storm-event-reconstruction

name

storm-event-reconstruction

description

Use when the user wants to reconstruct a hurricane, tropical storm, atmospheric river, bomb cyclone, or extreme wave/wind event from multiple data sources — phrases like "reconstruct Hurricane {Name}", "buoys during {Storm} landfall", "satellite imagery for {Event}", "storm surge at {Location} during {Storm}". Encodes the multi-source composition pattern (in-situ + satellite + track + surge + atmospheric) and the time/space windowing heuristic.

Storm event reconstruction

A storm reconstruction is a multi-source composition workflow. Done well, it answers "what happened on the ground / in the water during this event" by stitching together in-situ buoys, satellite imagery, official track data, water-level gauges, and atmospheric model output.

See recipe.md for a full Hurricane Ian (2022) walkthrough.

The five sources to compose

Source	Collection	What you get
In-situ waves / wind / SST	NDBC (offshore), CDIP (wave-specific)	Time-series buoy observations
Satellite imagery	GOES_R (visible/IR every 1–10 min), sentinel-1-grd (SAR for cloud-penetrating views)	Geostationary IR + SAR
Track and intensity	NOAA_AOML_HDB	Best-track lat/lon, MSW, MSLP per 6-hour fix
Storm surge / water level	COOPS, NOS_COOPS	Coastal tide gauges with surge anomalies
Atmospheric forecast / reanalysis	HRRR	Hourly 3km CONUS atmospheric model fields

Time and space windowing heuristic

For a storm that made landfall at time T and location L:

Source	Time window	Spatial window
Track (NOAA_AOML_HDB)	T - 7 days → T + 2 days (full lifecycle)	Whole-basin bbox
Buoys (NDBC)	T - 48h → T + 24h	Storm radius × 2 around the track
GOES (GOES_R)	T - 24h → T + 12h	Bbox covering the eye + outer bands
Surge (COOPS)	T - 24h → T + 48h	Coastal stations within 200 km of L
HRRR (HRRR)	T - 24h → T + 12h	Same as GOES bbox

Storm radius scales with intensity: Cat 1 ~50–100 km, Cat 5 ~100–200 km. Double it for the bbox.

Pattern: multi-tool composition

# 1. Get the official track first (sets time/space scope for everything else)
track = search_datasets(
  collections=["NOAA_AOML_HDB"],
  q="<storm name>",
  datetime="<basin season>",
  limit=10
)

# 2. From the track, derive bbox (max storm extent) and landfall time T

# 3. NDBC buoys near the track during the eyewall passage
ndbc = search_datasets(
  collections=["NDBC"],
  bbox=<storm bbox>,
  datetime="<T-48h>/<T+24h>",
  filter="properties->>'cube:variables'->>'wave_height'->>'max' > 4",
  filter_lang="cql2-text",
  limit=30
)

# 4. GOES-R imagery during eyewall passage
goes = search_datasets(
  collections=["GOES_R"],
  bbox=<storm bbox>,
  datetime="<T-12h>/<T+6h>",
  limit=20
)

# 5. Coastal water-level gauges along the landfall coast
surge = search_datasets(
  collections=["COOPS"],
  bbox=<coastal strip near landfall>,
  datetime="<T-24h>/<T+48h>",
  q="water level",
  limit=20
)

# 6. (optional) HRRR atmospheric reanalysis for environment
hrrr = search_datasets(
  collections=["HRRR"],
  bbox=<storm bbox>,
  datetime="<T-24h>/<T+12h>",
  limit=10
)

Variable thresholds for "extreme" filters

If the user asks for buoys that recorded extreme conditions:

Variable	Cat 1 threshold	Cat 3+ threshold
wave_height (max)	4 m	6 m
wind_speed (max)	33 m/s	50 m/s
pressure (min)	990 hPa	950 hPa
gust (max)	40 m/s	60 m/s

Use cql2-filter-builder for the actual CQL2 syntax.

Multi-storm comparisons

If the user wants to compare two storms:

1. Run the above pattern for each storm separately
2. Don't try to compose both into a single search (different bboxes, different time windows)
3. In the output, normalize by relative time (T - landfall_time) for cross-storm comparison

Atmospheric river / bomb cyclone variants

Same pattern, but:

• Drop NOAA_AOML_HDB (no tropical track data)
• Add GOES_R for water vapor channel + cloud features
• Add HRRR early (it's the workhorse atmospheric source for non-tropical events)
• Use longer time windows (atmospheric rivers persist 3–7 days)

Output expectations

Tell the user upfront:

• "I'll need to make 4-5 tool calls to fully reconstruct this event."
• "Buoy availability depends on which buoys were operational and within the storm's path."
• "Some buoys go offline during eyewall passage (lost data), which itself is a signal."