<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Geophysical Fluid Dynamics on Vladimir Lapin</title><link>https://vlap.github.io/tags/geophysical-fluid-dynamics/</link><description>Recent content in Geophysical Fluid Dynamics on Vladimir Lapin</description><image><title>Vladimir Lapin</title><url>https://vlap.github.io/images/decor/vladimir_lapin.jpg</url><link>https://vlap.github.io/images/decor/vladimir_lapin.jpg</link></image><generator>Hugo</generator><language>en-us</language><lastBuildDate>Mon, 15 Jun 2026 14:00:00 +0200</lastBuildDate><atom:link href="https://vlap.github.io/tags/geophysical-fluid-dynamics/index.xml" rel="self" type="application/rss+xml"/><item><title>Training Neural Surrogates for Geophysical Fluid Dynamics: 2D Shallow Water Equations on PDEBench</title><link>https://vlap.github.io/posts/2026-06-15-neural-surrogates-shallow-water-equations/</link><pubDate>Mon, 15 Jun 2026 14:00:00 +0200</pubDate><guid>https://vlap.github.io/posts/2026-06-15-neural-surrogates-shallow-water-equations/</guid><description>Single-step validation loss is dangerously misleading when training neural operators on fluid PDEs. Here is what we learned building an end-to-end training pipeline for the 2D Shallow Water Equations on PDEBench, formatting for The Well, and evaluating autoregressive rollouts.</description></item></channel></rss>