Methodology

Status: experimental, personal research forecast by @derecho_series (#onwx). It is not an official forecast or warning. Always follow Environment and Climate Change Canada (ECCC) alerts.

Pipeline

Every run fetches the latest hourly Open-Meteo forecast (2 m T/Td, 10 m wind & gusts, MSLP, precipitation, CAPE, lifted index, precipitable water, freezing level, and temperature / height / wind / relative humidity at 925, 850, 700 and 500 hPa) on a ~0.5° × 0.6° point mesh over 41–47.5°N, 85.5–72°W (wide enough to include the upwind lakes), plus Great Lakes surface-water temperatures at open-water sample points.
Deep-layer (sfc–500 hPa) and low-level (sfc–850 hPa) bulk shear are computed; storm-relative helicity is a shear-based proxy.
Products are computed every 3 h from now to +48 h, interpolated to 0.05° and smoothed for contouring.
At +6 h and +12 h the Control Thunderstorm Volatility Index is blended 50/50 (where its grid overlaps) with the ML spatial forecaster — a ConvLSTM + U-Net residual (Δ-state, GraphCast-inspired) model trained on ERA5 1.5° Ontario grids — initialised from the last four 6-hourly Open-Meteo frames.

Severe weather

Severe Thunderstorm Parameter (1–10)

The original machine-learning model (formerly “Severe Weather Index”), restored Nov 1 2026 with the exact pipeline that made the Sept 2 2026 render. For each map time the Ontario spatial forecaster (a ConvLSTM + U-Net residual model, GraphCast-inspired, trained on ERA5 Ontario grids with constraints drawn from @derecho_series posts) takes the four 6-hourly Open-Meteo frames ending 6 hours earlier and predicts the fields 6 hours ahead. The original index is then computed on those predicted fields: CAPE, 0–6 km shear and storm-relative helicity, scored against the rest of the map plus absolute terms (CAPE/2000, shear/25 m/s, helicity/200), with a significant-severe bonus. Original colour scale, levels every 0.5.
Environment ceiling (Nov 1 2026): because part of the original score is relative to the rest of the map, quiet days used to show high values in the relatively best area. The map now keeps the original pattern but caps it with an absolute ceiling built from the same ingredients as the Control damper:
Parameter = min(original value, C), C = 10 × f(CAPE) × f(shear) × f(profile). f(CAPE) runs 0→1 from 100 to 1200 J/kg; f(shear) runs 0.55→1 for 0–6 km shear of 8→18 m/s; f(profile) = f(2 m dew point 8→13 °C) × (0.6→1 for 700–500 hPa lapse rate 5→6.5 °C/km), fading to 1 as CAPE rises from 1000 to 2500 J/kg. The ceiling has no floor, so on a stable day it is near 0 and values under 1 are not drawn: a blank map.
Validation (Open-Meteo historical-forecast archive, 14 cases): event-day maxima 6.1–9.3 (6 of 6 at 6 or higher; Sept 2 2026 stays at 8.4). Mundane-day maxima 0.1–8.1 (2 of 8 at 6 or higher, down from 8 of 8 without the ceiling): Aug 5 2024 (6.7) and May 15 2025 (8.1, at the south-west edge of the map) still read high because CAPE and moisture were genuinely high.

Control Thunderstorm Volatility Index (1–10)

The previous live version of the index (map-relative scores with floors and a CAPE gate, plus the ML blend), unchanged, with a damper for marginal environments:
Index = 1 + (Parameter − 1) × D. D is a product of smooth tapers, each running from 0.4 at the marginal end to 1 at the robust end: CAPE 300→1000 J/kg, 2 m temperature 18→22 °C, 2 m dew point 12→16 °C, 700–500 hPa lapse rate 5.5→6.5 °C/km. Because CAPE already reflects the temperature profile, the temperature, dew point and lapse-rate tapers fade out as CAPE rises from 1000 to 2500 J/kg. Damping only applies when the profile and CAPE are both marginal. When the ingredients are robust, D = 1 and nothing changes.
Validation (Open-Meteo historical-forecast archive, 14 cases, without the ML blend): event-day maxima: Parameter 9.1–10.0, Control 8.5–10.0 (6 of 6 at 6 or higher). Mundane-day maxima: Parameter 7.9–10.0 (8 of 8 at 6 or higher), Control 2.9–10.0 (mean 6.2, 3 of 8 at 6 or higher). The damper removes most false alarms on cool, low-CAPE days. It cannot separate hot, high-CAPE days without storms (Jul 8 2025, May 15 2025 stay near 10). The earlier multi-ingredient rebuild has been retired.

Lightning Probability (%)

Physically motivated, not calibrated against lightning observations (no lightning flash-density truth is available in the ERA5 stores this pipeline can reach): buoyancy σ((CAPE−250)/200)·σ((0.5−LI)/1.2), charge-separation-zone instability from the 700–500 hPa lapse (≈ −10 to −20 °C layer), 700 hPa moisture and model precipitation as an updraft proxy. Treat the percentages as relative, not reliable frequencies.

Winter

Snowsquall Parameter (0, rain/mix 1–3, snow 1–10)

Output scale: blank (0) = no lake-effect precipitation possible; green 1–3 = lake-effect rain or mixed precipitation (hard-capped at 3): the lake-effect ingredients are met but the profile is too warm for snow (850 hPa above −5 °C, a warm layer, or 2 m above +2 °C); snow colours 1–10 only where the temperature profile supports snow. Map labels show each lake’s surface temperature (Open-Meteo surface temperature over open water at run start) and observed 2 m temperatures (METAR reports from aviationweather.gov at run time). Validation (ERA5): quiet days 8/8 blank for snow; 4 autumn cold-outbreak cases (assumed lake-effect rain, not radar-verified) gave rain/mix on 2/4 (max 3.0 on 2017-10-25), none on 2/4; snow events showed no rain/mix.
A physically constrained lake-effect model (replaces the earlier heuristic). For each lake (Superior, Michigan, Huron, Georgian Bay, Erie, Ontario) the column over the water is scored with that lake’s surface-water temperature:

Lake–850 hPa ΔT: σ((T_lake − T₈₅₀ − 13)/1.5) — the classic ≥13 °C threshold.
Convective depth: mixing depth / inversion height of a lake-heated parcel through the 925/850/700/500 hPa profile, plus the 850–700 hPa lapse rate (lake–700 ΔT is reported as a diagnostic).
Snow in the cloud layer: σ((−T₈₅₀ − 4)/1.5). Moisture: relative humidity at 850 and 700 hPa.
Wind: 850 hPa speed window (organised bands ~6–25 m/s) and low directional shear 850–700 hPa (σ((30° − Δdir)/7)).
Upper-air pattern (modulates, does not gate): 500 hPa trough curvature (Laplacian of height), 850 hPa cyclonic vorticity, 850 hPa cold advection, and wrap-around flow (lower pressure ~400 km to the east-northeast).
Open water: lake temperature above freezing and no ice cover.

Ingredients are multiplied (each is necessary). Bands are then placed by tracing upwind from every grid point along the 925–850 hPa steering wind: fetch over each lake, the mean column score over that water, and distance inland from the lee shore (~90 km e-folding) set the value; a 2 m temperature guard removes rain cases.
Hard gates (the value is 0, blank, when any fails, with a smooth taper just above each threshold): lake–850 ΔT ≥ 10 °C; 850 hPa T ≤ −5 °C; no layer above 0 °C at 850/700 hPa; 2 m T ≤ +2 °C; open, ice-free water; 850 hPa RH ≥ 55%; 850–700 hPa direction change ≤ 60°; 850 hPa wind 3–30 m/s; over-water fetch ≥ 50 km; no capping inversion below ~1000 m.
ERA5 validation (11 documented lake-effect times, 8 warm/summer times and 6 cold-season non-squall times, 2007–2022): event peaks averaged 4.25 (median 4.48, lowest 1.22); warm/summer peaks averaged 1.04 (highest 1.25); cold non-squall peaks averaged 1.15 (highest 1.88). Values ≥ 4 occurred in 6/11 event times, 0/8 warm/summer and 0/6 cold times; ≥ 3 in 9/11 events and 0/6 cold times. The cold non-squall days are assumed (not radar-checked). A small, partly tuned sample — not a skill score.

Snow totals (cm)

Hourly precipitation × a Kuchera-style temperature-dependent snow-to-liquid ratio, counted only in hours the temperature profile types as snow, accumulated since run start.

Precipitation Type and Intensity

Each grid point gets a type and a 1–10 intensity, shown as light-to-heavy shades of that type's colour: rain (green), snow (blue), freezing rain or ice pellets (pink) and mixed rain and snow (purple). Type: snow when 2 m T ≤ 1 °C and the warmest of 925/850 hPa ≤ 0.5 °C; freezing rain or ice pellets when 2 m T ≤ 0 °C with a warm layer above 0.5 °C; mixed when neither applies but 2 m T ≤ 2 °C and the warm layer ≤ 1.5 °C; otherwise rain. Blank below 0.1 mm/h. Intensity is the hourly precipitation rate (liquid equivalent) against these lower thresholds for levels 1 to 10:
Rain: 0.1, 0.25, 0.5, 1, 2, 3, 4.5, 6, 8, 10+ mm/h
Snow: 0.1, 0.2, 0.3, 0.5, 0.75, 1, 1.5, 2, 2.5, 3+ mm/h (about 1 to 3+ cm of snow per hour at 10:1)
Freezing rain / ice pellets: 0.1, 0.2, 0.3, 0.5, 0.75, 1, 1.5, 2, 3, 4+ mm/h
Mixed rain and snow: 0.1, 0.2, 0.4, 0.6, 1, 1.5, 2, 3, 4, 5+ mm/h
Snow and ice are scaled lower than rain because the same liquid rate is far more disruptive when frozen. Click the map for the type, intensity and rate.

Freezing Rain Ice Accretion (mm)

Precipitation typed as freezing rain (2 m T ≤ 0 °C with a warm nose > 1.5 °C at 925 or 850 hPa) × 0.6 accretion efficiency, accumulated. Ice pellets: warm nose 0.5–1.5 °C.

Precipitation

Precipitation (1 h, 6 h, 24 h, total) and Rain total (mm)

Model liquid-equivalent precipitation over the trailing window or since run start. Rain total counts only hours typed as rain from the temperature profile.

Heavy Rain / Flash Flood Potential (1–10)

1 + 9 × max(σ((1 h−15)/5), σ((6 h−35)/10)) × (0.5 + 0.3·σ((PWAT−30)/5) + 0.2·σ((q₈₅₀·V₈₅₀ − 0.12)/0.04))

— rainfall rate, precipitable water and 850 hPa moisture flux. No soil-moisture or urban-drainage information.

Wind

10 m Winds and Max Wind Gust (km/h)

Open-Meteo 10 m sustained wind speed (arrows show direction) and the highest model gust in the 3 h before the valid time. Gusts are raised by up to 10% within ~25 km of open water when the lake is more than ~5 °C warmer than the air (deeper mixing in cold air over warm water). No downslope term — southern Ontario has little terrain for it.

Regions: Southern Ontario and Newfoundland

Region selector

The map has two regions: Southern Ontario (default) and Newfoundland (the island; Labrador is not included). Old links without a region still open Southern Ontario; Newfoundland links add r=nl. Both regions are built in the same run, with the same 14 layers. Newfoundland maps use a satellite basemap, Newfoundland town labels and Newfoundland time (Newfoundland Standard Time / Newfoundland Daylight Time). Colour is drawn only over the island and about 25 km of coastal water, so there is no rectangular edge.

How each layer is made for Newfoundland

Severe Thunderstorm Parameter: the same original machine-learning pipeline, with its 6 × 11 model grid moved (same size and spacing) to 46.5–51.5° N, 60.4–51.4° W, then capped by the same environment ceiling. Caveat: the model was trained only on southern Ontario warm-season (May–September) data, so in Newfoundland it is used outside the area it learned. Its relative part can still pick out the "least stable" area on a quiet day; the ceiling is what keeps such days near zero. Treat it as experimental and lean on the ceiling and the Control index.
Control Thunderstorm Volatility Index: same formula as Ontario (no machine-learning blend). The convective available potential energy, shear and helicity thresholds were tuned in Ontario; Newfoundland's cool marine air means values will usually be low.
Snowsquall Parameter: the ocean version: sea-surface temperatures from the Open-Meteo marine service for four basins (Gulf of St. Lawrence, north-east coast and Labrador Sea, east coast, south coast) replace lake temperatures; the same air-sea temperature difference, column-depth and wind gates apply. Wind arrows are shown. Sea-surface labels show each basin's temperature at the run start. Caveat: no sea ice handling, and land terrain lift (Long Range Mountains) is not modelled.
Snow totals, precipitation (1 h, 6 h, 24 h, total), rain total, freezing rain ice: identical method to Ontario.
10 m winds and gusts: model winds and gusts with no lake enhancement (the Ontario lake-enhancement step is not applied). Wreckhouse-type downslope gusts in the south-west are not resolved.
Flash flood potential and lightning probability: same formulas as Ontario; neither is calibrated for Newfoundland.

Combination maps (both regions)

2 m Temperature with 10 m Wind and 2 m Dew Point with 10 m Wind: filled temperature or dew point with 10 m wind arrows.
850 hPa Temperature with 850 hPa Wind: about 1.5 km up; the 0 °C line is a quick rain/snow guide, the arrows show warm or cold air being carried in.
Sea-Level Pressure Isobars with 1 h Precipitation: white isobars every 4 hPa over the hourly precipitation fill.
Feels-Like Temperature: the Open-Meteo apparent temperature (wind chill when cold, humidex-like when warm).
Visibility / Fog: model visibility under 8 km only; below 1 km is fog. Model visibility is a rough guide and often misses marine fog timing.
Freezing Level Height: height of the 0 °C level above sea level.
Precipitation Type: rain, snow, freezing rain or ice pellets where at least 0.1 mm/h falls, typed from the 2 m, 925 hPa and 850 hPa temperatures (same rule as the rain and ice layers).

Synoptic maps (both regions)

These three layers use a wider view than the other maps (Southern Ontario: 30–60°N, 106–56°W; Newfoundland: 34–64°N, 84–30°W) so whole weather systems fit on screen. Model data from Open-Meteo on a 1.25° mesh is smoothed onto a 0.5° grid.

500 hPa anomaly (regional view) — the wide synoptic view around each region. Black lines are 500 hPa heights every 6 decametres, labelled in white boxes; the 540 decametre line (a common rough rain/snow guide) is drawn thick and blue. Shading is the height anomaly in metres (−500 to +500 m) against the 1991–2020 average for the same time of year: greys near zero, yellow → orange → red → brown for ridging (higher heights than normal), blue → green → purple → pink for troughing. The base period is 1991–2020, from the National Centers for Environmental Prediction / National Center for Atmospheric Research Reanalysis 1 monthly long-term means (National Oceanic and Atmospheric Administration Physical Sciences Laboratory), interpolated to the day. The share image shows the lowest and highest anomaly (MIN|MAX) in red.
500 mb anomaly – North America — the same style over the whole continent (about 15–75°N, 170°W–10°W), identical for both regions with the selected region outlined in red. Heights come from the European Centre for Medium-Range Weather Forecasts model (0.25° grid, via Open-Meteo), sampled every 2.5° and smoothed; the anomaly uses the same 1991–2020 base period. The share image is on a Lambert conformal map with state and province borders.
Town and city names — every map shows place names. On the site more towns appear as you zoom in (major cities first, then regional centres, then smaller towns) and labels that would overlap are skipped; the wide synoptic and North America views show major cities only.
Highs / lows — sea-level pressure isobars every 4 hPa. High (H) and low (L) centres are found automatically as local maxima and minima of the smoothed pressure that stand at least 2 hPa above or below their surroundings, and are labelled with the central pressure.
Fronts — simulated surface fronts drawn on the pressure map. The method uses the thermal front parameter (Renard and Clarke, 1965; Hewson, 1998) on 850 hPa equivalent potential temperature: a front is drawn along the warm edge of a strong temperature-and-moisture contrast (at least about 3 K per 100 km), only where it is at least 400 km long and mostly lies in a sea-level pressure trough. The 850 hPa wind across the line decides the type: wind pushing toward the warm side marks a cold front (blue triangles), toward the cold side a warm front (red half-circles), little movement a stationary front (alternating blue and red), and a cold-type line close to a deep low (below 1000 hPa) is shown as occluded (purple). Over high terrain (above 1200 m), where 850 hPa is underground, no fronts are drawn.
Limits: this is an automatic analysis, not a forecaster's hand analysis. It can miss weak or shallow fronts, break one front into pieces, and mislabel occlusions. Treat it as a guide to where the boundaries are.

Water levels

Newfoundland significant wave height

Open-Meteo Marine significant wave height (the average of the highest third of waves) with mean wave direction arrows; click for the wave period. Shown over water within about 230 km of the island.

Storm surge model

A point model at coastal ports (Newfoundland: St. John's, Argentia, Port aux Basques, Harbour Breton, Bonavista, Lewisporte, Botwood, Corner Brook, Lark Harbour; Southern Ontario: Lake Erie, Lake Ontario, Lake Huron and Georgian Bay ports). Surge is the sum of:

Wind setup: steady balance of wind stress against the water slope, setup = shelter × (air density / water density) × drag × wind² × onshore fraction × fetch / (gravity × depth). Mean 10 m wind over the previous 3 hours (not gusts); drag coefficient (0.75 + 0.067 × wind) × 10⁻³. Only the onshore wind component counts at ocean ports; each port has its own fetch (capped at 250 km), effective depth (charted coastal depths, not a full bathymetry grid) and a shelter factor for enclosed harbours and bays. Setup is limited to a quarter of the depth.
Lakes (Southern Ontario), primary layer "Storm surge / seiche" (physics model): a closed lake tilts, so setup is signed along the long axis: water piles up at the downwind end and drops at the upwind end (for example Buffalo up and Toledo down in a south-west gale on Lake Erie). Wind stress uses the over-lake mean wind of the previous 3 hours with a speed-dependent drag. Lake Erie uses an effective depth of 11 m (its shallow western basin), Lake Ontario 45 m, Lake Huron 35 m, Georgian Bay 30 m. No tides.
Lakes (Southern Ontario), control layer "Seiche control": a simple seiche (set-up) formula. The control layer uses a hand calculation: set-up = (3.2 × 10⁻⁶ ÷ (gravity × mean depth)) × wind² × lake length, where the set-up is the measurable seiche, gravity is 9.81 m/s², and wind is the mean of the sustained wind and the gusts over several hours. The constant 3.2 × 10⁻⁶ is the air-to-water density ratio times a storm drag coefficient, so it is the same wind-stress balance as the physics above. Here the wind is the over-lake mean of sustained wind and gusts over the previous 5 hours, only its component along the lake's long axis counts, and the lake tilts: water rises at the downwind end and drops at the upwind end (Buffalo up and Toledo down in a south-west gale on Lake Erie), with 0.6 of the set-up at each end, scaled by each port's position along the axis. Lengths and mean depths: Lake Erie 338 km and 19 m, Lake Ontario 311 km and 86 m, Lake Huron 330 km and 59 m, Georgian Bay 190 km and 44 m. No tides.
Seiche height map (Southern Ontario lakes): the same two seiche calculations spread over the whole water surface of Lake Erie, Lake Ontario, Lake Huron, Georgian Bay and Lake St. Clair (Lake Michigan and Lake Superior are outside this map). Each lake tilts along its long axis with the wind: the water rises at the downwind end (red), drops at the upwind end (blue), and passes through a near-zero line in the middle (dashed). The pressure effect (1 cm of rise per hectopascal below 1013) is added everywhere. Colours run from −2.5 to +2.5 m in 0.2 m bands, near-white around zero, and arrows show the wind over the water. Two layers: "Seiche map" (physics model) and "Seiche control map". The ends of each lake are scaled by an end gain (1.88 for the physics map, 1.47 for the control map), each end reaching full strength at its own tip, so that the map matches the gauges at the far ends where the shallow, narrowing basins pile the water higher than a straight-line tilt. The gain was fitted to 23 December 2022 (gauges: Buffalo about +3.2 m, Toledo about −2.9 m); the map gives Buffalo +3.2 m and Toledo −2.9 m (physics), +3.1 m and −2.9 m (control). Checked on the held-out 31 October 2019 storm (gauges: Buffalo +1.9 m, Toledo −1.6 m): physics +2.0 m and −1.8 m, control +2.2 m and −2.1 m. Values beyond ±2.5 m show in the darkest band. The point forecasts on the "Storm surge / seiche" and "Seiche control" layers are not changed by this scaling.
Inverse barometer ("lift up"): +1 cm for every hPa below 1013 hPa (and −1 cm per hPa above).
Wave setup (Newfoundland): 0.19 × significant wave height × onshore wave fraction × a port exposure factor.
Tide (Newfoundland): Fisheries and Oceans Canada tide predictions for each port, shown relative to the mean level over the run window. Total water level = tide + surge.

Not included: resonance and sloshing after the wind stops (lake seiches oscillate for hours), coastally trapped waves, river flow, ice cover, the long-term lake level, and runup on individual beaches. Real surge lags the wind by a few hours on Lake Erie.

Validation (hindcasts, no tuning)

Weather from the Open-Meteo reanalysis archive; waves from the Open-Meteo Marine archive (not available for 2021, so Hurricane Larry has no wave setup). Observed surge: Fisheries and Oceans Canada observed minus predicted levels, and the United States National Oceanic and Atmospheric Administration hourly levels for Lake Erie (minus the 5-day pre-storm mean). Peak values are the largest absolute value in the window.

Case	Observed peak (m)	Model peak (m)	Peak error (m)	Timing (h)	Root-mean-square error (m)	Correlation
Hurricane Larry, St. John's	-0.55	+0.37	+0.93	-4	0.20	0.16
Hurricane Larry, Argentia	+1.40	+0.40	-1.00	+1	0.32	0.20
Post-tropical Fiona, Port aux Basques	+0.88	+1.99	+1.12	+3	0.65	0.73
Post-tropical Fiona, Argentia	+0.58	+0.94	+0.36	+8	0.43	0.33
Hurricane Larry, Bonavista	+0.46	+0.38	-0.07	+1	0.13	0.53
Lake Erie seiche Dec 2022, Buffalo	+2.37	+2.51	+0.14	+5	0.37	0.97
Lake Erie seiche Dec 2022, Toledo	-2.88	-2.22	+0.66	+2	0.35	0.93
Lake Erie Halloween storm Oct 31 2019, Buffalo	+1.90	+1.56	-0.34	-3	0.18	0.93
Lake Erie Halloween storm Oct 31 2019, Toledo	-1.60	-1.31	+0.29	-4	0.26	0.87

Lake Erie is the strongest result (correlation 0.87–0.97, peaks within 0.15–0.66 m). Newfoundland is much weaker: wave setup is overdone at harbour tide gauges during Fiona (gauges sit in sheltered water), and the reanalysis winds are too weak and too smooth for a small fast hurricane like Larry, so Larry's surge is missed. Treat the Newfoundland surge layer as an indication of risk, not a forecast of water height.

Southern Ontario seiche: the physics model and the control formula

Lake Erie hindcasts, same weather and observations as above. Both are shown as separate layers, like the Severe Thunderstorm Parameter and its Control Thunderstorm Volatility Index: Storm surge / seiche is the physics model (wind stress with a speed-dependent drag over the lake length and depth, 3-hour mean lake wind, plus the inverse barometer) and stays primary because it fits the timing best (lower root-mean-square error in nine of ten cases); Seiche control uses the control formula (5-hour mean of sustained wind and gusts, 0.6 of the set-up at each end, plus the inverse barometer) and matches the peak heights more closely in most cases. Where the two agree, confidence is higher. Applying the full set-up at one end, as in a single hand calculation for one shore, roughly doubles the peaks against the gauges, so the site splits the tilt between the two ends.

Case	Observed peak (m)	Physics model peak (m)	Physics root-mean-square error (m)	Physics correlation	Control peak (m)	Control root-mean-square error (m)	Control correlation
Lake Erie Dec 2022, Buffalo	+2.37	+2.51	0.37	0.97	+2.67	0.59	0.95
Lake Erie Dec 2022, Toledo	-2.88	-2.22	0.35	0.93	-2.39	0.48	0.90
Lake Erie Oct 2019, Buffalo	+1.90	+1.56	0.18	0.93	+1.88	0.29	0.92
Lake Erie Oct 2019, Toledo	-1.60	-1.31	0.26	0.87	-1.75	0.39	0.85
Lake Erie Dec 2021, Buffalo	+1.97	+1.99	0.43	0.78	+2.09	0.52	0.80
Lake Erie Dec 2021, Toledo	-1.37	-1.73	0.40	0.73	-1.88	0.49	0.75
Lake Erie Mar 2021, Buffalo	+1.53	+0.83	0.22	0.70	+1.11	0.22	0.76
Lake Erie Mar 2021, Toledo	-0.62	-0.68	0.20	0.60	-0.99	0.27	0.61
Lake Erie Nov 2020, Buffalo	+2.17	+1.97	0.33	0.88	+2.18	0.50	0.88
Lake Erie Nov 2020, Toledo	-2.03	-1.71	0.26	0.93	-2.02	0.40	0.90

Shoreline inundation and erosion coefficients

Two 1–10 coefficients drawn as colour-coded shoreline segments of about 1.5 km (layers Shoreline inundation and Shoreline erosion, under Water levels). The water level on each segment comes from the control formula Sh = [(3.2e-6)/(g·zm)]·[W²·l]: in Southern Ontario from the Seiche control layer (5-hour mean lake wind, 0.6 of the set-up, placed along the lake axis); in Newfoundland from the new Surge control layer, which applies the formula with the onshore wind over the coastal fetch and the depth at the nearest port, capped at a quarter of the depth, plus the inverse barometer, the predicted tide at the nearest Fisheries and Oceans Canada port and a wave setup of about 0.1 of the wave height. Waves are fetch-limited on the lakes and from the Open-Meteo Marine model at sea, reduced by the angle of approach.

Inundation compares the total water level (water level plus wave runup) with the height of the land behind the shore, read from the Copernicus 30 m elevation model along a 600 m inland profile, or with the crest of a structure. Erosion builds up over 12 hours from the time the water plus half the runup stays above the toe of the shore, weighted by wave energy and by how easily the shore erodes.

Shore type	Erodibility
Bluff (glacial till)	1.0
Sand, low beach	0.7
Rocky shore with sediment	0.4
Limestone bedrock	0.3
Bedrock, bedrock cliff, harbour	0.2
Pier, wharf, groyne	0.15
Breakwater, seawall	0.1

Protected structures (harbours, piers, wharves, breakwaters, seawalls) come from OpenStreetMap. In Newfoundland they are included with low erodibility and an overtopping test against the crest height: the tagged height where present, otherwise the elevation at the structure, otherwise an estimated 3.0 m (marked "estimated" in the popup). In Southern Ontario only natural shores are shown. Each popup gives the segment type, both coefficients, water level, tide, runup, total level, crest or backshore height and erodibility.

Hindcasts: Lake Erie east end, Dec 23 2022 — 10 at the peak on both coefficients, 34 of 79 segments at 7 or more for inundation, water level 1.45 m on the south shore toward Buffalo at 11:00 AM; the Buffalo gauge rose about 2.4 m. Post-tropical Fiona, Sep 24 2022 — southwest coast 10 on both, 33 of 93 segments at 7 or more for inundation and 76 for erosion; the Port aux Basques wharf (crest estimated at 3.0 m) is overtopped.

Limitations: the elevation model measures surfaces (buildings and trees raise it); the Natural Earth shoreline is off by hundreds of metres in places, so backshore heights have floors of 1 m on the lakes and 2 m at sea; tide is taken at the nearest port; crest heights are mostly estimated; ice cover, beach width and storm duration beyond 12 hours are ignored. The coefficients rank relative risk and are not calibrated against surveyed damage.

Limitations

~0.5° input mesh; individual squall bands, thunderstorms and lake-breeze features are not resolved — the maps show environments.
Helicity and downdraft CAPE are proxies, not computed from full soundings. Lightning Probability is not calibrated.
The Snowsquall Parameter ignores terrain lift (e.g. Dundalk Highlands, Tug Hill) and lake-to-lake connections; Superior and Michigan lie mostly outside the live grid.
The ML model was trained on a small May–Sep sample at 1.5° and is no longer blended into the thunderstorm indices.
Validation is on small hand-picked ERA5 samples, not operational verification.

Data: Open-Meteo (CC BY 4.0); ERA5 (Copernicus/ECMWF) via WeatherBench2 and ARCO-ERA5 for validation. Basemap © OpenStreetMap contributors © Esri. Map library: Leaflet.