geri-kmap: Global Ecosystem Research Infrastructure sites

An interactive map and database of the sites of the six research infrastructures (RIs) in the Global Ecosystem Research Infrastructure (GERI):

• NEON: National Ecological Observatory Network (United States)
• eLTER: Integrated European Long-Term Ecosystem, critical zone and socio-ecological Research
• ICOS: Integrated Carbon Observation System (Europe)
• TERN: Terrestrial Ecosystem Research Network (Australia)
• SAEON / EFTEON: South African Environmental Observation Network
• CERN: Chinese Ecosystem Research Network

GERI's NSF AccelNet project (award 2301655) harmonizes observations across these networks, starting with ecological drought. geri-kmap catalogues where the RIs operate and cross-walks each site's identifiers across registries. It also records which sites the GERI harmonized products cover. Those products live on the CyVerse Data Store at /iplant/home/shared/geri.

Today the catalogue holds 1,893 sites in 52 countries. They come from DEIMS-SDR, the NEON Data API, the ICOS Carbon Portal, TERN's OzFlux server and AusPlots API (820 surveillance plots), SAEON DataCite station records, CERN's station list and the GERI Data Store. All 117 site ids in the GERI harmonized products resolve to catalogue sites. See docs/.

The catalogue also carries what Loescher et al. (2026), "Toward a global ecosystem network-of-networks" (Ecosphere 17: e70789, doi:10.1002/ecs2.70789), adds:

• the RI profiles of its Table 1;
• the 33 MexFlux towers of its GENNI use case, kept outside GERI;
• the 251 partner organizations of its landscape survey.

See docs/genni.md.

Repository layout

A two-stack project, built on the engine of tyson-swetnam/lto (itself forked from tyson-swetnam/cod-kmap):

1. Python data pipeline (scripts/, schema/, data/):
   • Fetchers read each RI registry into data/raw/R-GERI-*/ (committed).
   • scripts/load_geri_sites.py writes those files into a local DuckLake store of record (DuckDB 2.0, SQLite catalog at db/geri_catalog.sqlite; data files in db/lake/; both gitignored).
   • scripts/export_parquet.py writes Parquet and GeoJSON to public/. The committed public/parquet/ is the canonical copy of the data.
2. Static MapLibre + DuckDB-Wasm site (index.html, src/, public/), published to GitHub Pages with no build step.

Quick start

Python 3.11 or newer (DuckDB 2.0).

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python scripts/rebuild_lake_from_parquet.py   # public/parquet/ → db/geri_catalog.sqlite + db/lake/
python scripts/qa.py                          # data-quality + schema-constraint gate
python -m http.server 5173                    # then open http://localhost:5173/

Refresh from the registries:

python scripts/fetch_deims.py && python scripts/fetch_neon.py && python scripts/fetch_icos.py
python scripts/fetch_ozflux.py && python scripts/fetch_saeon_stations.py && python scripts/fetch_geri_datastore.py
python scripts/fetch_elter_dar.py && python scripts/fetch_ausplots.py && python scripts/fetch_mexflux.py
python scripts/fetch_landscape_survey.py && python scripts/fetch_geri_papers.py
python scripts/summarize_harmonized.py --download   # only when the harmonized products change (~1 GB)
python scripts/load_geri_sites.py
python scripts/apply_facility_fixes.py
python scripts/check_endpoints.py                    # requests every data-access URL (after import_lto_archives.py if lto changed)
python scripts/load_geri_archives.py
python scripts/load_geri_paper.py
python scripts/qa.py
python scripts/export_parquet.py && python scripts/export_view_caches.py

License

Code: MIT (see LICENSE). Site records keep the terms of their source registries; see docs/sources.md.