Discovering ocean data with AQUAVIEW

AQUAVIEW speaks the STAC standard, so the standard client just works — one line in the language you already use, no sign-in for discovery.

from pystac_client import Client

STAC = "https://service.aquaview.org/stac"
cat = Client.open(STAC)
print("Connected to AQUAVIEW ✓")

library(rstac)

stac_url <- "https://service.aquaview.org/stac"
cat <- stac(stac_url)
cat("Connected to AQUAVIEW ✓\n")

AQUAVIEW federates the whole ocean-data landscape into one searchable place. 608,400 datasets, ~100 sources. Instead of tracking down data across many separate servers, clouds and portals, you discover and reach it in one place.

608,400 datasets  ·  100 sources  ·  one catalog, one API

The transfer skill: you won’t know collection IDs cold. Each source carries a title, description and keywords — already fetched once in A2 — so we filter locally: no extra calls, and you can see exactly what matched. Read the result carefully: it searches how a source describes itself, so find_sources("wave") misses PacIOOS. Rule: a provider you half-remember → find_sources(); datasets in your study area → a bounding-box search (Part B); browse → Explore.

# local filter over id+title+description+keywords
HAY = {c.id: haystack(c) for c in collections}
def find_sources(term, n=6):
    return [cid for cid, h in HAY.items()
            if term.lower() in h][:n]

# rstac: free-text the fetched list over
# id + title + description + keywords
find_sources("glider")

glider      → CALOOS, CCHDO, CENCOOS, IOOS, MARACOOS, NANOOS
wave        → CDIP, CENCOOS, GFS, NDBC
chlorophyll → NEFSC          # matched via keywords, no ID known
vessel      → CERULEAN, MARINECADASTRE_AIS, …

Whichever door you come in, a result is the same STAC item with the same schema — so the data looks identical downstream.

The core move: a bounding box around the fleet and a date range. One call reaches across every source — no per-server hunting.

HAWAII = [-161, 18, -154, 23]   # W,S,E,N
hits = cat.search(bbox=HAWAII,
                  datetime="2025-01-01/..")
print(hits.matched(), "datasets")

hits <- cat |> stac_search(
    bbox = c(-161, 18, -154, 23),
    datetime = "2025-01-01T00:00:00Z/.."   # R: full RFC3339
  ) |> get_request()

4,228 datasets match  # 2025–present, across the Hawaiʻi box

A wide search returns everything intersecting your box and time — that’s discovery working, not failing. Here it’s dominated by Cerulean SAR oil-slicks. So you narrow with collections= — the real skill — and pull one dataset per relevant source. Same columns, wildly different sources. Mind the SST source: satellite SST lives in plain COASTWATCH; COASTWATCH_WC is the West-Coast node — over Hawaiʻi it carries HF-radar currents and ocean colour.

source     id        title
CERULEAN   5214248   Oil slick 5214248
CERULEAN   5214247   Oil slick 5214247       # dense here — but not what a glider needs
CERULEAN   5214246   Oil slick 5214246

source              id                   title
COASTWATCH_WC       ucsdHfrH1_Lon0360    Currents, HF Radar, US Hawaiʻi State
PacIOOS             ww3_hawaii_lon180    WaveWatch III (WW3) Hawaiʻi Wave Model
OBIS                345c1b88-2bab…       DFO Maritimes Cetacean Sightings
MARINECADASTRE_AIS  mc_dataset_67919     Vessel Routing Measures
IOOS                sg626-20250729…      Seaglider sg626 track

Before downloading anything, ask how much of each kind exists in your box — one cheap count per source, no data moved. This is the bar chart in the live notebook.

Ocean colour / HF radar (COASTWATCH_WC)  ████████████  293
Satellite SST (COASTWATCH)               ████          91
Hawaiʻi waves & models (PacIOOS)        ███           61
Species-occurrence context (OBIS)        ██            42
Vessel-traffic context (AIS)             █             27
Glider Data Assembly Center (IOOS)       █              6

We don’t know any glider IDs. We know a box and a time — that’s enough. Search the IOOS Glider DAC inside the Hawaiʻi box and read what comes back. Every item carries its variable list, so you check it has what you need before downloading.

gliders = list(cat.search(bbox=HAWAII,
    collections=["IOOS"], datetime=WHEN).items())
for g in gliders:
    p = g.properties
    print(g.id, p["start_datetime"][:10],
          "→", p["end_datetime"][:10])

gl <- cat |> stac_search(collections = "IOOS",
        bbox = hawaii, datetime = when) |> get_request()
for (f in gl$features)
  cat(f$id, f$properties$start_datetime, "\n")

6 glider deployments in the box since 2025
  sg626-20250729T1452   2025-07-29 → 2025-10-29   # completed 3-month Seaglider — our pick (reproducible)
  sg511-20260722T1831   2026-07-22 → 2026-09-14   # flying right now — the “your turn” swap
  sg511-20250729T1532 · sg148-20260722T1939 · ng457-20260713T0000 · ng344-20260713T0000

sg626: 60 variables — time ✓ latitude ✓ longitude ✓ depth ✓ temperature ✓ salinity ✓
       also aboard: sbe43_o2, aa5013_o2, bb2f_470/695/700   ·   tabular assets: csv, nc, json

The csv asset points at the Glider DAC’s own ERDDAP. AQUAVIEW standardises discovery and metadata; it doesn’t hide the source — so we append ERDDAP’s native tabledap query to pick columns and a time window, and only that slice is transferred. Always constrain the time window — without it you’d ask for the whole three-month deployment.

T0, T1 = "2025-08-01T00:00:00Z", "2025-08-15T00:00:00Z"
url = (g.assets["csv"].href +
  "?time,latitude,longitude,depth,temperature,salinity"
  f"&time>={T0}&time<={T1}")
track = pd.read_csv(url, skiprows=[1],       # row 2 = units
                    parse_dates=["time"])

url <- paste0(g$assets$csv$href,
  "?time,latitude,longitude,depth,temperature,salinity",
  "&time>=", T0, "&time<=", T1)
track <- readr::read_csv(url, skip = 2, col_names = ...)

84,176 measurements over 13 days
  latitude    18.827 → 20.994      longitude  −158.135 → −156.176
  depth        −0.2 → 907.6 m      temperature  4.61 → 29.38 °C

The same comparison the CoastWatch glider tutorial makes — except both datasets came through one catalog. Five steps: find the blended SST by searching the box in COASTWATCH (noaacwBLENDEDsstDaily, night-only — fair against a glider’s top 10 m); open it lazily via OPeNDAP (strip .nc — the one character that matters; unconstrained it 413s); subset to the track box + window; daily-mean the glider’s shallowest 10 m; match ±0.1° pixels per day.

item = cat.get_collection("COASTWATCH").get_item("noaacwBLENDEDsstDaily")
ds  = xr.open_dataset(item.assets["nc"].href
                      .removesuffix(".nc"))     # OPeNDAP, lazy
sst = ds["analysed_sst"].sel(latitude=…, longitude=…,
      time=slice(T0[:10], T1[:10])).load() - 273.15
daily = track[track.depth <= 10].resample("1D", on="time").mean()
daily["sat"] = [satellite_at(t, r.lat, r.lon)      # ±0.1° box
                for t, r in daily.iterrows()]

nc <- nc_open(sub("\\.nc$", "", item$assets$nc$href))
# ncdf4 reads by index: find the lat/lon/time index ranges,
# ncvar_get(start=, count=) only the track box + window,
# then the same daily ≤10 m mean and ±0.1° match-up.

mean bias    −0.039 °C   (glider minus satellite)
RMSE          0.111 °C
correlation   0.941
n             14 days

Everything so far you could have done against one ERDDAP server. This you cannot: the same box and window, asked of sources that have nothing to do with each other — one loop.

track_box = [lon.min()-PAD, lat.min()-PAD, lon.max()+PAD, lat.max()+PAD]
for cid in ["COASTWATCH","PacIOOS","MARINECADASTRE_AIS",
            "OBIS","CERULEAN","IOOS"]:
    n = cat.search(bbox=track_box, collections=[cid],
                   datetime=WHEN).matched()

for (cid in providers)
  n <- (cat |> stac_search(bbox = track_box,
         collections = cid, datetime = when, limit = 1) |>
         get_request())$numberMatched

   89  COASTWATCH           satellite SST / ocean colour
   49  PacIOOS              wave + circulation models
   26  MARINECADASTRE_AIS   vessel traffic (noise context for PAM)
   39  OBIS                 species occurrence datasets
  235  CERULEAN             SAR oil-slick detections
    6  IOOS                 other gliders in the same water

The bounding box is the search area. Left: the concept — your box and a glider track over the islands. Right: that exact view live on Explore (note it counts what’s in the viewport, so its number differs from matched(), which counts the whole box). Click to open the real map.

conceptYour search area — bbox + glider track

liveThe same view in AQUAVIEW Explore

Preconfigured deep-link — centred on the fleet with SST, waves, OBIS, AIS & glider layers already on. Open in AQUAVIEW →

AQUAVIEW’s MCP server is live at https://mcp.aquaview.org/mcp. Point any MCP client at it — in Claude Desktop, add it under mcpServers — then ask in plain English and you get the same STAC items you just found by hand. Same catalog, third door.

{ "mcpServers": { "aquaview": {
    "url": "https://mcp.aquaview.org/mcp" } } }

# literally, in the AQUAVIEW chat:
"What SST and vessel-traffic data
 is near 20°N 157°W
 in August 2025?"

SST · HF-radar currents · AIS vessel traffic — described in plain language,
using the same items you searched by hand. Same catalog, third door.