, look elsewhere.