{"success":true,"data":{"author":"AODN","bucketAccessType":"public","bucketListable":true,"bucketName":"aodn-cloud-optimised","bucketPrefix":"radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr","category":"Earth Observation & Geospatial","cloudProvider":"aws","createdAt":"2026-10-04T00:00:00.000Z","description":"The South Australia Gulfs (SAG) HF ocean radar system covers about 40,000 square kilometres bounded by Kangaroo Island to the east and the Eyre Peninsula to the north, a dynamic region where remnants of the Leeuwin Current arrive from the west, water of varying density exchanges with Spencer Gulf and Gulf St Vincent, and upwelling occurs from the deep ocean on the southern side. It is a key area for aquaculture, fishing and shipping, and links the Great Australian Bight to the Bonney Coast. The radar is a WERA phased-array system with 16-element receive arrays at Cape Wiles (34.943 S, 135.681 E) and Cape Spencer (35.294 S, 136.879 E), operating at 8.512 MHz with a maximum range of 200 km; surface currents are measured within the coverage area. This store holds the real-time, non-quality-controlled (FV00) hourly averaged vector current maps as a single zarr cube: on 2026-10-04 it held 77,034 hourly maps on a 74 x 102 grid from 2009-11-23 to 2026-03-23, with a 385 day gap before the 2025-03-12 restart that covers the algal bloom period. It is the only HF-radar surface current record covering the 2025 Karenia harmful algal bloom period in the South Australian gulfs, since the delayed-mode sibling ends in February 2024.","displayTitle":"Ocean Radar - South Australian gulfs site - Sea water velocity - Real time","fuseMountBrowsable":true,"id":"e734d7da-fd62-5739-a268-bf960d73e3cd","license":"http://creativecommons.org/licenses/by/4.0/","objectCountEstimate":13,"probeStatus":"ok","region":"ap-southeast-2","schemaCard":{"sampleRows":[{"TIME":"2025-05-15 12:30:00","LONGITUDE":135.9449,"UCUR_quality_control":"3 or 4 (all 2025+ values)","LATITUDE":-35.1479,"UCUR":-0.085,"VCUR":-0.0944},{"TIME":"2025-05-15 12:30:00","LATITUDE":-35.5444,"UCUR":0.128,"VCUR":-0.1712,"LONGITUDE":135.7217},{"LONGITUDE":135.811,"VCUR":-0.3639,"TIME":"2025-05-15 12:30:00","LATITUDE":-35.941,"UCUR":0.2136}],"accessHints":[{"language":"bash","snippet":"# Task: look at the store root and one variable directory (do not list the whole store recursively in a notebook; 6,799 objects)\naws s3 ls --no-sign-request --region ap-southeast-2 s3://aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/\naws s3 ls --no-sign-request --region ap-southeast-2 s3://aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/UCUR/ | head -5\n# full schema + global attributes (abstract, metadata_uuid, license) in one 19 KB read\naws s3 cp --no-sign-request --region ap-southeast-2 s3://aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/.zmetadata - | python3 -c \"import json,sys; m=json.load(sys.stdin)['metadata']; print(m['.zattrs']['title']); print({k:v['shape'] for k,v in m.items() if k.endswith('.zarray')})\""},{"language":"python-local","snippet":"# Task: open at the mount with xarray in the geospatial image (zarr 3 reads this zarr-v2 store); drop the object-dtype 'filename' array\nimport xarray as xr\nds = xr.open_zarr('/mnt/archive', consolidated=True, drop_variables=['filename'])\nprint(ds.sizes, '|', str(ds.TIME.values[0])[:16], '->', str(ds.TIME.values[-1])[:16])\n# dimension names are UPPERCASE: TIME, LATITUDE, LONGITUDE. LATITUDE runs north->south (-34.82 .. -37.46)."},{"language":"python-anonymous","snippet":"# Task: open anonymously with plain zarr (works with zarr 2 and 3), decode TIME, and report what period the store really holds\nimport numpy as np, pandas as pd, zarr, s3fs\nfs = s3fs.S3FileSystem(anon=True, client_kwargs={'region_name': 'ap-southeast-2'})\ng = zarr.open_consolidated(fs.get_mapper('aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr'), mode='r')\nt = pd.DatetimeIndex(pd.Timestamp('1950-01-01') + pd.to_timedelta(g['TIME'][:], unit='D')).round('min')   # units: days since 1950-01-01\nprint(len(t), 'hourly maps,', t[0], '->', t[-1], '| UCUR shape', g['UCUR'].shape, 'chunks', g['UCUR'].chunks)\nprint('maps per year since 2020:', pd.Series(t.year).value_counts().sort_index().loc[2020:].to_dict())"},{"language":"python-anonymous","snippet":"# Task: one hour's surface-current map (UCUR, VCUR): costs two ~0.4 MB chunks, about 4 s anonymously\nimport time, numpy as np, pandas as pd, zarr, s3fs\nfs = s3fs.S3FileSystem(anon=True, client_kwargs={'region_name': 'ap-southeast-2'})\ng = zarr.open_consolidated(fs.get_mapper('aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr'), mode='r')\nt = pd.DatetimeIndex(pd.Timestamp('1950-01-01') + pd.to_timedelta(g['TIME'][:], unit='D')).round('min')\ni = int(np.searchsorted(t.values, np.datetime64('2025-05-15T12:30')))\nt0 = time.time(); u, v = g['UCUR'][i], g['VCUR'][i]\nlat, lon = g['LATITUDE'][:], g['LONGITUDE'][:]          # LATITUDE is DESCENDING (-34.82 -> -37.46)\nprint(t[i], 'read', round(time.time() - t0, 1), 's; valid cells', int(np.isfinite(u).sum()), 'of', u.size, '; max speed', round(float(np.nanmax(np.hypot(u, v))), 2), 'm/s')\nii, jj = np.where(np.isfinite(u))\nprint('first valid cell: lat', lat[ii[0]], 'lon', lon[jj[0]], 'UCUR', round(float(u[ii[0], jj[0]]), 3), 'VCUR', round(float(v[ii[0], jj[0]]), 3))"},{"language":"python-local","snippet":"# Task: SA-gulfs bloom-period subset with xarray at the mount: 2025-03-01 onward, lat -36.5..-32, lon 135..139.5 (slice LATITUDE north->south because it is descending)\nimport time, xarray as xr\nt0 = time.time()\nds = xr.open_zarr('/mnt/archive', consolidated=True, drop_variables=['filename'])\nsub = ds[['UCUR', 'VCUR']].sel(TIME=slice('2025-03-01', None), LATITUDE=slice(-32.0, -36.5), LONGITUDE=slice(135.0, 139.5)).load()\nprint(sub.sizes, round(time.time() - t0, 1), 's')      # measured 2026-10-04 with plain zarr: (5769, 47, 56), 243 MB, 67 s anonymously\nspeed = (sub.UCUR ** 2 + sub.VCUR ** 2) ** 0.5\nprint(speed.mean(['LATITUDE', 'LONGITUDE']).resample(TIME='1MS').mean().to_series().round(3))"},{"snippet":"# Task: SA-gulfs bloom-period subset (2025-03-01 onward, lat -36.5..-32, lon 135..139.5) of UCUR+VCUR with plain zarr; ~60 chunks per variable\nimport time, numpy as np, pandas as pd, zarr, s3fs\nfs = s3fs.S3FileSystem(anon=True, client_kwargs={'region_name': 'ap-southeast-2'})\ng = zarr.open_consolidated(fs.get_mapper('aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr'), mode='r')\nt = pd.DatetimeIndex(pd.Timestamp('1950-01-01') + pd.to_timedelta(g['TIME'][:], unit='D')).round('min')\nlat, lon = g['LATITUDE'][:], g['LONGITUDE'][:]\ni0 = int(np.searchsorted(t.values, np.datetime64('2025-03-01')))\nli = np.where((lat <= -32.0) & (lat >= -36.5))[0]; lj = np.where((lon >= 135.0) & (lon <= 139.5))[0]\nt0 = time.time()\nu = g['UCUR'][i0:, li[0]:li[-1] + 1, lj[0]:lj[-1] + 1]; v = g['VCUR'][i0:, li[0]:li[-1] + 1, lj[0]:lj[-1] + 1]\nprint('subset', u.shape, '(TIME, LATITUDE, LONGITUDE) from', t[i0], 'read', round(time.time() - t0, 1), 's,', round(2 * u.nbytes / 1e6), 'MB float64 in RAM')\nspeed = pd.Series(np.nanmean(np.hypot(u, v).reshape(len(u), -1), axis=1), index=t[i0:])\nprint('days with data', int(speed.resample('D').count().gt(0).sum()), '| monthly mean speed m/s:', speed.resample('MS').mean().round(3).dropna().to_dict())","language":"python-anonymous"},{"language":"python-anonymous","snippet":"# Task: pick the right store for a date. Delayed-mode (QC'd, FV01) sibling ends 2024-02-20; this realtime store resumes 2025-03-12. Read both TIME axes.\nimport pandas as pd, zarr, s3fs\nfs = s3fs.S3FileSystem(anon=True, client_kwargs={'region_name': 'ap-southeast-2'})\nfor name in ['radar_SouthAustraliaGulfs_velocity_hourly_averaged_delayed_qc', 'radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc']:\n    g = zarr.open_consolidated(fs.get_mapper(f'aodn-cloud-optimised/{name}.zarr'), mode='r')\n    t = pd.DatetimeIndex(pd.Timestamp('1950-01-01') + pd.to_timedelta(g['TIME'][:], unit='D')).round('min')\n    print(f\"{name.split('averaged_')[1]:16s}\", len(t), 'maps', t[0].date(), '->', t[-1].date(), '| maps in 2025+:', int((t >= '2025-01-01').sum()))"},{"language":"python-anonymous","snippet":"# Task: a full-length point series touches EVERY time chunk (chunks are 100 h x full grid) = ~250 MB per variable, ~2 min anonymously. Prefer maps or short windows.\nimport time, numpy as np, pandas as pd, zarr, s3fs\nfs = s3fs.S3FileSystem(anon=True, client_kwargs={'region_name': 'ap-southeast-2'})\ng = zarr.open_consolidated(fs.get_mapper('aodn-cloud-optimised/radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr'), mode='r')\nt = pd.DatetimeIndex(pd.Timestamp('1950-01-01') + pd.to_timedelta(g['TIME'][:], unit='D')).round('min')\ni0 = int(np.searchsorted(t.values, np.datetime64('2025-03-01')))      # one year window instead of 2009-2026\nt0 = time.time(); s = g['UCUR'][i0:, 20, 62]                           # grid cell near (-35.54, 135.72)\nprint('UCUR point series since 2025-03:', s.size, 'hours,', int(np.isfinite(s).sum()), 'with data, read', round(time.time() - t0, 1), 's')"}],"rowCount":"TIME 77,034 x LATITUDE 74 x LONGITUDE 102 per variable (7,548 grid cells, 581,452,632 cells per 3-D array); nine 3-D variables","columns":[{"name":"TIME","type":"float64 [77034], 'days since 1950-01-01T00:00:00+00:00', gregorian, axis T; decodes to hourly values at HH:30 UTC (middle of the averaging hour), 2009-11-23 17:30 to 2026-03-23 01:30; monotonic, no NaN; 2,505 gaps > 1 h, the largest 2021-11-24 to 2023-08-11 (625 d), 2024-02-20 to 2025-03-12 (386 d), all of 2015 (365 d), 2020-12-04 to 2021-09-08 (278 d)"},{"type":"float64 [74], degrees_north, WGS84, DESCENDING from -34.8234 to -37.4552, step 0.03605","name":"LATITUDE"},{"type":"float64 [102], degrees_east, WGS84, ascending 132.9540 to 137.4627, step 0.04464","name":"LONGITUDE"},{"name":"UCUR","type":"float64 [TIME,LATITUDE,LONGITUDE], m s-1, eastward_sea_water_velocity, hourly mean after rejection of obvious bad data, valid -10..10, NaN fill; ancillary NOBS1 NOBS2 UCUR_quality_control; 93% NaN over sampled chunks (97% in 2025)"},{"name":"VCUR","type":"float64 [TIME,LATITUDE,LONGITUDE], m s-1, northward_sea_water_velocity, hourly mean, same masking as UCUR; speeds observed 0.005-0.50 m/s on 2025-05-15 12:30, bloom-period median 0.17, p99 0.56"},{"type":"float64 [TIME,LATITUDE,LONGITUDE], m s-1, standard deviation of the velocity component over the hour; observed 0.0-0.095 on 2025-05-15 12:30","name":"UCUR_sd / VCUR_sd"},{"name":"UCUR_quality_control / VCUR_quality_control","type":"float64 [TIME,LATITUDE,LONGITUDE], IMOS/IODE flags 0 no_qc_performed, 1 good, 2 probably_good, 3 bad_potentially_correctable, 4 bad, 5 value_changed, 8 interpolated, 9 missing; NaN where the velocity is NaN. Observed regimes: 2009-2014 almost all 0 (plus a few 3/4); 2023-08 to 2024-02 a mix of 1/2/3/4; from 2025-03-22 onward ONLY 3 and 4 (e.g. 15,924 x3 and 8,036 x4 in the chunk starting 2025-03-22; 97 x3 and 87 x4 on the 2025-05-15 12:30 map)"},{"type":"float64 [TIME,LATITUDE,LONGITUDE], degrees 0-180, radar beam intersection angle (geometric dilution of precision); defined on all 7,548 cells even where UCUR is NaN; attrs say flag >= 3 if angle in [150,160) or (20,30], >= 4 if >= 160 or <= 20","name":"GDOP"},{"name":"NOBS1 / NOBS2","type":"float64 [TIME,LATITUDE,LONGITUDE], count of radial observations in the hour from station 1 (Cape Wiles) and station 2 (Cape Spencer); observed 1-3"},{"type":"object (vlen-utf8) [TIME], source NetCDF name per hour, e.g. IMOS_ACORN_V_20250515T123000Z_SAG_FV00_1-hour-avg.nc; drop_variables=['filename'] when opening with xarray/zarr 3","name":"filename"}],"verifiedRead":{"family":"zarr","path":".","facts":["the mount root is a Zarr store: open it with xarray.open_zarr on the root, not the chunk files","under 'radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/': 13 directories (e.g. GDOP, LATITUDE, LONGITUDE, NOBS1); 3 objects named like .zattrs, 24 B to 19 KB","under 'radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/filename/': 108 objects named like ##, 51 B to 4 KB"],"replacedStarterCell":"# SA gulfs HF radar, realtime non-QC surface currents: the bloom-period subset (2025-03-01 onward) over the SA bbox, with timing\nimport time, numpy as np, pandas as pd, xarray as xr\nt0 = time.time()\nds = xr.open_zarr('/mnt/archive', consolidated=True, drop_variables=['filename'])   # zarr v2 store, UPPERCASE dims\nsub = ds[['UCUR', 'VCUR']].sel(TIME=slice('2025-03-01', None), LATITUDE=slice(-32.0, -36.5), LONGITUDE=slice(135.0, 139.5))\nprint('subset', dict(sub.sizes), 'from', str(sub.TIME.values[0])[:13], 'to', str(sub.TIME.values[-1])[:13], '(store ends 2026-03-23)')\nsub = sub.load()                                        # ~60 chunks x 2 variables, 243 MB float64\nspeed = np.hypot(sub.UCUR, sub.VCUR)\nvalid = np.isfinite(sub.UCUR).sum(['LATITUDE', 'LONGITUDE'])\nprint('hours', int(sub.TIME.size), '| mean valid cells per hourly map', round(float(valid.mean()), 1), 'of', sub.LATITUDE.size * sub.LONGITUDE.size)\nmonthly = speed.mean(['LATITUDE', 'LONGITUDE']).resample(TIME='1MS').mean().to_series().round(3)\nprint(monthly.to_string())                              # mean surface speed (m/s) over cells with data, 2025-03 .. 2026-03\nprint('done in', round(time.time() - t0, 1), 's; every value here carries QC flag 3 or 4 in this FV00 stream, see caveats')","starterCell":{"error":"GroupNotFoundError: No group found in store '/mnt/archive' at path ''","seconds":1.92,"ok":false},"image":"531775725781.dkr.ecr.us-east-1.amazonaws.com/scigantic-notebook-geospatial:latest","library":"xarray/zarr","rssDeltaMb":20.4,"snippet":"import os\nimport xarray as xr\nds = xr.open_zarr(os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], '.'))\nprint(ds)","verifiedAt":"2026-10-04T20:49:50.756Z","summary":"xarray Dataset dims [TIME=77034, LATITUDE=74, LONGITUDE=102], 10 variables e.g. ['GDOP', 'NOBS1', 'NOBS2', 'UCUR', 'UCUR_quality_control', 'UCUR_sd', 'VCUR', 'VCUR_quality_control']","attempts":[{"ok":false,"error":"ValueError: binary content, not text","seconds":0.86,"path":"filename/45","family":"text-head"},{"ok":true,"path":"filename/45","seconds":0.65,"family":"bytes-head"},{"path":"filename/45","seconds":7.52,"error":"ValueError: did not find a match in any of xarray's currently installed IO backends ['netcdf4', 'h5netcdf', 'scipy', 'cfradial1', 'datamet', 'furuno', 'gamic', 'gini', 'hpl', 'iris', 'metek', 'nexradlevel2', 'odim', 'rainbow', 'rasterio', 'uf', 'zarr']. Consider explicitly selecting one of the installed engines","family":"netcdf-xarray","ok":false},{"family":"text-head","seconds":0.64,"error":"ValueError: binary content, not text","path":"filename/91","ok":false},{"path":"filename/91","ok":true,"family":"bytes-head","seconds":0.85},{"seconds":4.67,"family":"netcdf-xarray","ok":false,"path":"filename/91","error":"ValueError: did not find a match in any of xarray's currently installed IO backends ['netcdf4', 'h5netcdf', 'scipy', 'cfradial1', 'datamet', 'furuno', 'gamic', 'gini', 'hpl', 'iris', 'metek', 'nexradlevel2', 'odim', 'rainbow', 'rasterio', 'uf', 'zarr']. Consider explicitly selecting one of the installed engines"},{"family":"zarr","path":".","seconds":53.14,"ok":true}],"seconds":53.14},"overview":"IMOS Ocean Radar (ACORN) South Australia Gulfs (SAG) WERA HF radar, hourly averaged surface current vectors, real-time non-QC FV00 stream (GeoNetwork d83ef640-827e-4990-8c35-7c1a1aa587a9; the delayed-mode FV01 sibling is cb2e22b5-ebb9-460b-8cff-b446fe14ea2f). One consolidated zarr v2 cube, 6,799 objects, 1.43 GB, measured 2026-10-04: TIME 77,034 hourly steps (values at HH:30, the middle of the averaging hour) from 2009-11-23 17:30 to 2026-03-23 01:30 UTC, LATITUDE 74 (-34.82 to -37.46, descending, 0.036 deg) x LONGITUDE 102 (132.95 to 137.46, 0.0446 deg), i.e. roughly 4 km cells offshore of the Eyre Peninsula, Spencer Gulf mouth and western Kangaroo Island. Coverage is sparse: 77,034 of the 143,121 hours between first and last are present, with 2,505 gaps, and only 2-8% of grid cells hold a value in a given hour (184 of 7,548 on 2025-05-15 12:30). Access direction: chunks are [100 TIME, 74, 102], full grid per chunk, ~0.36 MB compressed, so the store wants MAPS and TIME WINDOWS. One hour of UCUR+VCUR costs two chunks, 4-11 s anonymously. A full-length point series touches all 771 time chunks of a variable (253 MB on disk), 127 s. The SA bloom subset (2025-03-01 onward, 5,769 hours, 59 chunks per variable) loaded UCUR+VCUR in 18 s full-width / 67 s with the bbox slice, 243-697 MB float64 in RAM; spatial slicing saves RAM but not I/O because every chunk spans the whole grid. All arrays are float64 with NaN fill; 49 of 771 UCUR chunks are absent (all-NaN hours in gaps).","sourceObjectKey":"radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/GDOP/.zarray","generatedAt":"2026-10-04T00:00:00Z","caveats":["Scale: this 'realtime' store was last extended on 2026-04-10 and its last map is 2026-03-23 01:30 UTC; on 2026-10-04 it was six months stale. Check ds.TIME[-1] before promising a user current conditions.","Quirk (image): the geospatial notebook image built before 2026-10-05 ships s3fs 0.4.2 (a 2020 synchronous release) next to zarr 3.1.6, so EVERY s3:// open through xarray or zarr there fails with \"Session.__init__() got an unexpected keyword argument 'asynchronous'\" (measured 2026-10-04 in a real pod). On that image read the FUSE mount instead: xr.open_zarr('/mnt/archive', consolidated=True, ...) works but the open took 39-240 s for these stores in the probe. `pip install -U s3fs` in the pod also fixes it. The starterCell already reads /mnt/archive; the python-anonymous zarr.open_consolidated(fs.get_mapper(...)) snippets need a modern s3fs.","Departure from canonical: delayed vs realtime hand-off. The QC'd delayed-mode sibling (radar_SouthAustraliaGulfs_velocity_hourly_averaged_delayed_qc.zarr, 'Ocean Radar - South Australian gulfs site - Sea water velocity - Delayed mode') runs 2009-11-23 17:30 to 2024-02-20 06:30 and has NO 2025+ data. This realtime store duplicates that whole period (same 2009-11-23 start, same gaps, 77,034 vs 74,523 maps; it additionally holds 2015 which delayed lacks) and then resumes on 2025-03-12 00:30 after a 385 day gap, ending 2026-03-23. Rule: for anything up to 2024-02-20 use the delayed store and keep flag 1/2; for 2025-03-12 to 2026-03-23 (the algal bloom period) this store is the only source.","DO NOT: filter this store on quality_control == 1. It is FV00 ('no quality flags are associated with the data' per file_version_quality_control), yet the flag arrays are populated: 2009-2014 values carry 0 (no_qc_performed), the 2023-08 to 2024-02 segment carries 1/2/3/4, and EVERY value from 2025-03-22 onward carries 3 or 4 (measured on chunks 713 and 720 and the 2025-05-15 12:30 map: 184 valid cells, 97 x3, 87 x4, zero 1s). Those 3/4 flags are not explained by the GDOP angle rule in the GDOP attrs (GDOP was 59-151 deg on the flagged cells). Treat the 2025+ flags as 'unassessed real-time' and use GDOP, NOBS1/NOBS2 and UCUR_sd/VCUR_sd for your own screening, or you will get an empty bloom-period dataset.","Quirk: dimension names are UPPERCASE (TIME, LATITUDE, LONGITUDE), unlike the lowercase time/lat/lon of the AODN satellite zarr stores; .sel(time=...) raises KeyError here.","Quirk: LATITUDE is descending (-34.82 north to -37.46 south). With xarray use .sel(LATITUDE=slice(-32.0, -36.5)) (north first); .sel(LATITUDE=slice(-36.5, -32.0)) silently returns zero rows. The SA bbox (lat -36.5..-32, lon 135..139.5) intersects this grid as 47 rows x 56 columns (lat -34.82..-36.48, lon 135.01..137.46), i.e. the eastern half of the footprint only; the gulfs themselves (north of -34.8) and Gulf St Vincent are OUTSIDE the radar footprint, which sits offshore of the gulf mouths.","Quirk: TIME is 'days since 1950-01-01' as float64; decoded values are HH:30 (middle of the averaging hour), so an hourly reindex must be anchored at :30. 77,034 of 143,121 possible hours are present; .resample('D') will produce NaN days inside the 2,505 gaps rather than erroring.","Scale: chunks are [100, 74, 102] (full grid per chunk, ~0.36 MB compressed, ~6 MB decompressed float64). One hourly map = 1 chunk per variable (4-11 s anonymously, 2026-10-04). A full-length single-cell time series = all 771 chunks = 253 MB per variable, 127 s; a 2025-03 onward point series = 59 chunks, 33 s. Load the bloom window once (UCUR+VCUR 2025-03 onward: 18 s full grid / 67 s sliced, 243-697 MB RAM) and slice in memory; all nine 3-D variables for that window would be ~3.1 GB float64 and will not fit a 4 GiB pod.","Quirk: coverage is sparse and shrinking. Over sampled 100-hour chunks 91-97% of cells are NaN; in 2025 a typical hourly map has ~184 valid cells of 7,548 and only 684 cells carry any value in a 100-hour window (1,768 in 2009). 49 of 771 UCUR chunk objects do not exist (all-NaN hours inside gaps), which zarr reads as NaN fill: that is normal, not a broken store.","Quirk: GDOP is defined on all 7,548 cells of every map (radar geometry), so np.isfinite(GDOP) is NOT a data mask; mask on UCUR/VCUR. NOBS1/NOBS2 are small integers (1-3) stored as float64.","Quirk: 'filename' is a zarr-v2 object (vlen-utf8) array along TIME. Open with drop_variables=['filename'] under xarray + zarr 3 (the geospatial image); plain zarr.open_consolidated reads it fine and it tells you the source NetCDF per hour (IMOS_ACORN_V_<UTC>_SAG_FV00_1-hour-avg.nc).","Quirk: the root .zattrs time_coverage_start/_end both read 2026-01-01T00:30:00Z and geospatial_lat/lon_min/max are those of the LAST ingested file, not the cube; the geospatial_* values (-37.45..-34.82, 132.95..137.44) happen to match the grid but the time coverage does not. Read TIME, not the attrs.","DO NOT: cite https://registry.opendata.aws/roda/aodn_radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc/ (404 on 2026-10-04); the GeoNetwork record d83ef640-827e-4990-8c35-7c1a1aa587a9 is the reference, license CC BY 4.0 (in .zattrs and the record), no DOI in the record."],"sampleHeaders":[{"truncatedAt":512,"key":"radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/GDOP/.zarray","bytes":"{\n    \"chunks\": [\n        100,\n        74,\n        102\n    ],\n    \"compressor\": {\n        \"blocksize\": 0,\n        \"clevel\": 5,\n        \"cname\": \"lz4\",\n        \"id\": \"blosc\",\n        \"shuffle\": 1\n    },\n    \"dimension_separator\": \".\",\n    \"dtype\": \"<f8\",\n    \"fill_value\": \"NaN\",\n    \"filters\": null,\n    \"order\": \"C\",\n    \"shape\": [\n        77034,\n        74,\n        102\n    ],\n    \"zarr_format\": 2\n}","encoding":"utf8"},{"truncatedAt":512,"key":"radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/GDOP/.zattrs","bytes":"{\n    \"_ARRAY_DIMENSIONS\": [\n        \"TIME\",\n        \"LATITUDE\",\n        \"LONGITUDE\"\n    ],\n    \"comment\": \"This angle is used to assess the impact of Geometric Dilution of Precision. If angle between [150; 160[ or ]20; 30], QC flag will not be lower than 3. If angle >= 160 or <= 20, then QC flag will not be lower than 4.\",\n    \"long_name\": \"radar beam intersection angle\",\n    \"units\": \"Degrees\",\n    \"valid_max\": 180.0,\n    \"valid_min\": 0.0\n}","encoding":"utf8"}],"format":"v2-hybrid","verifiedReadAttempt":{"starterCell":{"ok":true},"attempts":[{"path":"filename/45","seconds":0.84,"family":"text-head","ok":false,"error":"ValueError: binary content, not text"},{"path":"filename/45","family":"bytes-head","ok":true,"seconds":0.64},{"family":"netcdf-xarray","path":"filename/45","ok":false,"seconds":6.76,"error":"ValueError: did not find a match in any of xarray's currently installed IO backends ['netcdf4', 'h5netcdf', 'scipy', 'cfradial1', 'datamet', 'furuno', 'gamic', 'gini', 'hpl', 'iris', 'metek', 'nexradlevel2', 'odim', 'rainbow', 'rasterio', 'uf', 'zarr']. Consider explicitly selecting one of the installed engines"}],"facts":["the mount root is a Zarr store: open it with xarray.open_zarr on the root, not the chunk files","under 'radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/': 13 directories (e.g. GDOP, LATITUDE, LONGITUDE, NOBS1); 3 objects named like .zattrs, 24 B to 19 KB","under 'radar_SouthAustraliaGulfs_velocity_hourly_averaged_realtime_nonqc.zarr/filename/': 108 objects named like ##, 51 B to 4 KB"],"image":"531775725781.dkr.ecr.us-east-1.amazonaws.com/scigantic-notebook-geospatial:latest","at":"2026-10-04T21:18:52.424Z","outcome":"no-reader"},"mountPath":"/mnt/archive","starterSamplePath":".","sampleObjects":[".zmetadata",".zattrs","TIME/105","UCUR/723.0.0","VCUR/723.0.0","UCUR_quality_control/723.0.0"],"layout":{"pattern":"zarr v2, consolidated (.zmetadata 19 KB, .zattrs, .zgroup at the root) with 13 array directories: GDOP, NOBS1, NOBS2, UCUR, UCUR_sd, UCUR_quality_control, VCUR, VCUR_sd, VCUR_quality_control (each [77034,74,102] float64, chunks [100,74,102], blosc-lz4, dimension_separator '.', keys like UCUR/723.0.0, 57-712 KB each, 722 of 771 chunks present), TIME [77034] chunks [732] (106 keys), filename [77034] object vlen-utf8 chunks [732], LATITUDE [74], LONGITUDE [102]. 6,799 objects, 1,434,479,577 bytes; all chunk objects dated 2026-04-08/10, nothing newer."},"totalObjects":"6,799 objects, 1,434,479,577 bytes (aws s3 ls --recursive --summarize, 2026-10-04)","starterCell":"# SA gulfs HF radar, realtime non-QC surface currents: the bloom-period subset (2025-03-01 onward) over the SA bbox, with timing\nimport time, numpy as np, pandas as pd, xarray as xr\nt0 = time.time()\nds = xr.open_zarr('/mnt/archive', consolidated=True, drop_variables=['filename'])   # zarr v2 store, UPPERCASE dims\nsub = ds[['UCUR', 'VCUR']].sel(TIME=slice('2025-03-01', None), LATITUDE=slice(-32.0, -36.5), LONGITUDE=slice(135.0, 139.5))\nprint('subset', dict(sub.sizes), 'from', str(sub.TIME.values[0])[:13], 'to', str(sub.TIME.values[-1])[:13], '(store ends 2026-03-23)')\nsub = sub.load()                                        # ~60 chunks x 2 variables, 243 MB float64\nspeed = np.hypot(sub.UCUR, sub.VCUR)\nvalid = np.isfinite(sub.UCUR).sum(['LATITUDE', 'LONGITUDE'])\nprint('hours', int(sub.TIME.size), '| mean valid cells per hourly map', round(float(valid.mean()), 1), 'of', sub.LATITUDE.size * sub.LONGITUDE.size)\nmonthly = speed.mean(['LATITUDE', 'LONGITUDE']).resample(TIME='1MS').mean().to_series().round(3)\nprint(monthly.to_string())                              # mean surface speed (m/s) over cells with data, 2025-03 .. 2026-03\nprint('done in', round(time.time() - t0, 1), 's; every value here carries QC flag 3 or 4 in this FV00 stream, see caveats')"},"summary":"The South Australia Gulfs (SAG) HF ocean radar system covers about 40,000 square kilometres bounded by Kangaroo Island to the east and the Eyre Peninsula to the north, a dynamic…","title":"Ocean Radar - South Australian gulfs site - Sea water velocity - Real time","updatedAt":"2026-10-07T13:10:44.835Z","visibility":"public","websiteUrl":"https://catalogue-imos.aodn.org.au/geonetwork/srv/eng/catalog.search#/metadata/d83ef640-827e-4990-8c35-7c1a1aa587a9"}}