forecast-inventory

Files (6)

• SKILL.md
• datasets.json · fixtures/
• variables.json · fixtures/
• datasets.rq · queries/
• variable_names.rq · queries/
• inventory.py · scripts/

SKILL.md (first lines)

# Forecast inventory

The ecoforecast graph is a **STAC/schema.org catalog**. `variableMeasured`
names such as `prediction`, `mean`, `quantile97.5`, and `crps` are **file
columns**, not numeric series in RDF. This skill inventories that catalog.

Default endpoint (ask first):
`https://qlever.geocodes-aws.earthcube.org/graphspace/ecoforecast`

## Agent workflow

1. Confirm the ecoforecast endpoint (or JSON).
2. Pick `themes`, `models`, `columns`, or `sites-map`.
3. For depth/ocean plots, switch to `catalog-plot` on deepoceans instead.
4. Say clearly when a result is a column name rather than a value.

## CLI

```bash
EP=https://qlever.geocodes-aws.earthcube.org/graphspace/ecoforecast

uv run python skills/forecast-inventory/scripts/inventory.py themes \
  --endpoint $EP -o runs/themes.png

uv run python skills/forecast-inventory/scripts/inventory.py models \
  --endpoint $EP -o runs/models.png

uv run python skills/forecast-inventory/scripts/inventory.py columns \
  --endpoint $EP

uv run python skills/forecast-inventory/scripts/inventory.py sites-map \
  --endpoint $EP --limit 4000 -o runs/sites.png
```

| Mode | Output |
|---|---|
| `themes` | bars from STAC URL `/forecasts/<theme>/<variable>/models/<model>` (datasets without that path are `(no STAC theme)`) |
| `models` | bars of model ids (`schema:name` / URL) |
| `columns` | markdown table of variableMeasured names (no PNG) |
| `sites-map` | delegates to `catalog-map points` |

For CRPS / mean time series from the parquet `contentUrl`, use **`forecast-parquet`**. For one lead (one score per target date), use `forecast-parquet horizon`.

## Do not

- Treat `crps` / `prediction` as SPARQL numbers
- Mix this skill with DepBelowSurf charts

From earthcube/decoder-pi, Apache-2.0.

More from earthcube/decoder-pi