SFS Beta Seasonal Outlook

Niño-3.4 — all 31 members

Shaded fan: member percentile spread (10–90% and 25–75%). Bold white line: ensemble mean. The observed index leads in from the left (ERSSTv5, 1991–2020 base), and the gold dashes overlay the C3S seven-model multi-model mean from the same issue month for comparison — two independent forecast systems on one chart. Monthly values, anchored at the last observed month; toggle 3-month for the ONI convention.

Derived climate indices — member spread

Beyond ENSO: box and pattern indices computed per member from the same anomaly fields. IOD (Saji DMI), the South Atlantic Subtropical Dipole (Morioka convention, SW − NE), the Niño flavors, and a PDO built honestly for a forecast system: North Pacific SST anomalies (global-mean removed) projected onto the model's own reforecast EOF1 and normalized by the reforecast PC spread, in σ units. Fan: member 10–90% and 25–75%; line: ensemble mean.

Subseasonal degree days — Lower-48, all 31 members

Population-weighted (50 largest metros, base 65 °F) degree days over the subseasonal window — days 16–46, in the beta's 48-hour output steps, every member drawn. Weeks 1–2 are left to medium-range NWP; this is the range only a seasonal system covers with members. The grey band is the model's own hindcast climatology for the same init month and lead day (30 years × 11 members = 330 samples per day) — so a member outside the band is genuinely unusual, not just drifted. Cumulative departure runs the running total of each member minus the hindcast mean: the number a heating/cooling season actually settles on.

Subseasonal teleconnections — regime probabilities vs hindcast

PNA and EPO computed per member across the day-16–46 window (48-hour steps) and standardized against the model's own reforecast for that exact lead day — values are in hindcast-σ units, so the climatological base rate of any threshold is fixed and the member fraction is directly comparable to it. The strip shows the ensemble-mean state at a glance (red positive, blue negative); hover for member probabilities. Click a pill for the full member fan. PNA uses the Wallace–Gutzler centers and EPO the mid-Pacific/Alaska 500-hPa dipole — both are zero-sum center combinations, structurally immune to the uniform part of the beta's daily-stream height artifact. AO and NAO are withheld for now: the diagnosed artifact is polar-amplified (annular, +11 dam >60°N vs +3 midlatitude), which projects directly onto those indices and cannot be separated from real signal; they return when NOAA reconciles the beta's daily stream.

Anomaly maps — step through the subseasonal window

For each valid step (days 16–46): left, the 31-member mean deviation from the model's own hindcast trend, in detrended signal-σ units: the reference is the 1991–2020 reforecast mean plus the model's own linear trend extrapolated to the forecast date, so the background warming doesn't paint the whole map red — what remains is genuine forecast signal. The σ is the variability of hindcast ensemble means about that trend line at the same lead day; ±1 is a normal-strength signal, ±3+ exceptional. Right, the member spread divided by the hindcast spread: below 1 (orange) the ensemble is more confident than climatology, above 1 (purple) less. Scrub through the window to watch where the signal holds and where it dissolves.

500 hPa height

2 m temperature

Precipitation (global)

200-hPa velocity potential (χ)

MJO — 31-member RMM phase space to day 46

The Wheeler–Hendon RMM phase diagram from every SFS member's tropical U850/U200, pushed through the exact same machinery as the AIFS-ENS MJO product — same EOFs, same day-of-year climatology, same 120-day low-frequency filter, same per-mode scaling — so the two forecasts plot in the same coordinates and can be compared directly. This is the full three-channel WH04 projection: the missing OLR channel is built from −standardized model precipitation (tropical rain and OLR anticorrelate closely), standardized against the model's own 1991–2020 reforecast. The wind channels are raw — no bias or drift adjustment — so at week-4+ leads model drift can read as slowly growing amplitude. Teal spaghetti: members (every other day); red: ensemble mean (dashed grey: wind-only reference); black: the observed track from the official BoM RMM index.

Stratosphere — wave flux and the polar vortex, all members

Two classic lower-stratosphere diagnostics from the daily members: the 100-hPa eddy heat flux (45–75°N zonal mean — proportional to the wave activity propagating up into the stratosphere) and the 50-hPa zonal-mean wind at 60°N (the strength of the polar-night jet; sustained easterlies in winter mark a major warming). Every member against the model's own hindcast climatology for the same init month and lead day. Shown for the day-16–46 subseasonal window (48-hour steps). These panels come alive in the cold season, when pulses of wave flux and vortex swings set up weeks-ahead pattern changes.

100 hPa — daily polar view, first 46 days

The lower stratosphere from above: ensemble-mean 100-hPa height anomaly day by day (48-hour steps), with the absolute mean height contoured and the mean 100-hPa wind as vectors. Each frame is measured against the model's own 1991–2020 reforecast per-lead-day mean plus its linear trend evaluated at the forecast year, so the baseline evolves with the forecast. The monthly store stops at 200 hPa, so this is built from the daily members. Bias shared by forecast and reforecast cancels in the anomaly, but the near-real-time daily stream carries a height offset of its own — read the pattern, not the absolute magnitude.

Monthly anomaly maps — step through every lead

The same anomalies without seasonal averaging: one frame per forecast month, in the loop player — scrub or step to watch the pattern develop and decay month by month. Fixed color scale across all leads.