The seawall is now being asked to see the wave coming
the seawall is now being asked to see the wave coming: a peak aware graph transformer emulates storm surges from atmospheric fields for any station, machine learning trims peak timing errors in the pearl river estuary, and a surrogate trained on global tide and surge model runs maps extreme floods for new york under mid century climate.
the coast now files its own warning.
Context
The PACT paper on arXiv, submitted May 9, 2026, presents a peak-aware cross-attention graph transformer for station-level storm surge prediction from atmospheric forcing fields, reporting better RMSE and MAE than a spatio-temporal graph neural network baseline across tide-gauge stations on the US Northeast coast, and about 3.5 seconds to generate a full winter-season surge trajectory for one year after training.
A Frontiers in Marine Science paper on the Pearl River Estuary uses a GRU model for nine stations and 42 tropical cyclones from 2000 to 2023, with a support vector regression correction for peak timing that reduced timing error by 55 percent. An Earth's Future paper trains a deep learning surrogate on Global Tide and Surge Model runs with CMIP6 HighResMIP forcing and tests it for New York City under historical and mid-21st-century scenarios.
The note says PACT emulates storm surges 'for any station'. The paper evaluates multiple tide-gauge stations along the US Northeast coast, uses a station query to produce station-level forecasts, and reports that performance degrades markedly when moving from reanalysis to climate-model forcings, a reanalysis to GCM gap. So any station worldwide is unsupported here, not refuted.
The Pearl River Estuary paper's 55 percent is a reduction in timing error after SVR-based phase correction, and it also reports a 27 percent reduction in peak error at 12 to 18 hours. It trims timing errors as the note says. The model there is a GRU, not a graph transformer.
The New York paper trains on GTSM simulations, and the note's 'maps extreme floods for new york under mid century climate' matches its test. It is a surrogate of a physics model, so its output depends on the hydrodynamic runs it learned from.
'The coast now files its own warning' is the author's line. These are three separate studies in three regions.
Related work
- PACT: Peak-Aware Cross-Attention Graph Transformers for Storm Surge Emulation (arXiv 2605.09036, May 9, 2026) ↗Source for PACT, its peak-aware design, the US Northeast stations and the reanalysis to climate model gap.
- Machine learning based accurate storm surge peak and timing forecast in Pearl River Estuary (Frontiers in Marine Science, May 19, 2026) ↗Source for the Pearl River Estuary peak and timing error reductions.
- A Deep Learning Framework for Extreme Storm Surge Modeling Under Future Climate Scenarios (Earth's Future, 2026) ↗Source for the surrogate of the Global Tide and Surge Model tested at New York City.
Watch next
- Check whether PACT has been tested outside the US Northeast. Read the New York surrogate's errors on the rarest surges.
Sources
- PACT: Peak-Aware Cross-Attention Graph Transformers for Storm Surge Emulation (arXiv 2605.09036, May 9, 2026)arxiv.org
- Machine learning based accurate storm surge peak and timing forecast in Pearl River Estuary (Frontiers in Marine Science, May 19, 2026)frontiersin.org
- A Deep Learning Framework for Extreme Storm Surge Modeling Under Future Climate Scenarios (Earth's Future, 2026)agupubs.onlinelibrary.wiley.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 17:48 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →