How the Sounding Explorer works

Where the data comes from, what each panel shows, and how every value is ranked against the station's own archive. Everything runs client-side in your browser.

← Open the explorer

1. Data sources

A sounding can arrive from four places. The app tries them in order of freshness, and the status line names the source you're looking at.

A balloon takes 90 minutes to two hours to ascend, so no source has a complete 12Z sounding before roughly 13:30 UTC — that is the floor on "real-time", not a delay in the app. The University of Wyoming feed sends no CORS headers, so a scheduled job mirrors ~600 stations six times a day onto a data branch the browser can read; complete UTC days are also bundled into a permanent, growing archive (the 📅 date picker reaches it).

2. The physics engine

All analysis is SHARPlib compiled to WebAssembly (~250 KB) running in your tab — no backend. The same compute_sounding() entry point is compiled natively for the offline climatology builder, so a live value and its historical percentile always come from identical code.

3. What the panels show

Skew-T

Three parcel traces: long-dashed white = most-unstable (MU), dashed orange = surface-based (SB), dotted cyan = 100-hPa mixed-layer (ML). When the most-unstable air is at the surface, MU and SB coincide and draw as one line. The right-edge ticks mark each parcel's LCL/LFC/EL, and the Parcels table's CAPE/CINH rows belong to the same three parcels.

Parcels table — CAPE and ECAPE

ECAPE (Peters et al. 2023) is CAPE with entrainment: updraft dilution is derived from the storm-relative wind and shear, so it uses the kinematic profile as well as the thermodynamic one. The table shows CAPE and ECAPE per parcel plus the implied updraft speeds √(2·CAPE) and √(2·ECAPE) — the gap between them is the buoyancy entrainment takes away. The ECAPE/CAPE ratio can exceed 100%: the formulation nets a storm-relative kinetic-energy gain against the entrainment loss, so a ratio above 1 means vigorous inflow more than pays for the mixing.

Moist static energy

h is the moist static energy at each level and h* what h would be if the level were saturated. A boundary-layer parcel conserves its h, and it is buoyant wherever its h exceeds the environment's h* — that is the layer generating the CAPE. The summary rows give boundary-layer h, the minimum h* aloft, their difference (negative ⇒ conditionally unstable), column-integrated MSE, and the 12-hour change in both.

Hodograph

Height-coloured trace with Bunkers left/right movers. The effective inflow layer — the air the storm can actually ingest — is highlighted directly on the curve, and effective SRH is computed over that layer rather than a fixed 0–1 km. Corfidi MCS vectors are reported in the Kinematics table, not drawn. The effective-layer Bunkers vector is drawn only when it differs from the fixed-layer one by more than 3 kt.

Derived levels

Computed directly from the profile: WMO tropopause (lowest level where the lapse rate drops to ≤ 2 K/km and stays there on average through the next 2 km), cold point (column temperature minimum — the definition that matters in the deep tropics), PBL top (first level whose virtual potential temperature exceeds the surface value by 0.5 K), and column relative humidity (∫q dp ⁄ ∫q_sat dp, with a 700–500 hPa mid-level mean).

Winter

• Dendritic growth zone — the −12 to −18 °C layer, reported with bounds, depth and RH (a saturated DGZ grows efficient snow; a dry one doesn't).
• Snow ratio (Kuchera) — keyed on the warmest temperature in the column.
• Snow squall parameter — SHARPlib's.
• Precipitation type (Bourgouin) — the energy-area method: melting energy of any warm nose vs refreezing energy of the cold layer beneath. Both numbers are shown beside the verdict so you can second-guess it.

Fog and cloud layers

The fog probabilities are empirical: three logistic models (fog, ceiling ≤ 1,000 ft, ceiling ≤ 3,000 ft) trained on 87,568 sounding–METAR pairs from 14 co-located stations over 2006–2025, using features from the lowest 3 km (peak low-level RH is the dominant one). Cross-validated AUC is ~0.9 both year-blocked and station-held-out, which is why it runs globally. The probabilities describe conditions at sounding time — they are not a forecast. The panel also reports the classic FSI screen, the bulk Richardson number (is the surface layer decoupled?), and dew-vs-frost. The cloud-layer detector uses height-dependent RH thresholds (over ice below freezing) calibrated against ~49k ceilometer-verified bases; boxes are coloured by phase and the dashed teal band is the DGZ. Thin or broken cloud between sounding levels can be missed.

4. Climatology — where a value ranks

Each station's entire IGRA period of record is analysed once, offline, and reduced to percentile breakpoints and record extremes at 73 day-of-year anchors, each pooling a ±10-day window across every year on record. The indices ranked: precipitable water, 850/700/500 hPa temperature and dewpoint, 500 hPa height, 1000–500 hPa thickness, freezing level, wet-bulb zero, 850/250 hPa wind speed, ECAPE and SHIP.

Nothing is flagged unless the index has at least 30 samples in the window. Soundings qualify for the climatology only with a real surface dewpoint, ascent to 400 hPa, and dewpoint on ≥60% of levels — gaps are never filled with invented data, and calm no-CAPE days count as ECAPE zeros rather than being skipped, so the percentiles describe all days, not just convective ones.

The 📊 Climatology button opens the annual curve for any ranked variable: shaded 25–75% and 10–90% bands, the median, red/blue record envelopes, and a gold dot marking the sounding on display.

5. The record watch

Six times a day, every station's newest sounding is compared against its climatology. The 🏆 map features stations whose latest launch broke a station record (all-time or for the time of year), with near-record (P95+) stations as context dots. Each label carries its sounding's launch time — records persist until the station launches again, so labels older than 48 h fade to unlabeled stars rather than pinning stale text to the map. The drill-down table under a station lists what stands and what this sounding broke.

6. The code

The repository is a plain GitHub Pages site — what you're reading and what runs in your browser are the same files. Browse the repository →

The data lives on orphan branches served CORS-open: skewt-data (latest soundings), skewt-archive (day bundles) and skewt-climo (per-station climatology).

7. Limitations

• Not for operational use — this is a personal research tool.
• Climatology covers active stations; long-closed sites have no percentiles.
• ECAPE/SHIP climatologies rest on fewer samples than level diagnostics, because they demand complete quality-gated soundings.
• Historical soundings are coarser because the data is — a 1960 hodograph has fewer levels, not worse rendering.

← Open the explorer