pith. sign in

arxiv: 2605.20925 · v1 · pith:MQEOV7KZnew · submitted 2026-05-20 · ⚛️ physics.ao-ph

Blending machine learning and physics-based approaches for weather and climate: a typology

Benjamin J Shipway , Caroline Bain , David Walters , Ben B. B. Booth , Ian Boutle , Robin T. Clark , Katherine L. Hill , Elizabeth Kendon
show 1 more author
Simon B. Vosper
This is my paper

Pith reviewed 2026-05-21 02:03 UTC · model grok-4.3

classification ⚛️ physics.ao-ph
keywords machine learningphysics-based modelsweather predictionclimate modelinghybrid approachestypologyblended modelingprediction systems
0
0 comments X

The pith

A typology classifies different ways to blend machine learning with physics-based models for weather and climate prediction.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to define a typology that organizes the many possible ways of combining machine learning techniques with established physics-based models. This matters to a sympathetic reader because it supplies a shared vocabulary for deciding how much and in what manner to introduce the speed and flexibility of machine learning without losing the reliability and interpretability that physics-based systems provide. The typology can classify current systems and map out paths for gradual or complete upgrades to prediction capabilities. If the framework works as described, the community gains a practical tool for planning the shift to next-generation systems that draw on both approaches.

Core claim

The authors present a typology of blended modelling approaches that distinguishes different strategies for integrating machine learning into physics-based weather and climate models. They argue that these blended approaches offer a practical route to innovation by pairing the speed and adaptability of machine learning with the robustness, trust, and interpretability of physics-based systems. The typology is intended both to classify existing modelling systems and to identify routes for incremental or wholesale development and implementation of new capabilities, thereby supporting informed decision-making and strategic planning across the community.

What carries the argument

The typology, a classification system that sorts blending strategies according to how machine learning components are combined with physics-based modelling systems and that outlines the benefits and limitations of each strategy.

If this is right

  • Existing modelling systems can be classified by their position within the typology to clarify how machine learning is currently used.
  • Different routes for gradual, incremental, or complete introduction of new machine learning capabilities become visible.
  • The combination of approaches accelerates incorporation of emerging science while preserving robustness and trust.
  • A common vocabulary supports clearer strategic planning for the transition to next-generation prediction systems.
  • The wider community can use the framework to navigate choices in hybrid model development.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The typology could be tested by applying it to a set of current operational models and checking whether the classifications align with how developers actually make blending decisions.
  • Similar classification schemes might prove useful in adjacent fields such as ocean or land-surface modelling where physics and data-driven methods also coexist.
  • If the typology is adopted, it could lead to shared benchmarks that compare the performance trade-offs of different blending levels.
  • The framework raises the question of how to quantify the right balance between machine learning speed and physics-based constraints for a given application.

Load-bearing premise

The proposed typology covers the main blending strategies that exist and that adopting its structured vocabulary will produce better informed decisions about model development.

What would settle it

A systematic review of existing weather and climate models that finds many blending methods falling outside the typology categories, or a follow-up study showing no change in planning practices after the vocabulary is introduced.

Figures

Figures reproduced from arXiv: 2605.20925 by Ben B. B. Booth, Benjamin J Shipway, Caroline Bain, David Walters, Elizabeth Kendon, Ian Boutle, Katherine L. Hill, Robin T. Clark, Simon B. Vosper.

Figure 1
Figure 1. Figure 1: We describe a spectrum of blending opportunities from independent-physics-based to independent-ML. Options in between make use [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Forecast tracks for Typhoon Ampil, showing NWP-only solution (red), hybrid [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AI downscaling emulators, trained on existing Regional Model simulations, [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
read the original abstract

The integration of machine learning (ML) with traditional physics-based models is reshaping the landscape of weather and climate prediction. On their own, ML-based and physics-based approaches each have significant benefits - but also challenges. Deploying both these approaches side by side has the potential to accelerate the pull through of emerging science in a trusted and practical way. But there are many choices that can be made to how we "blend" ML and established physics-based modelling systems to get the optimal benefits. This paper aims to provide a typology of blended modelling approaches and discusses some of the strategic benefits that come with them. It can be used not just to classify modelling systems, but also identify routes to gradual, incremental or wholesale development and implementation of new and emerging capabilities. These approaches provide a practical path to innovation by combining the speed and adaptability of machine learning with the robustness, trust, and interpretability of physics-based systems. By adopting a structured vocabulary and outlining the benefits and limitations of each approach, this framework supports informed decision-making and strategic planning, and can be used by the wider community to navigate the transition to next-generation prediction systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The paper proposes a typology of strategies for blending machine learning with physics-based models in weather and climate prediction. It describes the benefits and limitations of different blending approaches, argues that they combine ML speed and adaptability with the robustness and interpretability of physics-based systems, and positions the framework as a tool for classifying existing systems and guiding incremental or wholesale development toward next-generation prediction capabilities.

Significance. If the typology is adopted by the community, it could provide a shared vocabulary that supports more systematic decision-making when integrating emerging ML capabilities into operational weather and climate models. The manuscript's main contribution is conceptual organization rather than new empirical results or derivations; its value therefore rests on whether the categories prove useful for classification and planning.

minor comments (2)
  1. Abstract: the statement that the typology 'can be used not just to classify modelling systems, but also identify routes to gradual, incremental or wholesale development' is central but would be strengthened by a single concrete example of how one category maps onto an existing or planned modeling system.
  2. The section presenting the typology categories: each blending strategy is described at a high level; adding a short table that contrasts the categories on dimensions such as computational cost, interpretability, and data requirements would improve usability without altering the core argument.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their constructive review and recommendation of minor revision. The referee's summary accurately captures the manuscript's aim to provide a typology that supports classification and strategic planning for blended ML-physics systems in weather and climate prediction. We have no specific major comments to address point-by-point as none were raised in the report.

Circularity Check

0 steps flagged

No significant circularity in typology framework

full rationale

This paper is a conceptual classification and vocabulary paper that proposes a typology for blending ML and physics-based weather/climate models. It contains no equations, derivations, fitted parameters, or load-bearing mathematical steps. The central claims rest on descriptive categorization of strategies, benefits, and limitations rather than any self-referential logic, self-citation chains, or reductions of predictions to inputs. The framework is self-contained as a practical classification tool with no internal circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a conceptual typology paper with no quantitative models, data fitting, or derivations. No free parameters, new axioms, or invented entities are introduced beyond standard domain knowledge in atmospheric modeling.

pith-pipeline@v0.9.0 · 5759 in / 947 out tokens · 36564 ms · 2026-05-21T02:03:25.969964+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    V., Ford, R

    Adams, S. V., Ford, R. W., Hambley, M., Hobson, J. M., Kavčič, I., Maynard, C. M., Melvin, T., Müller, E. H., Mullerworth, S., Porter, A. R., Rezny, M., Shipway, B. J., & Wong, R. (2019). LFRic: Meeting the challenges of scalability and performance portability in weather and climate models. Journal of Parallel and Distributed Computing, 132, 383–

    work page 2019

  2. [2]

    R., & Ott, E

    https://doi.org/10.1016/j.jpdc.2019.02.007 Arcomano, T., Szunyogh, I., Wikner, A., Pathak, J., Hunt, B. R., & Ott, E. (2022). A hybrid approach to atmospheric modeling that combines machine learning with a physics-based numerical model. Journal of Advances in Modeling Earth Systems, 14, e2021MS002712. https://doi.org/10.1029/2021MS002712 Balaji, V., F. Co...

    work page doi:10.1016/j.jpdc.2019.02.007 2019

  3. [3]

    © Crown Copyright, Met Office 18 Subich, C., S. Z. Husain, L. Separovic, and J. Yang, 2025: Fixing the double penalty in data-driven weather forecasting through a modified spherical harmonic loss function. arXiv, 2501.19374, https://doi.org/10.48550/arXiv.2501.19374. Y.Q. Sun,P. Hassanzadeh,M. Zand,A. Chattopadhyay,J. Weare, & D.S. Abbot, Can AI weather m...

    work page doi:10.48550/arxiv.2501.19374 2025