Methods & sources

Reading the evidence

Mathematics and development

The eight working papers include detailed mathematical appendices. The development laboratory includes executable NumPy training, solver checks, an archived run record and a downloadable companion. Its learned conductance model is trained on constructed data only.

Figure specifications

All charts were generated from explicit formulas or the existing historical source JSON. CSV files accompany every static chart. Each interactive exports its current numerical state or curve as CSV. Opening figures and visual previews animate automatically when reduced motion is not requested. Every animation has a pause control; off-screen motion is suspended.

Pure mathematical functions · Interactive rendering · Historical T2M data · Series manifest

The 3D lattice loads pinned Three.js from jsDelivr with a native Canvas 2D fallback and a keyboard node selector. Equations use native HTML mathematical typography, with fractions and subscripts, avoiding runtime dependence on an external equation service. They are included in print styles.

Repositories and design references reviewed

These references were checked for this update. A reviewed repository is not necessarily a bundled dependency. JavaScript is used for browser interaction; Java applets are not required.

Scientific sources

1. Skamarock et al. (2019). A Description of the Advanced Research WRF Model Version 4. NCAR Technical Note NCAR/TN-556+STR.
2. Raissi, Perdikaris & Karniadakis (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 378, 686–707.
3. Das, P. et al. (2024). Hybrid physics-AI outperforms numerical weather prediction for extreme precipitation nowcasting. npj Climate and Atmospheric Science, 7, 282.
4. Zhang et al. (2023). Skilful nowcasting of extreme precipitation with NowcastNet. Nature, 619, 526–532.
5. Ravuri et al. (2021). Skilful precipitation nowcasting using deep generative models of radar. Nature, 597, 672–677.
6. NASA GPM. IMERG Version 07 Technical Documentation. Product documentation; consult the selected run and version for latency and support.
7. Funk et al. (2015). The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Scientific Data, 2, 150066.
8. NASA POWER. Hourly point API: historical source response embedded in the DLEGOS workbench.
9. Lam et al. (2023). Learning skillful medium-range global weather forecasting. Science, 382, 1416–1421.
10. Das, A., Kong, Sen & Zhou (2024). A decoder-only foundation model for time-series forecasting. Proceedings of ICML, PMLR 235, 10148–10167.
11. Google. WeatherNext model documentation. Living documentation: record the model generation, issue time, licence and dataset version for each experiment.
12. Roberts & Lean (2008). Scale-Selective Verification of Rainfall Accumulations from High-Resolution Forecasts of Convective Events. Monthly Weather Review, 136, 78–97.
13. Gneiting & Raftery (2007). Strictly Proper Scoring Rules, Prediction, and Estimation. Journal of the American Statistical Association, 102, 359–378.

Implementation and publication record

The existing Climate Physics AI research pages are preserved. This series adds six working papers, six opening interactive figures, additional animated explanations and seven static scientific charts. Article citations can be copied or downloaded as BibTeX. Browser print supports saving a reading copy as PDF.

The author and affiliation follow the research programme supplied by Adrian Dunkley. The prose and implementation are AI-assisted drafts for his review. These pages are not journal publications, do not imply institutional endorsement and do not establish completion or approval of a thesis.

Updated: 16 October 2026. Source code and checks are retained in the PhysicsAIWebsite repository. Visual browser verification was unavailable in this managed environment; publication checks cover content, local links, executable mathematics and JavaScript syntax.

Current literature and the scope of the contribution

The source collection was updated with primary records available by 11 October 2026. The 7 October Beucler et al. preprint distinguishes weather skill from credible response under altered forcing and calls for explicit forcing interventions and out-of-distribution stress tests. It is a recent review, not evidence that this project already meets those requirements. [R211]

PC-AFM uses soft conservation terms; it should not be called an exact normalisation. PISR concerns consistency with primitive equations in its examined data. WIND demonstrates an inverse-problem use of a learned atmospheric prior. These provide close comparators, not transferable Jamaican performance. No first-ever tesseract, conservation layer or physics-and-data inference claim is made. PC-AFM · PISR · WIND.

Animation repositories and verification choices

The implemented browser figures use JavaScript: Three.js for depth, voxels and the projected hypercube; D3.js for the source series and archived forecast scores; and an original isometric archive instrument made with Anatomy’s SVG kit. Source-code licences are retained with the vendored libraries. NASA Earth pages informed the restrained scientific presentation and source captions; no NASA affiliation is implied.

Repository review: 3Blue1Brown Manim supplies a programme-driven mathematical animation precedent; Manim.js provides a p5.js approach. They were inspected as alternatives. This site implements its live figures directly in Three.js/D3/Anatomy rather than claiming to have executed those video engines. Anatomy, Three.js, D3 and p5.js are linked for the implementation lineage.

BootLoops was inspected for numerical evidence discipline. The earlier rainfall assessment records use of its pinned Emitall component for arithmetic claim tracing. This website revision independently recounts temperature scores and checks pure numerical functions with planted-error controls. It does not claim to have run the entire BootLoops harness or to have certified atmospheric truth.

The geographic explorer has a static satellite fallback, keyboard controls, user-controlled playback and reduced-motion support. Browser rendering could not be visually exercised in the current environment; numerical, source, link, syntax and deployment checks are reported separately. A Java source companion illustrates the reduced shared-flux balance; Java is not used for browser applets. Its source is supplied for review and was not compiled here because a JDK was unavailable. Java companion.