NSW · 1365–2037 m
Where the wet-bulb is already below zero at a model’s grid terrain, the continuous line extends downward using a 6.5 °C/km environmental lapse-rate estimate, with the uncertainty band widened across a 4.5–8.5 °C/km range. It is useful below-terrain guidance, not a model level directly resolved there.
Dashed sections stop at the model’s grid terrain. They are upper bounds: the wet-bulb zero is somewhere at or below that height, but this view does not estimate how far below. Phase below the bound remains uncertain.
The shaded band is a heuristic uncertainty around a selected primary precipitation model’s snow level; switching that selection above switches the band with it. Each coloured snow-level line is emphasised where that model forecasts precipitation. A selected primary receives the strongest emphasis because it also supplies the main histogram and headline. With None selected, no band or headline is singled out, and every wet model line and phased precipitation track has equal weight. Phase colours and totals here are judged at Mid · 1701 m.
Model-grid SWE is ECMWF IFS’s native sf field (parameter 144),
expressed as liquid water equivalent at the model grid cell and its terrain. It is
shown as an independent comparison: it does not alter the resort-elevation wet-bulb
phase analysis and is not a settled-snow depth.
ECMWF IFS 9 km surface is a separate native-resolution deterministic IFS surface sample. Its precipitation bars are intentionally unphased: this feed does not supply the pressure-level temperature/humidity profile used by the snow-level method. Native SWE is shown independently and does not alter the 25 km profile or ensemble products. Sampled cell -36.520, 148.353 · model terrain 1644 m · Wed 30 0400hrs AEST · 1800Z.
ECMWF IFS official 0.25° QPF is one exact ECMWF cycle read from that centre's own
open-data service, not a provider's seamless blend of successive runs — the same
cycle its own snow-level line above is derived from. Its bars are phased: the same producer that reads this run's tp/sf
also reads its t/r/gh at 1000/925/850/700 hPa and surface 2t/2d (2 m relative
humidity is derived from those two — this stream carries no 2 m RH field directly),
so the wet-bulb method above has a genuine same-run profile to work from rather than
a borrowed one. Because only four levels are fetched, the 1000–500 hPa thickness
diagnostic is unavailable for this track specifically — the method never needs it,
but it is worth naming as a gap other tracks on this page do not have. This stream
carries no terrain-height field either, so the profile is anchored to this same
0.25° cell's terrain height as already resolved by the Open-Meteo ECMWF IFS model
beside it, rather than a second, independently derived figure for the same cell.
Bars are drawn across the 3 and 6 hour spans the
accumulations actually cover, at each span's mm/h rate, at NATIVE cadence — plotted,
not interpolated to hourly; the total is over this chart's hours only. This run's
precipitation reaches +360 h, well past them, but its profile — and so its
snow-level line and phased bars — only to +192 h; hours beyond that still
carry precipitation but no phase, drawn unclassified (the official cycle is 00Z/12Z
only, so it is always some hours older than the models' own cycle — see the run
stamp above). Selected cell -36.500, 148.250 on a 0.25° grid · Tue 29 2200hrs AEST · 1200Z.
Its native SWE is snowfall at that cell and the terrain the model gives
it, which at 0.25° is a ~25 km average sitting
well below this resort. A zero there means the model put none down at that
height — it is not a statement about snow at the resort, and the snow-level plot
above is what answers that question.
NOAA AI-GFS 0.25° is NOAA's experimental AI global model, one exact 0.25° cycle
read straight from NOMADS as GRIB2 rather than through a provider. Two things about
it differ from every other track here, and both are stated rather than smoothed over.
It publishes no snowfall field at all — its surface stream carries 10 m
winds, 2 m temperature, mean sea-level pressure and total precipitation, and nothing
else. So there is no native SWE figure beside its total and there never will be; the
snow shown in its bars is this page's own wet-bulb phase split of its precipitation,
which is a different quantity from a model snowfall diagnostic. And it carries no 2 m humidity, so unlike the models beside it the surface cannot
become a level of its profile — that profile starts at the lowest pressure level
above the model's terrain instead. Its profile is TMP/SPFH/
HGT at 1000/925/850/700/600/500 hPa, the same six levels the
ECMWF products on this page supply, with relative humidity derived from specific
humidity at each level. Because 500 hPa is fetched, the 1000–500 hPa thickness
diagnostic is available here — which the exact ECMWF run above cannot offer.
Its terrain is not borrowed: it comes from the GFS 0.25° analysis this model is
initialised on, so the elevation anchoring its profile is the model's own. Bars are
drawn across the 6 hour spans the accumulations
actually cover, at each span's mm/h rate, at NATIVE cadence; the total is over this
chart's hours only. This run's precipitation reaches +384 h, well past them,
but its profile — and so its snow-level line and phased bars — only to +240 h;
hours beyond that carry precipitation but no phase, drawn unclassified. Selected cell -36.500, 148.250 on a 0.25° grid · Wed 30 0400hrs AEST · 1800Z. Its architecture has not been verified against NOAA documentation,
so it is named only as NOAA publishes it.
NOAA HGEFS 0.25° is NOAA's hybrid AI-and-physics ensemble, and this line is its published ensemble mean — not a run, and not a member. Read it as the centre's own summary of members that disagree with each other, which is a systematically different quantity from a single forecast: on AIGEFS over the same window a sum of published means came to more than twice a sum of member medians, and both were correct answers to different questions. Three things about it are absences, and each is stated because each is permanent rather than a gap in this run. It publishes no individual members — the stream carries ensemble statistics only — so there is no P10–P90 beneath this line and there cannot be one. The spread it does publish is a standard deviation, and turning a mean and a standard deviation into a percentile range assumes the members are normally distributed about the mean; precipitation is bounded at zero and skewed, and where that assumption can be checked against real members it fails badly, so this page shows a mean with nothing around it rather than a band no member forecast. Its bars are unclassified, every one of them: the stream does publish the pressure levels the wet-bulb method needs, but a snow level derived from an ensemble MEAN profile is not any member's snow level — averaging temperature and humidity across members that disagree about the pattern yields an atmospheric state none of them forecast — so no phase is claimed and no snow-level line is drawn for this track. And it publishes no snowfall field, so there is no native SWE figure beside its total and there never will be. Bars are drawn across the 6 hour spans the accumulations actually cover, at each span's mm/h rate, at NATIVE cadence; the total is over this chart's hours only. This stream stops at +240 h — its own hard cap rather than a choice made here. Selected cell -36.500, 148.250 on a 0.25° grid · Wed 30 0400hrs AEST · 1800Z. Its architecture has not been verified against NOAA documentation, so it is named only as NOAA publishes it.
All precipitation amounts are mm liquid equivalent
These models are not on the same run. Their initialisation times differ by up to 6 h, so part of any difference you see below is lead time, not disagreement. Compare the shapes, not the hour-by-hour detail.
| Model | Model run | Total | Snow-phase | Uncertain | Rain | Snow share rain snow | State |
|---|---|---|---|---|---|---|---|
| ECMWF IFS | Tue 29 2200hrs AEST · 1200Z · 13 h old | 67.5 mm | 1.2 mm | 1.8 mm | 64.5 mm | 2% | ok |
| ECMWF AIFS | Wed 30 0400hrs AEST · 1800Z · 7 h old | 54.6 mm | 0 mm | 0 mm | 54.6 mm | 0% | partial |
| GFS | Wed 30 0400hrs AEST · 1800Z · 7 h old | 80.1 mm | 4.6 mm | 1.5 mm | 74 mm | 6% | ok |
| ICON | Wed 30 0400hrs AEST · 1800Z · 7 h old | 59 mm | 0 mm | 0 mm | 59 mm | 0% | ok |
| ACCESS-G | — | — | — | — | — | — | missing |
| UKMO Global | Tue 29 2200hrs AEST · 1200Z · 13 h old | 84.3 mm | 0 mm | 1 mm | 83.3 mm | 0% | partial |
| JMA GSM | Tue 29 2200hrs AEST · 1200Z · 13 h old | 64.8 mm | 0 mm | 0.4 mm | 64.4 mm | 0% | partial |
| ARPEGE World | Wed 30 0400hrs AEST · 1800Z · 7 h old | 27.2 mm | 0 mm | 0 mm | 27.2 mm | 0% | partial |
| GDPS | Tue 29 2200hrs AEST · 1200Z · 13 h old | 74.7 mm | 1 mm | 0.4 mm | 73.3 mm | 1% | ok |
This compares the latest two archived forecasts for each model using only their exact shared valid hours. Amounts are mm liquid equivalent. It shows how a forecast changed; it does not assess whether either forecast was correct.
History is still accumulating — one archived snapshot cannot show a revision yet.
History is still accumulating — one archived snapshot cannot show a revision yet.
History is still accumulating — one archived snapshot cannot show a revision yet.
History is still accumulating — one archived snapshot cannot show a revision yet.
Some of this run is missing. What is drawn is real; the gaps are gaps, not zeroes. (missing variables: wind_gusts_10m)
Some of this run is missing. What is drawn is real; the gaps are gaps, not zeroes. (series covers 6 of 7 days)
Some of this run is missing. What is drawn is real; the gaps are gaps, not zeroes. (missing variables: wind_gusts_10m)
Some of this run is missing. What is drawn is real; the gaps are gaps, not zeroes. (series covers 4 of 7 days)
This compares our dry-bulb freezing level against each model’s own published one — like against like. It checks the vertical profile the snow level is built on. It does not validate the snow level: no model here publishes a wet-bulb snow level to check against, and ECMWF IFS and AIFS publish no freezing level at all, so their profiles stand unchecked.
The snow level is interpolated between whichever pressure levels a model publishes above its own grid terrain. ECMWF IFS and AIFS publish 1000/925/850/700/600/500 hPa; GFS and ICON add 950/900/800. At an alpine grid point the 1000 and 925 hPa surfaces are underground and are discarded, which can leave an ECMWF crossing resting on the 850–700 hPa layer alone — roughly 1.5 km of vertical span. Each model’s actual bracket is stated in the primary panel’s own disclosure, and the uncertainty band widens to match.
Snow report · Webcams · Thredbo’s own pages — not forecast data