Operation Model

Operations are abstract interfaces that define what inputs they accept, what outputs they produce, and what parameters configure their behavior. The platform routes operations to the appropriate compute tier automatically. Users rarely need to specify where an operation runs.

Decided in ADR-0005 - DATA + COMPUTE + STYLE per layer. Pipeline eliminated.Decided in ADR-0007 - engines, connectors, and drivers are three orthogonal concerns.

Operation YAML Schema

Every operation is defined in a YAML file within its domain's operations/ directory.

Field	Type	Required	Description
id	string	Yes	Unique identifier (e.g., terrain_slope)
name	string	Yes	Human-readable name
description	string	Yes	What this operation does
version	string	No	Semantic version
category	string	Yes	Category within the domain (e.g., terrain, raster)
type	string	Yes	Transform signature (e.g., raster-to-raster, raster-vector-to-table)
default_implementation	string	No	Default implementation library (e.g., native, duckdb)
inputs	object	Yes	Typed inputs with descriptions
outputs	object	Yes	Typed outputs with format, dtype
params	object	No	Parameters with type, default, min/max, options
display_hints	object	No	Default view configurations for results
cache_policy	object	No	TTL, invalidation triggers
granularity	object	No	Resolution constraints
uncertainty	object	No	Error model, known limitations
execution	object	No	Resource profiles, cost hints
default_input	object	No	Default data source connector and params

# Example: terrain_slope operation
id: terrain_slope
name: Slope
description: >
  Calculate slope from DEM using GDAL DEMProcessing (Horn's method).
  Output is float32 degrees where 0 = flat and 90 = vertical cliff.
version: 2.3.0
category: terrain
type: raster-to-raster
default_implementation: native

inputs:
  - name: dem
    type: raster
    format: dem
    description: Input Digital Elevation Model (GeoTIFF)
    required: true

outputs:
  - name: slope
    type: raster
    format: cog
    dtype: float32
    unit: degrees
    description: Slope in degrees (0=flat, 90=vertical)

params:
  smooth:
    type: boolean
    default: true
    description: Apply Gaussian smoothing to reduce noise
  smooth_sigma:
    type: number
    default: 1.5
    description: "Gaussian sigma (~4-5m radius at 1m resolution)"
    min: 0.1
    max: 10.0

display_hints:
  map:
    renderer: maplibre
    palette: terrain
    opacity: 0.8
    rescale: "0,60"
    colormap: viridis
  info:
    fields: [slope_degrees]
    format: "{value}°"
    section: Terrain

cache_policy:
  ttl_days: 90
  invalidate_on: [source_update]

Inputs and Outputs

Inputs and outputs are typed. The type field indicates the data kind:

Type	Description	Common Formats
raster	Gridded raster data	geotiff, cog, dem
vector	Feature geometries	geojson, geopackage
table	Tabular data	parquet, csv, arrow
stats	Statistical summaries	json
raster[]	List of rasters	(same as raster)

An input with required: true MUST be provided. Inputs with required: false MAY be omitted, in which case the operation either uses a default or adjusts its behavior.

Parameters

Parameters configure operation behavior. Each parameter declares:

Field	Type	Description
type	string	string, number, integer, boolean, enum, array, object
default	any	Default value if not provided
required	boolean	Whether the parameter MUST be provided
description	string	What this parameter controls
min / max	number	Numeric bounds
enum	list	Allowed values for enum types

Display Hints

Operations declare default display configurations for their results. These are suggestions; layer-level style: blocks override operation-level display_hints.

Decided in ADR-0009 - operations provide display hints; layer style takes precedence.

display_hints:
  map:
    renderer: maplibre
    palette: viridis
    opacity: 0.8
    rescale: "0,60"
  table:
    renderer: tanstack-table
    sortBy: value
  chart:
    renderer: observable-plot
    chartType: bar
  info:
    fields: [slope_degrees]
    format: "{value}°"
    section: Terrain

Resolution priority: layer style: > operation display_hints > type defaults.

Compute Routing

When a layer uses compute.op, the platform routes it to a compute tier:

1. External override - if compute.backend: gee is set, route to Google Earth Engine
2. Data locality - compute near the data (cloud tier for remote data, local tier for local files)
3. Data size - small datasets may run in the browser tier; large datasets route to local or cloud
4. Availability - use the libraries available in the current tier

Google Earth Engine is the primary external compute option. All other routing is automatic. See Compute Routing for the full routing table.

Custom Operations

Users MAY define workspace-local operations by placing an operation.yaml and a compute.py in their workspace:

my-workspace/
  folia.yaml
  ops/
    custom_index/
      operation.yaml   # operation definition
      compute.py       # implementation

The workspace-local operation is available only within that workspace. Publishing the workspace to the catalog makes the operation available to others.

Operation Categories

Browse operations by category:

• Terrain - slope, aspect, hillshade, curvature, TPI, TRI, viewshed
• Raster - NDVI, reclassify, clip, mosaic, reproject, calc, and more
• Vector - buffer, dissolve, intersection, union, simplify, clip
• Analysis - zonal stats, weighted overlay, change detection, H3 index
• Tabular - filter, select, sort, join, union, aggregate
• Temporal - resample, rolling, align, interpolate, time range, period stats