Add GeoNames postal codes Parquet (1.83M rows, 121 countries) and dataset card

README.md

@@ -1,3 +1,88 @@
 ---
-license: apache-2.0
+license: cc-by-4.0
+pretty_name: GeoNames Postal Codes (Parquet)
+tags:
+  - geospatial
+  - postal-codes
+  - pincode
+  - geocoding
+  - geonames
+size_categories:
+  - 1M<n<10M
+configs:
+  - config_name: default
+    data_files: postal_codes.parquet
 ---
+
+# GeoNames Postal Codes — Parquet
+
+The [GeoNames postal code dataset](https://download.geonames.org/export/zip/)
+(`allCountries.txt`) converted to a single sorted, zstd-compressed Parquet file.
+It is used as a postcode → lat/lon lookup table for joining tabular locations
+onto gridded weather data (e.g. ERA5 Zarr) in
+[zarr-datafusion](https://github.com/stratoscale-io/zarr-datafusion).
+
+| | |
+|---|---|
+| Rows | 1,826,904 (one row per **place**, not per postcode) |
+| Distinct `(country_code, postal_code)` | 1,080,715 |
+| Countries | 121 |
+| File | `postal_codes.parquet`, ~19 MB, zstd(3), sorted by `country_code, postal_code, place_name` |
+| Source snapshot | GeoNames export of 2026-09-22 |
+
+## Schema
+
+| column | type | nullable | notes |
+|---|---|---|---|
+| `country_code` | string | no | ISO 3166-1 alpha-2 |
+| `postal_code` | string | no | text — many countries use letters (`EC1A`, `100-0001`) |
+| `place_name` | string | yes | locality / post office |
+| `admin_name1` / `admin_code1` | string | yes | state / province |
+| `admin_name2` / `admin_code2` | string | yes | county / district |
+| `admin_name3` / `admin_code3` | string | yes | community / sub-district |
+| `latitude`, `longitude` | float64 | no | WGS84 decimal degrees |
+| `accuracy` | int8 | yes | 1 = estimated, 4 = GeoNames place id, 6 = centroid of addresses or shape |
+
+## Caveats
+
+- **One row per place.** A postcode can have many rows (Delhi `110001` has 21);
+  collapse to one point per `(country_code, postal_code)` before joining.
+- **`accuracy` describes the method, not the error.** Countries without published
+  postal geometry are geocoded by place name. India, for example, has no
+  accuracy-6 rows, and over 100 Bengaluru `560xxx` pincodes share one fallback
+  district point ~28 km north of the city centre — including rows marked 4.
+- **UK, Canada and the Netherlands** are outward/short codes only here; GeoNames
+  publishes full codes separately (`GB_full`, `CA_full`, `NL_full`), not included.
+
+## Usage
+
+Direct HTTPS (supports range requests):
+
+```
+https://huggingface.co/datasets/Stratoscale/GeoNamesPincode/resolve/main/postal_codes.parquet
+```
+
+```python
+import duckdb
+duckdb.sql("""
+  SELECT * FROM 'hf://datasets/Stratoscale/GeoNamesPincode/postal_codes.parquet'
+  WHERE country_code = 'IN' AND postal_code = '110001'
+""")
+```
+
+## Reproduce
+
+```bash
+curl -O https://download.geonames.org/export/zip/allCountries.zip
+unzip allCountries.zip allCountries.txt
+zarr-cli cookbook/pincode-lookup/build_postcodes.sql
+```
+
+## License and attribution
+
+Licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), the
+license of the source data. Contains data from
+[GeoNames](https://www.geonames.org/) (© GeoNames, CC BY 4.0). The data is
+provided "as is" without warranty of accuracy, timeliness or completeness.
+Changes from the source: converted from TSV to Parquet, typed columns, sorted;
+no rows added, removed or modified.

postal_codes.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9273a1d4b75ab11b980a284557ba42414975fe4846b8ac6b646f046a302c46
+size 20177129