Meteorologist, oceanographer, Python solvem probler, experimenter, Raspberry Pi tinkerer, geek dad, manager, space enthusiast, ham radio lover, science columnist, ADHDer, synthetic intellect curious, linux admin cosplayer.

Post 48 — The Hemisphere That Caught Up — PAYOFF RESEARCH

Remit: exact numbers with sources for (1) N–S forecast-skill convergence, (2) observation-impact / FSOI ranking AMSU-A, (3) data-denial / OSE quantifying SH collapse.

Status: COMPLETE (with a few FLAGged figures noted). All numbers below are traced to a primary source fetched during this research unless marked NOT VERIFIED / FLAG.

1. North–South skill CONVERGENCE

Canonical anchor — Simmons & Hollingsworth 2002

A. J. Simmons and A. Hollingsworth, “Some aspects of the improvement in skill of numerical weather prediction”, Quarterly Journal of the Royal Meteorological Society 128 (581), pp. 647–677 (2002). DOI 10.1256/003590002321042135. (Same verification paper cited in Post 47.)

  • Over roughly the three years prior to the paper, 500 hPa geopotential-height forecasts in the Southern Hemisphere improved by about a one-day gain in predictability.
  • SH one-day forecast errors were reduced so much that SH medium-range forecasts became “almost as skilful” as those for the Northern Hemisphere — the canonical statement of the convergence. (Abstract/summary; full text paywalled — WebFetch returned HTTP 402. Numbers taken from the publisher abstract + secondary summaries, so treat the exact “~1 day / almost as skilful” phrasing as abstract-level, not a figure read off the chart.)

The “evolution of forecast skill” chart narrative — ECMWF Tech Memo 711 (English et al. 2013)

  • “Until around 2000 there was a large skill gap between the northern and southern hemisphere forecasts, and then the gap narrowed dramatically.” Cause of the gap: sparse in-situ observations in the SH. Closing after 2000 “reasonable to attribute … to improved satellite observations.”
  • Dee & Uppala (2009) (ERA-Interim vs operations over ~30 years): improved observations explain only about one quarter of the total forecast improvement; the remaining ~3/4 is better data assimilation and/or model. Over that window the only major satellite-data changes were ATOVS/AMSU-A arriving 1998 and AIRS in 2002. (Important nuance for the post: the convergence is satellite-DRIVEN but not satellite-ONLY — variational assimilation, the Post-47 engine, is what let the poorer SH data be used.)

Rough AC magnitudes (context, lower-confidence)

  • Secondary summary of ECMWF/reanalysis charts: through the 1990s, 5-day 500 hPa anomaly correlation rose from ~0.6 → ~0.75 (NH) and ~0.4 → ~0.625 (SH). FLAG — not read off a primary chart; use only as order-of-magnitude color, not as a quoted figure.
  • ECMWF’s modern “headline score” is the lead time at which HRES 500 hPa geopotential anomaly correlation reaches 80%; today NH and SH reach it at nearly the same lead time (the two curves now sit on top of each other). Exact present-day SH-vs-NH day numbers NOT pinned to a primary chart in this research — FLAG if a specific “both reach 80% AC at day X” number is wanted.

2. Observation impact / FSOI — AMSU-A ranked #1

ECMWF — Cardinali FSO/FEC (the headline “~25%” number)

C. Cardinali, “Monitoring the observation impact on the short-range forecast”, QJRMS 135: 239–250 (2009), DOI 10.1002/qj.366 — the original adjoint-based FSO paper. Reproduced with the ranked breakdown in C. Cardinali, “Data Assimilation: Observation Impact on the Short Range Forecast”, ECMWF Lecture Notes, June 2013 (data ~Sep–Oct 2011):

“The largest contribution to decreasing the forecast error is provided by AMSU-A (~25%); IASI, AIRS, AIREP (aircraft) and GPS-RO observations account for ~10% of the total impact respectively. TEMP and SYNOP surface pressure contribute ~5%, followed by AMVs and HIRS (~4%), then ASCAT and DRIBU (~3%). All other observations contribute less than 3%.”

So the ECMWF ranking, by share of total forecast-error reduction (24-h, dry energy norm): | Obs type | Share | |—|—| | AMSU-A | ~25% (largest single system) | | IASI | ~10% | | AIRS | ~10% | | AIREP (aircraft) | ~10% | | GPS-RO | ~10% | | TEMP (radiosonde) | ~5% | | SYNOP surface pressure | ~5% | | AMVs | ~4% | | HIRS | ~4% | | ASCAT | ~3% | | DRIBU (buoy) | ~3% |

Note: per-OBSERVATION (not per-type) impact is largest for conventional in-situ (DROP/DRIBU surface pressure, ships/buoys) — satellites win on VOLUME, in-situ wins per report. AMSU-A is #1 in total contribution.

FEC evolution — ECMWF Tech Memo 711 (English et al. 2013), Fig. 2

Percentage contribution to total forecast-error reduction, May 2012 → May 2013:

  • Microwave sounders (AMSU-A dominated): 26% → 33% (rose with ATMS + Metop-B AMSU-A/MHS).
  • Infrared sounders (AIRS/IASI): 20% → 16%.
  • Scatterometers: 5% → 8% (OSCAT + Metop-B ASCAT). Microwave sounders are the single largest category of the whole observing system.

Adjoint-FSO methodology origin — Langland & Baker 2004

R. H. Langland and N. L. Baker, “Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system”, Tellus A 56, 189–201 (2004) — the paper that introduced adjoint-based observation-impact estimation (with Baker & Daley 2000). Later NRL/NOGAPS FSO runs using this method also rank AMSU-A among the top contributors to forecast-error reduction. (The specific AMSU-A % for L&B is not quoted here — the paper is primarily the method; the ranked-percentage numbers above are ECMWF/Cardinali. Do NOT attribute a specific % to Langland & Baker 2004 without the paper in hand — FLAG.)

Met Office — Joo, Eyre, Marriott 2013

S. Joo, J. Eyre, R. Marriott, “The Impact of MetOp and Other Satellite Data within the Met Office Global NWP System Using an Adjoint-Based Sensitivity Method”, Monthly Weather Review 141, 3331–3342 (2013). Independent adjoint-FSO study; AMSU-A and IASI found to be the largest satellite contributors to short-range forecast-error reduction, consistent with ECMWF. FLAG: exact Met Office AMSU-A percentage NOT verified to the primary paper (paywalled); state qualitatively (“AMSU-A / IASI the leading satellite contributors”) not with a number. (Related: A. Lorenc & R. Marriott, “Forecast sensitivity to observations in the Met Office Global NWP system”, QJRMS 140, DOI 10.1002/qj.2122, 2014.)

3. Data-denial / OSE — the Southern Hemisphere collapse

Primary hard numbers — ECMWF Tech Memo 839 (Bormann et al., “Global observing system

experiments in the ECMWF assimilation system”, 2019). 500 hPa geopotential, two seasons combined:

  • Southern Hemisphere extratropics: microwave (MW) radiances are the DOMINANT impact — withholding them degrades the day-3 forecast error by 11%; statistically significant impact detectable out to day 9. Conventional obs, IR sounders, and GPS-RO each add ~2–3% at day 3.
  • Northern Hemisphere extratropics: conventional obs largest (10% day-3 degradation), then MW (6%); significant out to day 7.
  • The asymmetry (MW dominant in SH, conventional dominant in NH) is the quantitative core of the “hemisphere that caught up” story: the SH is carried by microwave satellite sounders.

Total satellite loss — ECMWF Tech Memo 711 (English et al. 2013), citing Radnóti et al. 2009

  • “Loss of the satellite data would still cause catastrophic degradation in the southern hemisphere, and very significant degradation in the northern hemisphere.”
  • Denial of a single AMSU-A instrument costs ~0.5% in forecast scores (context: how much a single sounder is worth once the full system is in place).

The “≈ 24 hours / ~1 day lost, SH back toward 1980s skill” figure — FLAG

  • A web summary (attributing to ECMWF OSE work) states that removing polar-orbiting satellites increases SH forecast error by ~50% in the short range and ~25–35% in the medium range, “typically equating to 24 hours of lost predictive skill.” NOT VERIFIED to a primary source in this research — could not confirm the exact 50% / 25–35% / 24 h phrasing in a fetched paper. Likely traces to Radnóti/English OSE work (Radnóti, Bauer, McNally, Horányi 2010/2012 ECMWF Tech Memos) but the specific numbers were not read off a primary document. Use with a hedge, or substitute the VERIFIED Tech Memo 839 numbers (11% at day 3, significant to day 9) plus the Simmons & Hollingsworth “~1 day gain” to make the same point on firmer ground.
  • The clean, defensible framing: without satellite radiances the SH analysis reverts toward the in-situ-only regime of ~1980 (the same “poorer observations as we had in 1980” phrase used in Tech Memo 711), i.e. roughly the pre-convergence gap of about one forecast day.

KEY NUMBERS FOR THE POST (verified, quotable)

  • ~1 day: the SH skill gap that closed (Simmons & Hollingsworth 2002); SH became “almost as skilful” as NH.
  • ~2000: when the NH–SH gap “narrowed dramatically” (Tech Memo 711).
  • ~1/4: fraction of 30-yr improvement due to better observations; ~3/4 due to better assimilation/model (Dee & Uppala 2009 via Tech Memo 711) — the variational-DA nuance.
  • ~25%: AMSU-A’s share of total forecast-error reduction, the single largest system (Cardinali FSO; IASI/AIRS/aircraft/GPS-RO each ~10%).
  • 26%→33%: microwave-sounder share of total FEC, 2012→2013 (Tech Memo 711).
  • 11% at day 3, significant to day 9: SH forecast-error degradation from denying microwave radiances — the dominant SH observing system (Tech Memo 839).
  • 10% / 6% at day 3 (NH): conventional vs microwave denial in the NH (Tech Memo 839).
  • 0.5%: cost of denying a single AMSU-A (Tech Memo 711).

Sources (URLs actually fetched or resolved during this research)

  • Simmons & Hollingsworth 2002, QJRMS 128:647–677, DOI 10.1256/003590002321042135 — https://rmets.onlinelibrary.wiley.com/doi/10.1256/003590002321042135 (abstract only; full text HTTP 402 paywalled)
  • ECMWF Tech Memo 711, English, McNally, Bormann et al. (Oct 2013) “Impact of satellite data” — https://www.ecmwf.int/sites/default/files/elibrary/2013/9301-impact-satellite-data.pdf (downloaded + pdftotext; FEC %s, gap-narrowing narrative, single-AMSU-A 0.5%, catastrophic-SH quote)
  • ECMWF Tech Memo 839, Bormann et al. (2019) “Global observing system experiments in the ECMWF assimilation system” — https://www.ecmwf.int/sites/default/files/elibrary/2019/18859-global-observing-system-experiments-ecmwf-assimilation-system.pdf (downloaded + pdftotext; SH 11%/day-3, NH 10%/6%, significance to day 9)
  • Cardinali 2009, QJRMS 135:239–250, DOI 10.1002/qj.366 — https://rmets.onlinelibrary.wiley.com/doi/10.1002/qj.366 (original FSO paper)
  • Cardinali 2013, ECMWF Lecture Notes “Data Assimilation: Observation Impact on the Short Range Forecast” — https://www.ecmwf.int/sites/default/files/elibrary/2013/16937-observation-impact-short-range-forecast.pdf (downloaded + pdftotext; the ranked AMSU-A ~25% / IASI-AIRS-AIREP-GPSRO ~10% breakdown)
  • ECMWF Newsletter 152, “Assessing the impact of observations using observation-minus-forecast residuals” — https://www.ecmwf.int/en/newsletter/152/meteorology/assessing-impact-observations-using-observation-minus-forecast (per-observation vs per-type impact context)
  • Langland & Baker 2004, Tellus A 56:189–201 (adjoint observation-impact method; citation via search, paper not fetched full-text)
  • Joo, Eyre, Marriott 2013, MWR 141:3331–3342 (Met Office adjoint FSO; citation via search, exact AMSU-A % NOT verified)
  • Lorenc & Marriott 2014, QJRMS 140, DOI 10.1002/qj.2122 — https://rmets.onlinelibrary.wiley.com/doi/abs/10.1002/qj.2122 (Met Office FSO, citation)
  • ECMWF “Quality of our forecasts” (headline 80% AC 500 hPa score definition) — https://www.ecmwf.int/en/forecasts/quality-our-forecasts

Not verified / FLAG list

  • Exact modern SH-vs-NH day-N anomaly-correlation values where the curves meet (only the qualitative “curves now coincide” + Simmons–Hollingsworth “~1 day / almost as skilful”).
  • The “50% short-range / 25–35% medium-range / ≈24 h lost” SH-without-satellites figure — web summary only, not confirmed in a fetched primary source; prefer Tech Memo 839’s 11%/day-3.
  • Langland & Baker 2004 and Joo et al. 2013 exact AMSU-A percentages (paywalled; ECMWF ~25% is the solid number).
  • 1990s AC magnitudes (0.6→0.75 NH, 0.4→0.625 SH) — secondary, order-of-magnitude only.