KELP v1 Knowledge Enhanced Local Prediction
UK + IRELAND / ENGLAND / SOUTH WEST
Saunton Surf Report & Forecast
Connecting to the shared forecast…
Standalone preview · background collection begins when installed on the Pi.
Loading forecast…
Current forecast
—ftExperimental face-height estimate —Swell peak / mean period
——
Wave face heights are first-pass model estimates, not observations. The displayed surf bands are rounded to 1–2 ft wide for readability; they are not confidence intervals.
Beach & directions North up
N ↑Imagery © Esri, Vantor, Earthstar Geographics, GIS User Community ━ Primary swell━ WindArrows show travel direction
Seven days ahead
Powered by KELP v1SURF · FTPERIODWINDDIRECTION
TIMEWAVE FACEPEAK PERIODQUALITYWIND · GUSTDIRECTION
All seven days are shown below. Select a row to update the beach map and tide panel. The main swell height, mean period and direction are Météo-France values and drive the wave-height model. DWD swell peak, height and direction are shown together as a separate comparison. ECMWF total-sea peak remains separate. Neither comparison changes the main forecast. Missing comparison values stay unavailable. Buoy checks compare total-sea Tp with total-sea Tp, not with swell mean periods.
Tell us what you think
Ideas, problems or suggestions? Leave a message. Your email is optional and will only be used to reply about your feedback. Neither is published.
CHECK WHAT IT ACTUALLY LOOKS LIKE
Beach camera
MEASURED LOCAL GEOMETRY
Beach & breaking-wave trial
Survey model · experimentalTrial water levels are scenarios, not verified tides. Entering an offset links the trial to the selected forecast hour, but does not verify that offset. Blue: water level · brown: surveyed bed · orange: estimated breaking point. Gaps remain unknown. Existing forecast heights remain the baseline while we validate the survey model.
Survey sources and model limits
Source: South West Coastal Monitoring / Channel Coastal Observatory. Polzeath: AG Surveys, 6 March 2026, OSGB36 / ODN, five measured profiles. Saunton main beach: HI1853 Bideford Bay multibeam exports, 22 September 2024, 2 m grids labelled ODN. Transects at OSGB northings 137100, 137300 and 137500 m. The survey source pack includes Crow Point’s February 2026 profiles, but those profiles are geographically separate and excluded from the main Saunton model. Raw source files and original metadata are retained in the survey source pack.
The trial transports each wave component’s shore-normal energy flux from the assumed input depth and finds the first depth-limited breaking point within the outer continuous measured section (breaking index 0.65). Local measured section direction replaces the baseline’s approximate direction only in this trial. Saunton grid NoData gaps remain disconnected; Polzeath profile gaps wider than 20 m are disconnected. Values between neighbouring measured samples are linearly interpolated. If breaking starts outside the measured section, a height is withheld. This one-dimensional model does not resolve headlands, currents, directional refraction, wave grouping or today’s sandbanks. Datum conversion provenance and the forecast tide’s ODN alignment remain unresolved. No accuracy improvement is established. Not to be used for navigational purposes.
Tide information
Provisional coastal model, not an official local tide table. Heights are relative to mean sea level (MSL), not chart datum. Event times below are approximate hourly extrema.
For authoritative local high/low water predictions use ADMIRALTY EasyTide. Direct UKHO integration requires an API subscription key and verified station selection.
How the wave-height model works Baseline v0.1 + survey trial v0.1
The primary wave feed is Météo-France. DWD EWAM is a separate comparison and is not used in the main wave-height calculation. The wave-height calculations continue to use each component’s mean period, with source provenance stored when you log an observation. Mean and peak periods are clearly labelled. The comparison models have not been established as more accurate locally. Each available swell component is adjusted for its approach angle, propagated shoreward with a simplified energy-flux and depth-limited breaking model, then combined by wave energy. This gives an estimate of significant breaking-wave face height, rather than the biggest set.
Starting assumptions: straight parallel seabed contours, input depth 20 m, breaking index 0.65, approximate seaward beach normals 270° at Saunton and 280° at Polzeath, correction 1.00. These remain the baseline model’s provisional assumptions. The separate survey trial above incorporates measured cross-sections and explores water-level effects; it has not been calibrated. Headland diffraction, currents and wave grouping remain unresolved. Missing wind-wave components are excluded. Displayed hourly bands are 1 ft wide below 6 ft and 2 ft wide above that, rounded around the estimate. Daily headline values refer to midday; bars show changes through the day. Narrow display bands do not mean that forecast uncertainty is this small. Wind affects the surface-condition indicator separately, not wave height.
Wave transformation reference. Quality stars are a provisional rule-based surf-quality score: 1 poor, 2 poor–fair, 3 fair, 4 good, 5 excellent; they are not forecast confidence. Body-height labels are approximate, based on a roughly 6 ft surfer. The latest 4–5 ft / poor reports from Matt are stored as observations, not substituted into the forecast or used to fit the whole model. The simplified model has not been validated for these beaches. Changing controls explores sensitivity; it does not establish accuracy.
Independent buoy checks Measurements are the reference
Bideford Bay checks the North Devon wave model; Perranporth is a regional north Cornwall check, not a Polzeath surf-height measurement. Models are sampled at each buoy’s coordinates. Live observations are not quality controlled.
If browser cross-origin rules block live buoy access, save the official table page as HTML and import it above, or import CSV: time,hs,tp,tz (ISO timestamps with timezone; metres and seconds). JSON backup imports are also supported. Historical checks use only snapshots fetched before the observation; retrospectively fetched values never count as forecast accuracy.
What was it actually like?
Help KELP learn. Accurate eyeball reports help KELP learn about each beach and improve future predictions for everyone. Report what you actually see—even when the forecast is wrong.