NYOFS: NOAA New York/New Jersey Operational Forecast System

As of 2026-10-06. Grid and file facts on this page were read from the NYOFS output itself (CO-OPS FMRC, NCEI and AWS copies of the 2026-09-30 and 2026-10-01 cycles); the status of each access path was tested live against forcingkit.fetchers.nyofs on the same day.

Warning

NYOFS delivers, but not as a z-v3 parent store. Every access path (CO-OPS FMRC, AWS S3 and NCEI) returns data for 2014 to the present. The output is the legacy (time, depth, eta, xi) layout rather than the z-v3 parent store. See Known issues below; use NECOFS: Northeast Coastal Ocean Forecast System (FVCOM) for the parent ocean in this region until NYOFS writes z-v3.

System overview

NYOFS is the operational nowcast/forecast system of the NOAA Center for Operational Oceanographic Products and Services (CO-OPS) for the Port of New York and New Jersey (implementation plan: NOAA Technical Report NOS CO-OPS 37, 2002). Its grid covers the Upper and Lower Bays, the lower Hudson, the East River to the western end of Long Island Sound, Newark Bay, the Kills and Raritan Bay. It forecasts water level and currents; its output carries no temperature or salinity.

Underlying model

  • Hydrodynamic model: the Princeton Ocean Model (POM), 3D, free surface, as described by Blumberg and Mellor (1987).

  • Grids: two curvilinear grids, a coarse grid for the harbour and its approaches and a fine grid over the Kill van Kull and its approaches.

  • Vertical coordinate: sigma, 7 uniformly spaced levels from 0 to 1 (positive down: 0 at the surface, 1 at the bed).

  • Cycles: four a day, at 05, 11, 17 and 23 UTC.

  • Output: per cycle, one nowcast file holding the 6 hours up to the cycle time (the 23 UTC nowcast holds 18:00 to 23:00, hourly) and one forecast file running forward from it. Files are netCDF3 (classic format), COARDS conventions, time in days since 2008-01-01.

Grids

Grid

Size (ny x nx)

Extent of wet cells (lon/lat)

Cell size over water

Coarse (nyofs)

134 x 73; 2840 wet

[-74.343, 40.417, -73.776, 40.907]

140 to 1040 m, median 340 m

Fine (nyofs_fg)

40 x 126; 2443 wet

[-74.198, 40.631, -74.024, 40.682]

70 to 151 m, median 117 m

forcingkit reads the coarse grid only. The deepest wet cell is 24.8 m on the coarse grid and 16.9 m on the fine grid.

Variables

Both grids use the same fields-file layout:

Name

Dimensions

Meaning

lon, lat

(ny, nx)

Cell positions, degrees.

mask

(ny, nx)

1 for water, 0 for land.

depth

(ny, nx)

Bathymetry, metres, positive down.

sigma

(sigma)

Level positions, 0 to 1, positive down.

zeta

(time, ny, nx)

Water surface elevation, metres.

u, v

(time, sigma, ny, nx)

Eastward and northward water velocity, m/s.

w

(time, sigma, ny, nx)

Upward water velocity, m/s.

air_u, air_v

(time, ny, nx)

Eastward and northward wind forcing, m/s.

u and v are earth-referenced (eastward_sea_water_velocity and northward_sea_water_velocity) and share the (ny, nx) points of lon and lat: they are not on staggered C-grid faces, so no face-to-centre interpolation is needed. Missing values are -99999.0.

Where the data lives

Location

Span

Files

CO-OPS FMRC aggregation

Rolling 7 days (428 hourly steps on 2026-10-06)

https://opendap.co-ops.nos.noaa.gov/thredds/dodsC/NYOFS/fmrc/Aggregated_7_day_NYOFS_Fields_Forecast_best.ncd: the coarse grid as one OPeNDAP dataset, with time, time_run and time_offset.

NCEI THREDDS

2014 to the present month

https://www.ncei.noaa.gov/thredds/dodsC/model-nyofs-files/{yyyy}/{mm}/<file>, all files of a month in one directory. Coarse-grid field nowcasts only (plus station files), four a day. Up to the 2024-09-08 cycles the name is nos.nyofs.fields.nowcast.{yyyymmdd}.t{cc}z.nc; from 2024-09-09 it is nyofs.t{cc}z.{yyyymmdd}.fields.nowcast.nc. A few days are missing (2024-06 has 25 of 30).

AWS S3 noaa-nos-ofs-pds

Complete from 2024-11-19; parts of 2024 before that

From 2024-11-19 (that day holds only the 23 UTC cycle), nyofs/netcdf/{yyyy}/{mm}/{dd}/nyofs.t{cc}z.{yyyymmdd}.fields.{nowcast|forecast}.nc, nyofs_fg... for the fine grid, and station files. Before that, nyofs/netcdf/{yyyymm}/ with the older names, for 2024-04 and 2024-07 to 2024-12 only. A coarse nowcast file is about 5.8 MB, a coarse forecast file about 51 MB.

A continuous hindcast is the chain of nowcast files: cycle t{cc}z supplies hours cc - 5 to cc, so the four files of a day cover it without overlap.

How forcingkit reads it

forcingkit.fetchers.nyofs.fetch_nyofs_boundary_conditions(start_date, duration_hours, bbox):

  1. Domain check. The box must lie inside the declared domain [-74.475, 40.389, -73.743, 40.940], the extent of the coarse grid.

  2. Source. Start times up to 6 days old use the FMRC aggregation. Older windows are read as the chain of nowcast files, one per cycle (_nowcast_cycles). For each cycle the fetcher tries AWS S3 first when the date is 2024-11-19 or later, then NCEI (_nowcast_file_urls). AWS files are netCDF3 and are read with the scipy engine; NCEI is read over OPeNDAP. If any cycle in the window is missing from both, the fetcher returns nothing, so the dispatcher falls back rather than writing a shortened store.

  3. Times. Record times are rounded to the hour (NYOFS stores them up to 15 s off). Where the FMRC series repeats an hour, the first non-missing value is kept.

  4. Subset. The bounding rectangle of the wet cells inside the box is read; cells outside the box or on land are NaN. u and v are passed through as stored, since they are already earth-referenced at the cell positions.

  5. Output. A Dataset with u and v only, dimensions (time, depth, eta, xi). eta and xi are integer indices. depth is linspace(-50, 0, 7), a placeholder for the 7 sigma levels that does not follow the local water depth.

The dispatcher then resamples to hourly, casts to Float32, and writes the store in one piece with schema = "legacy".

Known issues

Tested live on 2026-10-06 with the box [-74.10, 40.60, -74.00, 40.70] (Upper Bay and the Narrows), 3 hours; each returned 4 hourly records x 7 levels x 62 x 29:

Start

Path

Files read

2026-10-04 (2 days old)

FMRC

The aggregation.

2026-09-15 04 UTC

AWS S3, per-day layout

nyofs.t05z and nyofs.t11z.20260915.fields.nowcast.nc.

2024-11-19 20 UTC

AWS S3, first per-day date

nyofs.t23z.20241119.fields.nowcast.nc.

2024-10-15 12 UTC

NCEI, current names

nyofs.t17z.20241015.fields.nowcast.nc.

2023-12-15 06 UTC

NCEI, older names

nos.nyofs.fields.nowcast.20231215.t11z.nc.

Fixed on 2026-10-06: the file names and cycle times (NYOFS publishes one nowcast file per cycle, on 05/11/17/23 UTC, not one file per hour on 00/06/12/18 UTC), the AWS cutover (per-day layout only from 2024-11-19; earlier dates go to NCEI), the netCDF3 reader for AWS files, the velocity averaging (removed: u and v were being smoothed by one cell), the declared domain (it reached 40.2 N and 73.3 W, past the grid, so boxes in that margin ranked NYOFS first and fell back), and duplicate FMRC hours (one copy of a repeated hour can be all fill, which left all-NaN records in the output).

Still open:

  • Output layout. The legacy (time, depth, eta, xi) store has no geographic coordinates and no true depths, and lacks zeta, temp, salt, h and mask, all of which a z-v3 parent store carries. NYOFS has no temp or salt to give, so a z-v3 NYOFS store would need them from another donor. A model that reads z-v3 parents cannot use a NYOFS store until this is done; see Converting to z-v3.

  • Placeholder depths. depth is linspace(-50, 0, 7) in sigma-index order, but NYOFS sigma index 0 is the surface, so the first level is labelled -50 m.

  • Fine grid. The fine grid over the Kill van Kull (about 117 m) is not read.

  • Archive gaps. A window that includes a cycle missing from both AWS and NCEI (2024-06 lacks five days) is not served by NYOFS; the dispatcher falls back to NECOFS.

The /api/v1/obc response’s donor field names the donor that delivered (read from the store’s source attribute), and predicted_donor names the one ranked first.

Converting to z-v3

Scope of the follow-up that would make NYOFS a z-v3 parent, modelled on forcingkit.fetchers.necofs.iter_parent:

  1. Streaming interface. Add iter_parent(start_date, duration_hours, bbox, pad_cells, vertical_spacing_m) yielding the static fields once and then one record per hour, with PARENT_RECORD_DIMS and PARENT_UNITS; the dispatcher already streams any donor that has it. The nowcast chain fits this well: one 6 MB file gives six records.

  2. Static fields. lon, lat (full precision), h from depth, mask, and z/z_face on the requested vertical spacing, over the box padded by pad_cells. The coarse grid is curvilinear, so the regridding onto the parent’s regular lon/lat grid that NECOFS does for its unstructured mesh is needed here too (bilinear on the curvilinear grid, or the same scattered-point interpolator).

  3. Vertical. Convert sigma to depth per cell and hour, z = zeta - sigma * (h + zeta) with sigma positive down, then interpolate u and v onto the fixed z levels. This also removes the placeholder depths.

  4. Tracers. NYOFS has no temp or salt. Options, to be decided: take them from NECOFS over the same box and times (a second donor inside one store, recorded in the attributes), or write a constant-salinity, constant-temperature parent and flag it. The first is more faithful; the second is simpler and adequate for barotropic boundary forcing.

  5. Tests. Unit tests for the sigma-to-z conversion and the regridding on a synthetic curvilinear grid, and a live integration test that streams a few hours and checks the store against the z-v3 contract (store_is_complete, attributes, units).

Rough size: the NECOFS parent path (z_levels_for_depth, sigma_to_z, the scattered-point interpolators, _parent_static and iter_parent) is about 300 lines, with about 180 lines of tests. NYOFS can reuse z_levels_for_depth, sigma_to_z (which, like NYOFS, puts level 0 at the surface and takes the layer depths from step 3), the interpolator and _parent_static, and the dispatcher’s streaming writer needs no change, so the new code is mainly the reader loop and the tracer source.

Python usage

from forcingkit.fetchers.nyofs import fetch_nyofs_boundary_conditions

# A box in the Upper Bay; any start time from 2014 on.
ds = fetch_nyofs_boundary_conditions(
    "2023-12-15T06:00:00Z", 6, [-74.10, 40.60, -74.00, 40.70]
)
# ds.u, ds.v: (time, depth, eta, xi), or None if nothing could be read

Through the dispatcher, which ranks NYOFS first inside its domain and falls back on failure:

from forcingkit.dispatcher import dispatch_obc_request

path = dispatch_obc_request(
    "2026-10-04T00:00:00Z", 6, [-74.10, 40.60, -74.00, 40.70]
)

Tests

  • tests/unit/test_nyofs.py: metadata and domain checks, access-mode resolution, the hour-to-cycle mapping, file names for each archive era, reader choice and fallback between AWS and NCEI, time rounding and FMRC duplicate hours, velocity pass-through, and dispatcher registration. No network.

  • tests/integration/test_ofs_archive_paths.py: live reads of one fixed date per archive path (AWS per-day, NCEI current names, NCEI older names) for NYOFS and DBOFS.

  • tests/integration/test_nyofs_obc_fetch.py::test_nyofs_boundary_conditions: a live fetch 12 hours back, which exercises the FMRC path.

uv run pytest tests/unit/test_nyofs.py -v
uv run pytest tests/integration/test_ofs_archive_paths.py tests/integration/test_nyofs_obc_fetch.py -v

References

Citations

  • Blumberg, A. F. and G. L. Mellor (1987). A description of a three-dimensional coastal ocean circulation model. In N. S. Heaps (ed.), Three-Dimensional Coastal Ocean Models, Coastal and Estuarine Sciences 4, American Geophysical Union, 1-16. doi:10.1029/CO004p0001

  • NOAA (2002). Implementation Plan, Port of New York and New Jersey Operational Forecast System (NYOFS). NOAA Technical Report NOS CO-OPS 37, Silver Spring, MD. PDF

Institutions and data services