Particle X Climate

22 October 2026

Wilcox et al.

Takeaways, three ideas, and research directions.

I asked ten climate models what aerosols do, and they all agreed, to within a factor of 3.5. — Dr. D

Chosen paper

Distinct climate responses to regional aerosol reductions: an overview of the regional aerosol model intercomparison project

Laura J Wilcox, Bjørn H Samset, Camilla Weum Stjern, Robert J Allen et al. (30 authors in the Crossref record)

Year / venue: 2026, Environmental Research: Climate (IOP Publishing), "in press"

Status: a peer-reviewed Accepted Manuscript, "the version of the article accepted for publication including all changes made as a result of the peer review process", not the version of record.

Design (author documented): RAMIP Tier 1 experiments perturb SO2, BC and OC to SSP1-2.6 levels, globally and over four regions (East Asia, South Asia, Africa and the Middle East, North America and Europe), against an SSP3-7.0 baseline; There are 10 participating models, and 10 ensemble members were requested for each coupled transient experiment. Responses are shown for 2035–2049; ERF comes from paired 2050 fixed-SST runs (piClim-370*): "These simulations use pre-industrial SSTs, sea ice, and land use, with 2050 anthropogenic emissions of aerosols and greenhouse gases (Table 1)." (PDF p. 6); Historical ERF comes from RFMIP piClim-control and piClim-aer (Table 2 caption, PDF p. 9).

UKESM row, Table 2 (PDF p. 9): UKESM1-0-LL uses atmosphere MetUM-HadGEM3-GA7.1, aerosol "UKCA-GLOMAP-mode" and chemistry "UKCA-StratTrop". Its ECS and historical ERF are printed as "5.34, -1.11".

Licence: CC BY 4.0. The cover sheet says "this Accepted Manuscript is available for reuse under a CC BY 4.0 licence immediately".

Data: "CMORized data for the Tier 1 experiments presented here is available for download from the Centre for Environmental Data Analysis (CEDA)" (PDF p. 31). The UKESM1-0-LL collection is https://catalogue.ceda.ac.uk/uuid/63233813672245a2ba3c6de90c4cfaee (record checked 22 Oct; variable-level contents not checked).

DOI: https://doi.org/10.1088/2752-5295/aeb71a

HTML: https://iopscience.iop.org/article/10.1088/2752-5295/aeb71a (marked Accepted Manuscript, checked 22 Oct)

Site takeaways

What they got right (main result)

Air-quality policy is about to cut aerosol emissions in many parts of the world, and RAMIP asks how much that will change climate, where, and whose emissions matter most. Ten Earth system models ran the same experiments, with ten ensemble members requested for each. Sulfur dioxide, black carbon and organic carbon follow a clean-air pathway (SSP1-2.6) instead of staying high (SSP3-7.0), either everywhere or only over East Asia, South Asia, Africa and the Middle East, or North America and Europe. The forcing comes from paired 2050 runs with fixed sea surface temperatures, and the climate responses are averaged over 2035–2049.

Cutting emissions everywhere gives an effective radiative forcing of 0.71±0.27 W m−2 in 2050 and a global-mean warming of 0.31±0.12 K, where the ± is the spread between models (one standard deviation). The authors find the warming occurs largely over the Northern Hemisphere, causing a northward shift of the tropical rain belt and stronger Northern Hemisphere monsoons. Where the cut happens matters. In the authors' words, reductions over North America (part of the combined North America-and-Europe experiment) and East Asia have hemisphere-wide impacts and account for most of the global-mean response. South Asian cuts act mainly locally, African and Middle Eastern cuts change rainfall across the tropics, and a region's climate can be as strongly influenced by cuts made elsewhere as by its own. Ensemble size matters too: in the authors' South Asia example, a smaller ensemble can predict a rainfall response of the opposite sign to the 10-member mean. The paper states that "CMORized data for the Tier 1 experiments presented here is available for download from the Centre for Environmental Data Analysis (CEDA)".

What they didn't quite nail

The regional pieces do not add up. The four regional forcings sum to only 54% of the forcing from global cuts, and the authors call for caution in attribution. Part of the gap is South America, which the first design left out, and the authors are already running "Rest of the world" experiments to close it.

Two questions stay open. The models disagree more about future forcing than historical forcing (a factor of 3.5 spread for the 2050 global cuts, against 1.8 historically), and some, including UKESM1-0-LL, have relatively small future forcing compared with their historical values. The authors suggest forcing "may be state dependent in some models", perhaps through changes in cloud properties with warming or natural aerosol feedbacks, but do not test either. And because the ten models share no aerosol or chemistry scheme, RAMIP is well placed to ask how model structure shapes the response, a question the paper raises rather than answers. Over Africa the forcing differs between models even in sign in places, which the authors put down to different dust and cloud feedbacks.

Reader caveats (not paper flaws): this is tropospheric air-pollution aerosol under two emission pathways, with forcing from runs that use pre-industrial sea surface temperatures and 2050 emissions. The version read is the peer-reviewed Accepted Manuscript, ahead of the typeset version of record.

Ideas

Three ideas, each tied to an open problem the authors document. Each is a proposal; nothing here assumes what UKESM or the other models will show.

Idea 1

Why is UKESM1-0-LL's future aerosol forcing smaller than its historical forcing?

UKESM1-0-LL already has two fixed-SST forcing pairs: the RFMIP pair (1850 vs 2014 aerosol) and the RAMIP pair (SSP3-7.0 vs SSP1-2.6 aerosol in 2050, with 2050 greenhouse gases). The idea is to follow the forcing chain through each pair, globally and region by region: forcing per unit SO2 emission and per unit aerosol optical depth, sulfate formed per unit SO2, and cloud droplet number and size per unit optical depth.

The catch is that the two pairs differ in many ways at once, not just one. They differ in greenhouse-gas year, in how polluted the starting point is, in whether aerosol is added or removed, in the size and location of the change (the authors link the larger future spread partly to "the change in the spatial pattern of aerosol emissions"), in which species change, in the chemistry and oxidant state that sets how much sulfate forms, in natural emissions, and possibly in model version (several of these are our inference). Both pairs use pre-industrial sea surface temperatures, so "warming" here can only mean the land and atmosphere adjusting to 2050 greenhouse gases (inferred). The comparison can describe where UKESM's chain differs between the pairs, but it cannot pin the difference on any one cause or show saturation as the reason. An optional pair of new runs that changes only the greenhouse-gas year (2050 aerosol with 1850 greenhouse gases) would isolate that single factor and nothing more; if it also moved ozone or chemistry, it would isolate a greenhouse-gas-plus-chemistry year. That design is ours.

Similar base-state tests exist for other models. GFDL-ESM4.1 found the global-mean aerosol ERF "relatively robust" while "regional aerosol ERF differs strongly depending on the background climate and composition" (Zhang et al., 2026), and CESM2 showed "statistically insignificant increases in aerosol ERF when estimated in the different background climates" (White et al., 2025). We found no published UKESM1 test of this kind, which reflects our search, not a finding. The diagnostic stage needs no new simulations: the RFMIP pair is on ESGF, and the RAMIP UKESM runs list the cloud, optical-depth and sulfate-production variables on CEDA. Ananth's UKCA/GLOMAP background suits reading these diagnostics; that link is ours (inferred), not the paper's.

Idea 2

How much do natural aerosol changes add to UKESM1-0-LL's response? Africa as the test case

The authors name natural aerosol feedbacks as one possible reason for state-dependent forcing, and dust and cloud feedbacks for the African sign differences, but quantify neither. This idea would budget the changes in dust, sea salt and secondary organic aerosol in UKESM between the RAMIP reduction runs and their baselines, focusing on the global and Africa-and-Middle-East cuts, and ask how large they are next to the anthropogenic change.

The first step comes before any budget: check whether UKESM's natural emissions (dust, sea salt, DMS and BVOC) are interactive or prescribed in the fixed-SST runs. Allen et al. (2026b) print that these runs use "natural aerosol and precursor gas emissions from SSP3-7.0", and if UKESM's are prescribed there, the fast-response half of the idea is empty by design and the work rests on the coupled runs alone. UKESM computes several natural emissions online in its CMIP6 configuration (Mulcahy et al., 2020), but whether that holds here is unchecked (inferred). A simple cross-check is to compare the archived dust emission field between the paired runs. The public directory listings show dust and sea-salt fields but no DMS, secondary-organic or BVOC emission field, so coverage must be confirmed run by run. The closest precedent, Thornhill et al. (2021), quantified such feedbacks for a quadrupling of CO2, not for aerosol-policy cuts.

Idea 3

Does aerosol-scheme design organise the spread between models?

The authors describe RAMIP as "well suited to exploring the effects of model structural uncertainty", since no two models share an aerosol or chemistry scheme. The companion paper by Allen et al. (2026b) has already split the forcing into its radiation and cloud parts and classed the schemes as modal or bulk. This idea traces each model's chain from emission change to aerosol optical depth, droplet number and aerosol–cloud forcing, and asks whether forcing or droplet response per unit optical depth groups by scheme class. With nine or ten models this is a diagnostic, not a statistical test. UKESM1-0-LL, with the modal GLOMAP-mode scheme, is a natural worked case: Allen et al. report an East Asian spread from "−0.82 W m−2 in UKESM1-0-LL to 3.83 W m−2 in MIROC6". By our reading, neither paper relates the spread to scheme properties (inferred).

Feasibility depends on archived variables. The RAMIP data request lists optical depth and droplet number, and condensation-nuclei fields only "where possible". Allen et al. said their data would follow acceptance via ESGF nodes. When we looked, ESGF returned no RAMIP datasets, while the CEDA catalogue lists fixed-SST collections for eight models, including UKESM1-0-LL, but no piClim-370 record for two of Allen et al.'s nine fixed-SST models (GFDL-SPEAR_LO and GISS-E2-1-G); we downloaded no files and have not confirmed these are the exact data behind Allen et al. Ananth's GLOMAP-mode background fits the UKESM case (inferred).

The non-linearity left by the missing South America region is not an idea here, as the authors are already running "Rest of the world" simulations.

External citations used (checked on 22 October 2026 against Crossref, the open PDFs, CEDA or ESGF)

  1. Wilcox, L. J., et al. (2023). The Regional Aerosol Model Intercomparison Project (RAMIP). Geosci. Model Dev., 16, 4451–4479. https://doi.org/10.5194/gmd-16-4451-2023
  2. Allen, R. J., et al. (2026b). Decomposing the global and regional aerosol effective radiative forcing associated with strong versus weak air quality policies by Mid-21st century. Environ. Res.: Climate, 5, 025014. https://doi.org/10.1088/2752-5295/ae5418
  3. Zhang, et al. (2026). The Dependence of Emissions‐Based Anthropogenic Aerosol ERFs on Background Composition and Meteorological States in the GFDL‐ESM4.1 Model. J. Geophys. Res. Atmos., 131, e2026JD046386. https://doi.org/10.1029/2026jd046386
  4. White, et al. (2025). Absence of Aerosol Indirect Effect Dependence on Background Climate State in NCAR CESM2. J. Climate, 38, 147–163. https://doi.org/10.1175/JCLI-D-23-0755.1
  5. Jia, H. and Quaas, J. (2023). Nonlinearity of the cloud response postpones climate penalty of mitigating air pollution in polluted regions. Nat. Clim. Change, 13, 943–950. https://doi.org/10.1038/s41558-023-01775-5
  6. Thornhill, G., et al. (2021). Climate-driven chemistry and aerosol feedbacks in CMIP6 Earth system models. Atmos. Chem. Phys., 21, 1105–1126. https://doi.org/10.5194/acp-21-1105-2021
  7. Mulcahy, J. P., et al. (2020). Description and evaluation of aerosol in UKESM1 and HadGEM3-GC3.1 CMIP6 historical simulations. Geosci. Model Dev., 13, 6383–6423. https://doi.org/10.5194/gmd-13-6383-2020
  8. O'Connor, F. M., et al. (2021). Assessment of pre-industrial to present-day anthropogenic climate forcing in UKESM1. Atmos. Chem. Phys., 21, 1211–1243. https://doi.org/10.5194/acp-21-1211-2021
  9. Grosvenor, D. P. and Carslaw, K. S. (2020). The decomposition of cloud–aerosol forcing in the UK Earth System Model (UKESM1). Atmos. Chem. Phys., 20, 15681–15724. https://doi.org/10.5194/acp-20-15681-2020
  10. Ghan, S. J. (2013). Technical Note: Estimating aerosol effects on cloud radiative forcing. Atmos. Chem. Phys., 13, 9971–9974. https://doi.org/10.5194/acp-13-9971-2013
  11. CEDA collection record: UKESM1-0-LL output prepared for the Regional Aerosol Model Intercomparison Project (RAMIP). https://catalogue.ceda.ac.uk/uuid/63233813672245a2ba3c6de90c4cfaee
  12. ESGF (CEDA index) dataset listing: UKESM1-0-LL piClim-aer and piClim-control, Amon rsut.

Research directions

The three ideas form one programme around a single sentence in the paper: some models, UKESM1-0-LL among them, have relatively small future aerosol forcing compared with their historical values, and the authors suggest forcing "may be state dependent". Clouds and natural aerosol are the two candidate routes, and underneath both sits the question of whether the answer depends on how a model represents aerosol in the first place.

Idea 1 is the cheapest start, because the forcing pairs already exist; but since they differ in many ways at once, comparing them shows where UKESM's forcing chain changes without naming a single cause, and only new runs that change the greenhouse-gas year alone can isolate that one factor. Idea 2 turns the natural-aerosol route into a budget for one model, starting with whether its natural emissions can respond at all, with Africa as the test region. Idea 3 widens the lens to every model and asks whether the spread follows the way each scheme turns emissions into cloud droplets.

The guard-rails hold throughout: this is tropospheric air-pollution aerosol under two emission pathways, forcing comes from fixed-SST runs, responses are averaged over 2035–2049, and nothing here predicts what the models will show. Ananth's GLOMAP-mode experience is relevant to reading UKESM's droplet and size diagnostics; that link is ours (inferred), not the paper's. Each step tests a specific sentence in the paper. Idea 1's optional runs that change only the greenhouse-gas year would be new simulations; everything else uses output that is already archived or listed in the RAMIP data request.

Mrs. Vanguard