Six ways the land forecast can evolve

A single forecast gives one answer. An ensemble is more useful when the question is how many plausible answers remain.

The new nwmmediumlandens dataset brings NOAA's National Water Model medium-range land ensemble to GribStream. It follows snowpack, near-surface soil saturation, and evapotranspiration across the 1 km CONUS+ grid with six time-lagged members forced by GFS.

A time-lagged view of medium-range uncertainty

NOAA produces four medium-range cycles each day. The six members are time-lagged and GFS-forced: they represent land forecasts driven by forcing from different recent GFS cycles rather than six interchangeable copies of one deterministic run.

That distinction matters. The member spread can reveal places where the forecast land state is sensitive to the atmospheric forcing represented by the ensemble. A narrow range means the members remain close; a broad range highlights disagreement worth investigating. The spread is useful guidance, but it is not by itself a calibrated probability.

The forecast lengths are asymmetric:

• Member 1 reaches forecast hour 240, or ten days.
• Members 2–6 reach forecast hour 204, or 8.5 days.
• Land fields are available at 3-hour lead-time intervals.

Comparisons using all six members therefore extend through hour 204. Beyond that point, only member 1 remains.

Six land variables on one consistent grid

Every member carries the same compact set of fields:

• SOILSAT_TOP — saturation fraction in the top two soil layers, approximately the upper 0.4 m.
• SNEQV — snow water equivalent.
• SNOWH — physical snow depth.
• FSNO — fraction of the ground covered by snow.
• SNOWT_AVG — average snow temperature weighted by snow-layer mass.
• ACCET — total evapotranspiration accumulated during the forecast.

Together they let users compare not only how much snow is forecast, but how much water it stores, how wet the near-surface soil may become, and how much moisture returns to the atmosphere. Those relationships matter for snowmelt scenarios, agricultural water planning, winter operations, and watershed situational awareness.

The grid contains 4,608 by 3,840 cells in a Lambert conformal projection with 1 km spacing. It covers the contiguous United States and adjacent contributing areas of Canada and Mexico.

Land state, not streamflow

This collection contains NWM's regular land grid. It does not include the separate channel-routing, reservoir, or higher-resolution terrain-routing outputs, and it should not be described as a streamflow dataset.

For the immediate land forecast, nwmshortland provides a new deterministic cycle every hour through hour 18. The medium-range ensemble answers a different question: how snow and soil conditions may diverge over the following week as the forcing scenarios separate.

GribStream indexes compatible source objects from January 1, 2025 onward. NOAA describes the AWS source as a rolling four-week archive, although older objects are currently retained; history beyond that stated window should be treated as opportunistic rather than guaranteed permanent availability. Operational delays, missing cycles, and future NWM upgrades can also create gaps.

The NWM medium-range land ensemble inventory lists the exact selectors, units, members, lead times, and currently indexed dates.

Authoritative sources:

• NOAA National Water Model overview: https://water.noaa.gov/about/nwm
• NOAA NWM output-file contents: https://water.noaa.gov/about/output_file_contents
• NOAA NWM Registry of Open Data on AWS: https://registry.opendata.aws/noaa-nwm-pds/