Forecasts

Two complementary forecast surfaces: IEM MOS (text/JSON, both SDKs) and gridded NWP (GRIB2, Python today). Both feed dataset(include_forecast=True); Python additionally exposes the per-model NWP path for cross-model comparison.

Overview

                ┌──────────────────────────┐
dataset(...)    │  IEM MOS         (both)  │  →  fcst_* columns
include_forecast│  gridded NWP     (py)    │  →  fcst_*_nwp_<model> columns
                └──────────────────────────┘

• IEM MOS is the default forecast in research(). Text/JSON, station-level point forecasts, fully wired on both SDKs.
• NWP (gridded) wraps NOAA Big Data Program, ECMWF Open Data, and MSC Datamart. 11 NCEP-family models are wired end-to-end in Python today (HRRR, HRRRAK, GFS, GEFS, GDAS, NBM, RAP, RRFS, RTMA, URMA, CFS). 13 others ship URL patterns + QC rules but raise structured errors (see Wiring status below). TypeScript: deferred to v2.0+ — see NWP in TypeScript.

dataset() with forecasts

dataset(include_forecast=True) (and its research() alias) wires both the IEM MOS forecast and the per-NWP-model forecast end-to-end. Settlement-day bucketing uses station LST via settlement_date_for(observed_at, station) so post-midnight tail rows roll into the correct calendar settlement.

Python

import mostlyright

# Default — IEM MOS forecasts populate fcst_* columns
df = mostlyright.dataset(
    "KNYC", "2026-05-01", "2026-05-07",
    include_forecast=True,
)
print(df[["date", "obs_high_f", "fcst_high_f", "fcst_model"]])

# Additionally include NWP forecasts per model
df = mostlyright.dataset(
    "KNYC", "2026-05-01", "2026-05-07",
    include_forecast=True,
    forecast_models=["hrrr", "gfs"],
)
print(df[[
    "date", "fcst_high_f",
    "fcst_high_f_nwp_hrrr", "fcst_high_f_nwp_gfs",
]])

TypeScript

import { dataset } from "mostlyright";

// IEM MOS (NBE default) — TS-wired forecast path
const rows = await dataset("KNYC", "2026-05-01", "2026-05-07", {
  include_forecast: true,
});
console.log(rows[0].fcst_high_f, rows[0].fcst_low_f, rows[0].fcst_model);

NWP models in TS raise NwpNotAvailableError — see NWP in TypeScript.

Standalone IEM MOS

The TS SDK ships iemMosForecasts() as a standalone surface — useful when you want raw MOS rows without going through dataset().

Python

# Python folds IEM MOS into dataset(); call sites that want raw MOS
# use the internal _fetchers (not yet a public surface).
import mostlyright

df = mostlyright.dataset("KNYC", "2026-05-01", "2026-05-07", include_forecast=True)

TypeScript

import { iemMosForecasts } from "@mostlyrightmd/weather";

const rows = await iemMosForecasts("KNYC", "2026-05-01", "2026-05-07", {
  model: "nbe",
});
console.log(rows[0].tempC, rows[0].source); // 20.0, "iem.archive"

Supported NWP models

Wiring status legend — wired = fetch + decode + QC end-to-end. reserved = schema-declared (URL patterns + QC rules + idx dispatch present) but forecast_nwp() raises today. Reserved models flip to wired in follow-up releases.

NCEP family (USA — NOAA BDP + NOMADS)

Model	Wiring	Coverage	Cycle freq	Historical depth	Notes
hrrr	wired	CONUS 3km	hourly	2014-07-30	High-resolution rapid refresh
hrrrak	wired	Alaska 3km	3-hourly	2018-01-01	HRRR for Alaska
gfs	wired	Global 0.25°	6-hourly	2021-01-01	Standard global model
gefs	wired	Global ensemble 32m	6-hourly	2017-01-01	Default member c00; opt-in via member=
gdas	wired	Global 0.25° short	6-hourly	2021-01-01	GFS analysis system
nbm	wired	Regional blend	hourly	2020-01-01	National Blend; fxx=0 auto-bumps to 1
rap	wired	CONUS 13km	hourly	2020-01-01	Rapid refresh
rrfs	wired	CONUS 3km	hourly	2024-01-01	HRRR successor (pre-operational)
rtma	wired	CONUS 2.5km analysis	hourly	2024-01-01	Real-time mesoscale (fxx=0 only)
urma	wired	CONUS 2.5km analysis	hourly	2024-01-01	Un-Restricted MA (fxx=0 only)
cfs	wired	Global 1° (4-member)	6-hourly	2011-01-01	Climate Forecast System

All 11 NCEP-family models wired in Python.

ECMWF family (Global — Open Data)

Model	Wiring	Cycle freq	Historical depth	Notes
ecmwf_ifs_hres	reserved	6-hourly	2022-01-01	Deterministic IFS HRES
ecmwf_ifs_ens	reserved	6-hourly	2022-01-01	Ensemble IFS
ecmwf_aifs_single	reserved	6-hourly	2024-02-25	AI single
ecmwf_aifs_ens	reserved	6-hourly	2024-02-25	AI ensemble

MSC Canadian family

Model	Wiring	Cycle freq	Historical depth	Notes
hrdps	reserved	hourly	live-only	High-Resolution Deterministic Prediction
rdps	reserved	6-hourly	live-only	Regional Deterministic Prediction
gdps	reserved	12-hourly	live-only	Global Deterministic Prediction
geps	reserved	12-hourly	live-only	Global Ensemble Prediction
reps	reserved	12-hourly	live-only	Regional Ensemble Prediction

MSC models raise HistoricalDepthError(archive_depth=None) (live-only Datamart, 24h retention) — different from NwpModelNotAvailableError so callers can branch on the recovery action.

Legacy + experimental

Model	Wiring	Notes
hafs	reserved	Hurricane Analysis & Forecast System
nam	reserved	North American Mesoscale — retires 2026-08-31 (NWS scn26-47)
href	reserved	High-Resolution Ensemble Forecast — retires 2026-08-31
hiresw	reserved	High-Resolution Window — retires 2026-08-31

Calling a retiring legacy model emits DeprecatedModelWarning with the retirement date.

Standalone NWP (Python)

from mostlyright.weather import forecast_nwp
from datetime import datetime, UTC

df = forecast_nwp(
    "KNYC", "hrrr",
    cycle=datetime(2026, 5, 27, 12, tzinfo=UTC),
    fxx=24,
)
print(df[["station", "valid_time", "temp_c"]])

Rate limits / concurrency

• AWS BDP / Google Cloud / Azure — tolerate aggressive parallelism. Default httpx.Limits per FORECAST-LIMITS.md.
• NOMADS — capped at NOMADS_CONCURRENCY_CAP=4. mostlyright fails fast on 403.
• MSC Datamart — undocumented; mostlyright is polite (N≤4).
• ECMWF Open Data — 500 simultaneous connections per IP; mostlyright uses N≤8.

Errors

Error	When it fires
NwpModelNotAvailableError	Reserved model called (ECMWF, HAFS, legacy)
HistoricalDepthError	Cycle older than the model’s archive depth, OR MSC×5 (live-only)
NoLiveForNwpError	The wired mirror has no live cycle reachable
GribIntegrityError	Decoded GRIB2 bytes failed structural validation
DeprecatedModelWarning	NAM / HREF / HiResW after 2026-08-31 retirement

See also

• NWP in TypeScript — the v1.x deferral story + workarounds
• dataset() — composable forecasts via include_forecast=True
• Data quality / QC — physics-bound flags applied to forecasts
• API reference: mostlyright.weather.forecast_nwp · mostlyright.forecasts