Skip to content

README.md

Latest commit

 

History

History
68 lines (52 loc) · 2.48 KB

README.md

File metadata and controls

68 lines (52 loc) · 2.48 KB
Raw

INLASPDERasters.jl

Raster ↔ SPDE glue: extract covariate values from Rasters.Raster sources onto SPDE mesh vertices, and return posterior-field predictions as Rasters.Raster objects.

Companion package to INLASPDE.jl.

Status

v0.4.0. The package ships a working raster-SPDE bridge and is exercised end-to-end by the Meuse SPDE vignette:

  • extract_at_mesh — sample a Raster at mesh vertex coordinates.
  • predict_raster — Gaussian-approximation and sample-based SPDE → raster projection.
  • quantile_rasters — joint mean / sd / lower / upper credible-interval rasters from per-vertex summaries.
  • Exceedance — quantity wrapper for tail probabilities under the sample-based path.

CRS mismatches between mesh and raster are surfaced at the API boundary (see mesh_crs keyword, ADR-041).

API at a glance

using INLASPDE, INLASPDERasters, Rasters, Random

# 1. Extract a covariate raster onto mesh vertices.
elev_raster = Raster("elevation.tif")
elev_at_vertices = extract_at_mesh(elev_raster, mesh)

# 2. Project a fitted SPDE component back onto a target grid.
template = Raster(zeros(nx, ny), (X(xs), Y(ys)))
r_mean  = predict_raster(model, res, template;
    component = SPDE2, quantity = :mean)
r_lower = predict_raster(model, res, template;
    component = SPDE2, quantity = :lower)

# 3. Sample-based path for exceedance probabilities P(η > c | y).
rng = Xoshiro(123)
r_exc = predict_raster(rng, model, res, template;
    component = SPDE2, quantity = Exceedance(0.5), n_samples = 1000)

predict_raster(model, res, template) requires the SPDE2 component to retain its mesh — construct it via SPDE2(mesh::INLAMesh; …) rather than the back-compat SPDE2(points, triangles; …) form (ADR-036).

Why a separate package, not a weakdep of INLASPDE

  • Rasters.jl pulls GDAL_jll, Proj_jll, NetCDF_jll, and other binary artifacts totalling hundreds of MB. A weakdep triggered by using Rasters would inflate the install / precompile cost for everyone who happens to load Rasters in the same session, regardless of whether they care about SPDE.
  • Raster-specific covariate extraction belongs in one dedicated place, not sprinkled across an extension.

See also