Historical summer drought and wetness

Settlement hydroclimate is sourced from the NOAA Old World Drought Atlas (OWDA) v1.0, a tree-ring reconstruction of annual summer Palmer Drought Severity Index (PDSI) across Europe and the Mediterranean.

• Pinned release: OWDA v1.0, dataset DOI 10.25921/rjm6-mq74.
• Authoritative file URL: https://www.ncei.noaa.gov/pub/data/paleo/drought/owda.nc
• Publication: Cook et al. (2015), Old World megadroughts and pluvials during the Common Era, DOI 10.1126/sciadv.1500561.
• Pinned file: exactly 228226363 bytes; SHA-256 c044aa52e9e81932841b642b6977fa6f84beb9fe73c3db502b90f4295b1d65bd.
• Grid: 0.5 degree point grid, 114 longitudes by 88 latitudes.
• Time coverage: AD 0 through 2012.
• Maintainer audit/derivation input: NetCDF-4 classic-model/HDF5.

Release maintainers run just init-owda to download or verify the pinned file at target/world-data-sources/raw/climate/owda.nc. Preparation is atomic: a temporary sibling is size- and checksum-verified before replacement. An adjacent ignored .owda-source.json records the version, URL, both DOIs, size, and checksum. A mismatched existing cache fails closed; use python scripts/init_owda.py --force only to replace it. Override the compiler path with --drought-netcdf. The raw file is never a standard developer input and is never included in a world-data input bundle. The importer uses a pure-Rust, read-only NetCDF-4 decoder; no NetCDF/HDF5 native library enters the shared schema or database module.

Standard developer input and release derivation

Developers install the reviewed bundle's target/world-data-sources/prepared/owda/settlement-profiles-1544.json and run the ordinary just compile-world. That bounded profile contains exactly one sorted entry per bundled Viabundus settlement, its selected-year/20-year summary, and truthful direct or nearest sampling classification. It contains neither OWDA coordinates nor annual values.

After auditing the raw file and preparing the matching Viabundus .viabundus-source.json sidecar, a release maintainer produces that file deterministically with:

cargo run -p adventuresim-world-import -- \
  --drought-netcdf target/world-data-sources/raw/climate/owda.nc \
  --derive-owda-profiles target/world-data-sources/prepared/owda/settlement-profiles-1544.json

The command refuses a missing grid sample rather than emitting a deceptive profile, generates (or validates) settlement-ids-1544.json, hashes both matching Viabundus records into the output, and is limited to the reviewed 1544 release. Bundle verification checks those hashes, the exact settlement-ID coverage, integer bounds, and sampling classification.

Source boundary

The parser requires the documented lon=114, lat=88, and time=2013 dimensions; monotonically spaced longitude, latitude, and annual time axes; and an f64 pdsi(lon, lat, time) variable with the source long name and units. It reads the twenty summers ending in the selected world year. A cell must have all twenty finite, three-decimal values or twenty NaNs denoting no source coverage. Partial histories, infinities, unexpected precision, changed axis orientation, and values outside the audited PDSI range fail at the source boundary.

The downloaded 228 MB file was verified directly. Across the complete dataset, 8,826,343 of 20,194,416 values are finite; values span -11.996 to 13.431 and have exactly three decimal places. The 1525–1544 window contains 5,414 complete grid cells with stable coverage.

Canonical model and sampling

PDSI is stored as a bounded signed integer in thousandths, not a raw float. Standard thresholds derive extreme, severe, moderate, or mild drought; normal conditions; and mild through extreme wetness. Each settlement stores:

• its reconstructed 1544 summer PDSI;
• mean PDSI over 1525–1544;
• the number of moderate-or-worse drought summers (PDSI at most -2);
• the number of moderately-or-more wet summers (PDSI at least 2).

The history constructor guarantees that both counts fit the twenty-year window, cannot sum past it, include the current summer in the correct category, and admit the stored mean under the bounded drought, normal, and wet value ranges. Canonical data distinguishes reconstructed profiles from complete inferred profiles; there is no unknown state.

Sampling first uses the containing half-degree grid point. For a coastal or otherwise missing cell, it selects the physically nearest reconstructed point, with longitude distance scaled by latitude, up to 1.5 degrees. A location beyond that limit receives a neutral inferred profile. An audit of all 6,041 Viabundus settlements active in 1544 produced 5,458 direct samples, 583 nearest neighbors, and no fallbacks.

The data can directly affect current harvests, seasonal moisture and yields, water stress and availability, fire risk, forage, river conditions, and travel. The twenty-year history can inform reserves, prices, migration pressure, and whether a single dry or wet summer is locally unusual. Selected-year drought does not permanently assign or alter a settlement's biome.

Attribution and redistribution boundary

The repository and compiled world contain only bounded per-settlement derived values: selected-year PDSI, twenty-year mean, drought/wet counts, and the reconstruction/inference classification with concise provenance. They do not contain source grids, annual series, the complete NetCDF file, journal prose or figures, or rendered journal/source maps. Reusers should cite both the NOAA dataset DOI and Cook et al. paper DOI above. The local source cache remains ignored and must not be committed or redistributed through this repository.