Progress journal | What the beach data can tell us

Where things stand

Predicting City beach widths: the best rule so far is still the plain seasonal average (about 10 ft average error). Adding waves didn't help on the three survey dates we could test.
Satellite shorelines: after fixing the method, they matched a 2016 airborne LiDAR survey near the pier to within about 1 m on average (3–4 m typical error per line). All the images were from late April, so this covers one area in one season. A dense 2017–2026 satellite record for the whole cell has now been built.
Wave-driven shoreline model: it reproduced a published model at Torrey Pines and beat a seasonal-average baseline there on held-out years, when driven by the waves that actually happened. Its settings didn't carry over to a second Torrey site, so we don't assume they apply at San Clemente.
The 2024 sand placement: the record now adds up (about 114,000 + 82,000 cubic yards). A test of whether satellites can see it is being written down before anyone looks at the 2024–25 satellite data.

Open questions

Did the 2024 federal placement near the pier move the shoreline enough to show up in satellite data, and for how long? (A test is being written; no result yet.)
Can a model fitted to local satellite data beat simple seasonal and "same as last time" predictions on a year it hasn't seen (2025)?
Do forecasts made in advance, with predictions locked before the images exist, beat simple baselines? The first lock-in is 12 Oct 2026.
Could sediment from the 2024 Airport Fire burn area reach the beaches near Doheny, and could we even detect it?

Journal

10 Oct 2026

Placement record reconciled

Later City reports show the federal project placed about 114,000 cy (April–May 2024) and then 82,000 cy (November–December 2024). Together that is 196,000 cy, matching the county's total for the central beaches. No as-built map by location is public. (c29a4ac)

Fire-to-beach question scoped

Overlaying the Airport Fire perimeter on public watershed maps: about 87% of the burned area drains to the San Juan Creek mouth at Doheny. A desk calculation of what the satellites could detect suggests that a beach-width signal would be visible only after a very large sediment flush (5 of 108 scenarios). The test will be designed so that "too small to see" is reported as exactly that, not as proof of no effect. (9949bad, 39f3c22)

In progress not yet reviewed

We drafted the written rules for the 2024 placement test, fixed using only 2017–2023 data. We also prepared a full evidence packet for the Torrey Pines model test, so others can check it.

9 Oct 2026

Satellite shorelines pass a check against LiDAR

We compared satellite shorelines with the May 2016 airborne LiDAR at 41 lines near the pier. The rule was set in advance, and the result passed: bias about −1 m, typical error 3–4 m. All accepted images were from late April, so this is one area in one season, not proof that it works everywhere. The software has a random step, so we reran it with 10 fixed random seeds; it passed every time. That measures the software's randomness, not ten independent checks. (dfa8c54, 78466a3)

Torrey Pines model test

We reproduced the published Yates et al. (2009) model at Torrey Pines within pre-set tolerances. Then we tested it on six years it had never seen (2011–2016). Driven by the observed waves, its typical error was 7.6 m, against 12.5 m for a seasonal average, a pass. A simpler model did slightly better. The settings did not transfer to a second section, and with average (not actual) waves the model did not beat the seasonal average. So it shows no forecasting value beyond the reach of a wave forecast. (89aaa2d, eedf13d)

Dense satellite record

We downloaded and processed Sentinel-2 imagery from 2017 to 2026 for 103 shore-normal lines across the cell, with cloud and processing-version rules fixed beforehand. (e5f1952, e6146ec)

9 Oct 2026 dropped

We planned a fresh test of the satellite record against 2017–18 City surveys, then dropped it by its own pre-set rule: only one line was geometrically stable enough to score. (ca68e2e)

8 Oct 2026 failed

Prediction test on City widths: on 36 later widths, the seasonal-average rule scored 10.0 ft average error, and the wave-adjusted rule 12.0 ft. No improvement from waves. A separate check also failed: using the 2016 LiDAR to reproduce City widths. Approximate transect geometry dominates the error. We asked the City for surveyed transect coordinates and profiles (8 Oct). (Phase 2 summary; 49d18a0)

7–8 Oct 2026

The project started. We inventoried public data (City surveys, CDIP wave model points, NOAA tides, satellite imagery, placement history) and published the main report. (16b3c05)

Limits

Nothing here shows predictive skill at San Clemente yet. The Torrey Pines results come from a different beach.
The satellite–sensor offset (Sentinel-2 vs Landsat) isn't pinned down. Two Sentinel-2 dates and three Landsat dates aren't enough.
The 2024 placement analysis isn't fully blind. Some 2022–25 satellite and City numbers near the pier were seen before its rules were written, and we say so in the test.
No as-built placement map, post-construction survey or fill grain size is public.

12 Oct: first locked-in forecast of satellite shorelines for the coming week. It gets scored after the images arrive.
Run the 2024 placement test once, after its rules are reviewed and committed.
Write down the rules for fitting a San Clemente model and testing it on 2025.
Fire-sediment test rules due by 23 Oct, or earlier if a large storm flow occurs.

What we're testing, and what has held up