Raster Operations

Raster operations transform gridded data: band math, reclassification, clipping, mosaicking, reprojection, and normalization. Implemented using GDAL and rasterio, running in the local or cloud compute tier.

Part of the geo domain.

Operations

Operation	Type	Description	Key Params
raster_ndvi	raster → raster	Normalized Difference Vegetation Index: (NIR - Red) / (NIR + Red). Output: float32, -1 to 1.	(none, inputs are NIR and Red bands)
raster_reclassify	raster → raster	Remap values to new classes. Supports manual breaks, equal interval, quantile, value map, and crosswalk-based mapping.	method, breaks, class_values, n_classes, value_map, crosswalk
raster_clip	raster + vector → raster	Clip raster to a polygon boundary or bounding box.	all_touched (bool), crop (bool)
raster_mosaic	raster[] → raster	Combine multiple rasters into one. All inputs MUST share a CRS.	method (first, last, max, min, mean), nodata
raster_reproject	raster → raster	Reproject to a new CRS with configurable resampling.	crs (required), resolution, resampling (nearest, bilinear, cubic, lanczos)
raster_calc	raster → raster	Band math expressions using numpy. Bands accessible as b1, b2, etc.	expression (string, required)
raster_normalize	raster → raster	Normalize values to a target range (min-max scaling).	method, output_range
raster_mask	raster + vector → raster	Mask raster values outside a geometry.	(geometry input)
raster_resample	raster → raster	Resample raster to a new resolution.	resolution, resampling method
raster_rasterize	vector → raster	Convert vector features to raster grid.	attribute, resolution

Examples

NDVI from Sentinel-2

layers:
  analysis/ndvi:
    type: raster
    compute:
      op: raster_ndvi
      inputs:
        nir: { layer: source/sentinel2-b8 }
        red: { layer: source/sentinel2-b4 }
    style:
      palette: ndvi
      rescale: "-0.2,0.9"

Reclassify with Manual Breaks

layers:
  analysis/slope-classes:
    type: raster
    compute:
      op: raster_reclassify
      inputs:
        raster: { layer: terrain/slope }
      params:
        method: manual
        breaks: [5, 15, 30, 45]
        class_values: [1, 2, 3, 4, 5]
        labels: ["Flat", "Gentle", "Moderate", "Steep", "Cliff"]

Reclassify with Crosswalk

Crosswalks enable value remapping between classification systems using the crosswalk registry.

Decided in ADR-0025 - YAML crosswalk registry with SKOS predicates.

layers:
  analysis/harmonized-landcover:
    type: raster
    compute:
      op: raster_reclassify
      inputs:
        raster: { layer: source/nlcd }
      params:
        crosswalk: nlcd-to-esa-worldcover

Band Math Expression

layers:
  analysis/ndbi:
    type: raster
    compute:
      op: raster_calc
      inputs:
        raster: { layer: source/sentinel2 }
      params:
        expression: "(b11 - b8) / (b11 + b8)"

Clip and Mosaic

layers:
  processing/clipped:
    type: raster
    compute:
      op: raster_clip
      inputs:
        raster: { layer: source/dem }
        geometry: { layer: source/study-area }
      params:
        all_touched: true
        crop: true

  processing/combined:
    type: raster
    compute:
      op: raster_mosaic
      inputs:
        rasters:
          - { layer: source/tile-a }
          - { layer: source/tile-b }
          - { layer: source/tile-c }
      params:
        method: first