Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Historical land-use world data

The world compiler derives the 1544 land-use profile from HYDE 3.5 c9. HYDE is a global 5-arcminute historical reconstruction: it is regional evidence, not an exact observation of an individual settlement.

Manual preparation contract

Download the three HYDE 3.5 c9 April 2025 NetCDF inputs from Utrecht University's public HYDE vault:

https://geo.public.data.uu.nl/vault-hyde/hyde35_c9_apr2025%5B1749214444%5D/original/gbc2025_7apr_base/NetCDF/

The directory is protected by an interactive anti-bot page, so the repository initializer does not automate this download. Use a normal browser and retain the release filenames. The separate general_files.zip release input is still required for the matching general_files/garea_cr.asc cell-area grid.

Place exactly these release files in the Git-ignored target/world-data-sources/raw/hyde35-land-use/ directory, or point --land-use-dir at another directory:

  • cropland.nc / cropland — cropland area in km²
  • grazing_land.nc / grazing_land — grazing-land area in km²
  • urban_area.nc / urban_area — urban area in km²
  • general_files.zip / general_files/garea_cr.asc — HYDE grid-cell area in km²

For normal development, just init-world-data installs these exact four files into that directory from the pinned reviewed source-separated input bundle. Its HYDE component retains a separate notice and exact file inventory; the archive is not a combined derived world artifact. Manual browser retrieval is the fallback for preparing or independently auditing the HYDE component. See wiki/reference/world-data-bundles.md.

The importer requires NetCDF-4 inputs with time, lat, and lon dimensions in that order, matching 4,320×2,160 global 5-arcminute coordinate grids. It requires HYDE 3.5's 365_day time axis and verifies matching axes before sampling. It streams only the requested settlement cells from each large time slice rather than retaining the full grids in memory.

HYDE expresses land use as areas. The compiler brackets the requested world year in HYDE's time axis, linearly interpolates cropland, grazing, and urban area, and divides each by garea_cr.asc. At 1544 this interpolates 44% from the 1500 snapshot toward 1600. The remaining area is natural/seminatural land. Small source overlap (up to 5%) is normalized deterministically; greater overlap, malformed values, partial nodata, or missing source files fail the build. A cell with no usable complete profile receives the documented deterministic fallback and is counted in the build report.

Canonical land use is stored as bounded basis-point fractions for cropland, grazing land, built-up land, and natural/seminatural land. The four fractions sum to exactly 10,000 and support agriculture, livestock, encounters, and forest-cover fallbacks. The strategic map compiler separately reads raw interpolated HYDE cropland km² for every source cell intersecting the playable bounds. It does not reuse rounded settlement profiles. Largest-remainder rounding gives each HYDE cell a whole-square quota while preserving the rounded global area with less than 0.5 km² residual.

Those quotas are allocated to exact EPSG:3035 1 km squares by deterministic four-neighbour region growth. All settlements seed one global frontier; settlement-less HYDE cells receive a deterministic fallback seed, and adjacency crosses HYDE boundaries. Final inferred roads are merged before this allocation and therefore participate in both suitability and package identity. Bounded uniform-grid indexes measure exact point-to-line-segment distance without scanning every feature for every square. Suitability strongly prefers settlement proximity, gives roads only short-range influence, prefers an access band near water rather than banks, requires at least 75% of sixteen explicit within-square samples to be non-water passable land, penalizes 15-degree slopes/local relief, and weakly penalizes modern canopy. HYDE's observed historical cropland takes precedence over Jung potential-natural wetland vegetation because the latter describes the uncultivated counterfactual; mapped water remains ineligible. A HYDE cell clipped by the playable boundary may request more area than the arbitrary boundary can represent with complete, usable canonical squares. Rules version 2 deterministically saturates only those clipped edge cells at their usable-square capacity, including zero, and reports the omitted area. Un-clipped cells may saturate only when canonical-grid/raster discretization is at most 2 km² and at most 5% of the cell quota. Larger coastal or usable-land contradictions still fail the build. Soil is deliberately omitted: the current SoilGrids stage exposes settlement samples rather than a bounded map-wide raster, and expanding that compiler surface is disproportionate to this allocator.

The selected square IDs, HYDE source digest, rules version, residual, and counts determine the final terrain package identity. Exact square boundaries are used for both foraging legality and map paint.