Emerging low-cloud feedback and adjustment in global satellite observations

Ceppi, Paulo; Wilson Kemsley, Sarah; Andersen, Hendrik; Andrews, Timothy; Kramer, Ryan J.; Nowack, Peer; Wall, Casey J.; Zelinka, Mark D.

doi:10.5194/egusphere-2025-5206

Preprints

https://doi.org/10.5194/egusphere-2025-5206

Preprints

24 Oct 2025

| 24 Oct 2025

Emerging low-cloud feedback and adjustment in global satellite observations

Paulo Ceppi, Sarah Wilson Kemsley, Hendrik Andersen, Timothy Andrews, Ryan J. Kramer, Peer Nowack, Casey J. Wall, and Mark D. Zelinka

Abstract. From mid-2003 to mid-2024, a decrease in low-cloud amount enhanced the absorption of solar radiation by 0.22±0.07 W m^-2 decade^-1 (±1σ range), accelerating the energy imbalance trend during that period (0.44 W m^-2 decade^-1). Through controlling factor analysis, here we show that the low-cloud trend is due to a combination of cloud feedback and adjustments to aerosols and greenhouse gases (respectively 0.07±0.01, 0.06±0.01, and 0.05±0.03 W m^-2 decade^-1), which jointly account for 82 % of the trend. The contribution of natural climate variability is weak but uncertain (0.03±0.07 W m^-2 decade^-1), owing to a poorly constrained trend in boundary-layer inversion strength. Importantly, the observed low-cloud radiative trend lies well within the range of values simulated by contemporary global climate models under conditions close to present day. Any systematic model error in the representation of present-day global energy imbalance trends is thus likely to originate in processes other than low clouds.

Received: 20 Oct 2025 – Discussion started: 24 Oct 2025

Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Chemistry and Physics.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Paulo Ceppi, Sarah Wilson Kemsley, Hendrik Andersen, Timothy Andrews, Ryan J. Kramer, Peer Nowack, Casey J. Wall, and Mark D. Zelinka

Status: final response (author comments only)

RC1: 'Reviewer comment on egusphere-2025-5206', Anonymous Referee #1, 26 Nov 2025

Ceppi et al. use cloud-controlling factor (CCF) analysis to investigate the role of low clouds in the observed trends of Earth's absorbed solar radiation. They combine the CERES-FBCT satellite product with reanalysis and model data to quantify how CCFs influence low clouds and how those in turn affect the amount of absorbed solar radiation. They decompose the trend into unforced and forced components, where the latter is further split into cloud feedback (to surface warming) and rapid aerosol and GHG adjustments. They find that all three of them contribute to a forced radiative trend, with cloud feedback being the strongest contributor, whereas the unforced component seems smaller but very uncertain. In my view the study is a well-founded, important and timely contribution to the dynamically evolving understanding of the observed radiative trends and certainly suitable for publication as an ACP letter, although I think that some points would benefit from clarification.
Before providing my specific comments, I'd like to mention that I'm wondering how these results based on local CCFs fit (or not) with recent work that links the observed cloud and SW absorption trends primarily to large-scale circulation trends rather than local "within-regime" factors (primarily Tselioudis et al. 2025). Could the authors add a brief discussion about this?
Overall, I recommend to publish this paper as an ACP letter subject to minor revisions.
Specific comments (including minor technical ones):
L5: "The contribution of natural climate variability is weak but uncertain [...], owing to a poorly constrained trend in boundary-layer inversion strength"; It remains somewhat unclear to me why the large uncertainty related to trends in EIS should affect the unforced component much more than the forced component. Can this be clarified?
L9 (and elsewhere): "... processes other than low clouds."; maybe this is a bit meticulous, but "low clouds" are not a process. Maybe reformulate to something like "... processes unrelated to low clouds." or so.
L37: "SW low-cloud anomalies [...] made a large contribution amounting to half of this decadal trend [of Earth’s global energy imbalance]"; It seems that the CERES total SW trend is even something like 0.8W/m2/dec, couteracted by a LW trend of around -0.3W/m2/dec (e.g., Fig. 3 in Myhre et al. 2025). Do I infer correctly that (i) the contribution of low clouds to the total SW trend is only around a quarter and that (ii) the bulk of the rest would then likely be due to mid- and high-level clouds (with more of a LW compensation)? Maybe that's worth to metion.
Figure 1: (i) I recommend to add units to the decadal trend numbers. (ii) In panel d, it would be good to mention/remind that this is about LOW-CLOUD trends, e.g., by changing the title into "Low-cloud trend contributions". (Also in the corresponding part of the caption.)
L50: "... nearly all of the RSWlow trend is associated with decreasing cloud amount, as opposed to decreasing optical depth"; I'm wondering if observational uncertainties in the way how cloud amount and optical thickness are distinguished (including the choice of a threshold from which point something is considered cloud vs. no cloud) might affect this. Related, the near-global averaged C_low seems to have much less of a trend component than R_SWlow, and in particular does not mirror the increase of R_SWlow of the last few years, making me wonder if this could be related to recent aerosol-induced changes in cloud optical thickness after all.
Fig. 2: I recommend to add units to the decadal trend numbers. (This also holds for Figs. A3 and A4.)
Fig. 3: Given that GHG adjustment seems to be an important contributor, I'd like to see a version of panel f with zoomed colorbar, plus a minimum explanation/hypothesis in the text indicating the possible physics behind this adjustment.
L81: "... given the known large decadal variations in low-cloud feedback."; is this indeed meant in the sense of decadal variations that would change (temorarily) the "background state" which would then result in modified low-cloud FEEDBACK during that limited time period? Or does this actually not relate to the feedback but just the clouds themselves, so something like "... given the known large decadal variations in low-cloud cover.", which I consider much more plausible? Or, a third option, is it about decadal variations in DIAGNOSED low-cloud feedback, given that limited observational periods will certainly affect estimates of the feedback?
Paragraph starting L80: Is it possible that the smallness of the uncertainty in the forced R_SWlow component is partly due to the assumption that the GMST trend is completely forced?
L97: "Comparing with amip minimises differences in CCF trends between models and observations, thus highlighting the role of the cloud-radiative sensitivities."; Is "minimises" here really the case? I mean, aspects like EIS, which as you show are more related to T_700hPa than T_srf, may still be rather unconstrained by the prescribed SST. Maybe "reduces" would be more appropriate?
L138: "the observed substantial low-cloud radiative trend cannot be interpreted as evidence of an unexpectedly strong low-cloud feedback that climate models are systematically missing (Goessling et al., 2025)"; The Goessling et al. paper does not make a strong statement about this being the main culprit, but mentions an upper-range low-cloud feedback just as one of three possible contributors (besides aerosols and variability).
Paragraph starting L165: In principle one can retrace the CCFs to earlier literature where there's more explanation and physical argumentation around them. However, given that this chain of studies is somewhat long/complex, I would find it helpful if brief explanations of the physical rationale (and definition, see next point) of these seven CCFs could be repeated here.
L167: "sea-surface temperature (SST) advection, SSTadv"; if I'm not mistaken, this is about temperature advection by near-surface winds, where the SST gradient strongly governs the air-temperature gradient, but it's not the same as actual "sea-surface temperature advection", which sounds like ocean surface velocities were involved. Some clarification would be good.
L203: Does this equation need to be applied iteratively from top (low pressure) to bottom (high pressure), so that U(p_i) is then always something like sum(L_n(p<p_i))? And does that ultimately yield a total cloud cover that is consistent with the CERES total cloud cover, or am I thinking wrong here?
L235: "The sensitivities Θi are calculated via ridge regression, where all variables have been deseasonalised, and the CCF predictors have been standardised"; I do not quite see the justification of computing "all-year" sensitivites given that they could well vary considerably seasonally (maybe as much as regionally in places?) due to seasonal changes of the background conditions, in particular in the extratropics. Is there evidence that this is not the case? I think that should be clarified.
L265: Given that dX/dT_for and dX/dT_GHG are based on different sets of models, I'm wondering if dX/dT_for and thus the resulting dX/dT_SST would be similar if just the same subset of models was used?
L266: Similarly, here dX/dT (obs-based) and dX/dT_for (model-based) stem from different datasets, so I'm wondering how the resulting dX/dT_unfor would look like if they were obtained from consistent data. For example, if one would use a single ensemble member of a CMIP historical/scenario model simulation as surrogate observation and base dX/dT_for on a (large) ensemble of just that same model, would the diagnosed unforced components exhibit quasi-random patterns of similar magnitude (compared to the right column of plots in Fig. A2)? Could that provide evidence for the validity of the method, whereas, if magnitudes are much smaller, would that suggest that the "unforced" components found here may contain considerable amounts of in reality forced changes?
L293+294: "for the GHG adjustment trend we take the spread in CMIP6 model-simulated GHG adjustment as a measure of uncertainty, using eight models with available data" and "we combine our eight estimates of the RSWlow GHG adjustment trend with the 20 estimates of the sum of other trend contributions [...], yielding a 160-member ensemble"; my understanding is that the GHG adjustment is derived from the piClim-ghg/control simulations, and according to Tab. A1, that data is available from ten models, not eight. Have I misunderstood this?
L326: "the CCF trends are in disagreement, with CAMS showing a weak decrease and MERRA2 a weak increase in mass concentration. This is contrary to our expectation of a clear decrease in sulfate aerosol concentration, particularly following the introduction of new shipping regulations in 2020"; Could this also be related to natural variations in aerosol concentrations (e.g., wildfires, even if sulfate is not a typical wildfire aerosol)?
References:
Tselioudis et al. (2025), Contraction of the World’s Storm-Cloud Zones the Primary Contributor to the 21st Century Increase in the Earth’s Sunlight Absorption, https://doi.org/10.1029/2025GL114882
Myhre et al. (2025), Observed trend in Earth energy imbalance may provide a constraint for low climate sensitivity models, https://doi.org/10.1126/science.adt0647

Citation: https://doi.org/10.5194/egusphere-2025-5206-RC1
RC2:
'Comment on egusphere-2025-5206', Anonymous Referee #2, 23 Dec 2025
General Comments:
This paper uses satellite observations, reanalysis, and climate model data in a cloud controlling factor (CCF) analysis framework to estimate the contribution of low cloud changes to the shortwave (SW) component of the trend in Earth’s energy imbalance (EEI) for July 2003 to June 2024. They find that low cloud changes alone contribute 0.22 Wm-2 per decade to the trend in SW top-of-atmosphere (TOA) radiation, which is substantial compared to the observed EEI trend of 0.44 Wm-2 per decade. Unfortunately, the paper does not state what the compensating LW contribution from low cloud changes is. The authors further show that the low cloud SW trend is a result of low cloud feedback, sulfate aerosol adjustment, and GHG adjustment, which account for 82% of the SW low-cloud trend. The authors claim that natural climate variability plays a minor role in explaining the trend. Comparisons between observation-based results and CMIP6 AMIP and historical global climate simulations suggests that the observed SW low-cloud trends lie within the range of the simulated trends. The implication is that underestimation of the observed EEI trend by state-of-the-art climate models reported in earlier studies is unlikely to be due to model representation of low cloud changes.
This is a very interesting paper that makes an important contribution to our understanding of the trend in EEI. I only have two major comments and numerous minor ones.
Major Comments:
Considering recent work (likely) published after this manuscript was submitted, it would be helpful if the authors commented on the work by Park and Soden (2025; https://www.science.org/doi/10.1126/sciadv.adv9429), who did a very similar analysis but reached a very different conclusion about the role of aerosols in explaining the SW low-cloud trend. Park and Soden (2025) used many of the same datasets and analysis steps as in the present paper. While the present paper finds that aerosols contribute 0.06+/-0.01 Wm-2 per decade (29%) to the 0.21 Wm-2 per decade low-cloud SW trend, Park and Soden (2025) find the aerosol contribution is negligible (−0.006+/-0.028 W m−2 per decade) primarily due to compensation between decreases in aerosol concentration in the northern hemisphere and increases in the southern hemisphere. Importantly, the difference between estimates from these two studies exceeds the stated uncertainties. A key methodological difference is that Park and Soden (2025) use the natural logarithm of sulfate mass concentration at 925 hPa (SO4) from MERRA-2 & CAMS, the natural logarithm of MODIS Aerosol Index, and cloud droplet number concentration from MODIS while the present study uses MERRA-2 AOD in the CCF regression. Park and Soden (2025) considered two approaches: that of Wall et al (2022) and a scheme that explicitly accounts for aerosol activation rate when determining susceptibility of the SW low-cloud radiative effect to variations in aerosol concentration. Considering the discrepancies between the two studies, it seems worthwhile for the authors to include a comment about the Park and Soden (2025) paper and perhaps revise their apparent confidence in their aerosol result. A further complicating factor is that CMIP6 climate models simulations suggest an even larger contribution by aerosol-cloud interactions (Hodnebrog et al., 2024), but those are questionable given that the aerosol forcing data used are outdated.

The authors state that low cloud changes make a substantial contribution to the trend in EEI but only quantify the SW component. For completeness, they should also quantify the compensating LW trend contribution from low cloud changes. If that is not feasible, instead of comparing the SW low cloud contribution to the EEI trend (i.e., 0.44 Wm-2 per decade), they should compare its contribution to the total trend in SW, which is much larger (~0.8 Wm-2 per decade).

Minor Comments:
Line 28-33:
Why start in July 2003? The combined Terra and Aqua data are available since July 2002, and Terra-only since March 2000.

Note that climate model forcing and adjustments are also used alongside observations to diagnose low cloud radiative trends.

Line 37: Please note that SW is defined positive downward.
Line 39: “…amounting to half of this decadal trend”.
What about LW? Is the trend contribution zero? The text makes it seem like low cloud changes alone account for 50% of the trend in Earth’s energy imbalance. That is only true if their contribution to the LW trend component is zero, which is assumed but not shown.
Line 43: “…low clouds exhibit a positive radiative sensitivity to surface temperature…”
This is unclear/vague. Please explain that “positive radiative sensitivity” implies increased absorption of solar radiation.

Figure 1d:
It seems the trends in Fig. 1d are derived from 12-month running mean time series, which exhibit substantial autocorrelation. However, there is no mention of how the trend uncertainties are calculated. Do the uncertainties account for autocorrelation in the data?

It seems odd to include the cloud feedback contribution in the “Forced” category (second bar from bottom of Fig. 1d). Isn’t it more appropriate for this to be a separate category (i.e., SST-mediated response to forcing?)

Lines 87-88: In light of the recent paper by Park and Soden (2025), who also used MERRA-2 and CAMS but considered the natural logarithm of sulfate mass concentration at 925 hPa as opposed to log(AOD), and found a negligible trend contribution from aerosols, is this an accurate statement. It seems that the aerosol contribution is highly uncertain.
Lines 150-151: “We use gridded global satellite observations of cloud amount and top-of-atmosphere radiative fluxes from the CERES Flux-By-Cloud-Type (CERES-FBCT) product (Sun et al., 2022)”.
CERES-FBCT is available daily and monthly. Which of these is used here?
Line 155 : “Note that CERES-FBCT provides cloud-radiative effect rather than true cloud-induced radiative anomalies”
FBCT provides all-sky and clear-sky TOA fluxes along with TOA fluxes for 42 pc-tau cloud types along with their amounts in each gridbox. The all-sky flux is thus weighted sum of clear-sky and individual cloud type fluxes, where the weights are given by the respective clear or cloudy sky amounts. Cloud-radiative effect can be computed from all-sky and clear-sky, but it’s not part of the data product.
Lines 156-158: “These are however expected to be much smaller for SW than longwave (LW) fluxes, particularly since we exclude regions poleward of 60◦ where surface albedo changes are largest (Raghuraman et al., 2023).”
While this may be true for monthly anomalies, it isn’t so obvious when considering trends, as is done here. For example, we know that there is a strong trend in water vapor, which affects SW TOA flux.
Line 159: Please clarify what RSWlow is. The prior sentences discuss CRE, but Line 159 mentions low cloud SW anomalies. Is RSWlow the low cloud flux weighted by low cloud amount or is it CRE for low clouds? If it’s more like CRE, consider using “CRE” instead of RSW in the name.
Lines 162-163: “The cloud-radiative changes diagnosed from CERES-FBCT provide a close match to those obtained from CERES-EBAF (Loeb et al., 2024b)”
It is worth pointing out that this statement is true only for the Terra+Aqua period. It has not been demonstrated for the NOAA-20 period, which is considered in this study.
Also, FBCT is a daytime only product whereas EBAF accounts for both daytime and nighttime LW.
Line 195: “CERES-FBCT data consists of clear-sky radiative flux Rclr, cloudy-sky radiative flux Rcld, and cloud amount.”
It’s worth pointing out that all-sky TOA flux in a gridbox is the sum of Rclr plus the sum of f(p, tau)*Rcld(p, tau) for all (p, tau) cloud types, where f(p, tau) is the cloud amount for cloud type (p, tau). Thus, the contribution to all-sky from a given cloud type is Rcld(p, tau)*f(p, tau).
Lines 196-198: “Following previous literature, we categorise clouds in the lowest two bins (p > 680 hPa) as low clouds. We use only the shortwave (SW) component of the fluxes, denoted RSWclr and RSWcld.”
This is confusing. How does RSWcld differ from RSWlow, introduced in line 159? It seems RSWcld is the cloudy SW radiative flux for a cloud in a given (p, tau) bin from the FBCT product while RSWlow is a derived quantity (see line 210) analogous to cloud radiative effect for a specific (p, tau) cloud type. Using such similar symbols (RSWcld and RSWlow) to define these quantities is unnecessarily confusing. Consider using something like “CRE_SW_low” instead of RSWlow to make the distinction clearer.
Also, is RSWcld = cloudy-sky radiative flux (Rcld) times cloud amount? Please clarify.
Line 203: Equation for non-obscured low cloud amount Ln. A more straightforward equation for this is: Ln(p, tau) = L(p, tau) / (L+Clr), where L is total low cloud amount and Clr is the clear-sky amount.
Line 205: “absence of clouds”
Should “clouds” be preceded by “low” here?
Line 206:“We can then calculate the low-cloud contribution to top-of-atmosphere radiative flux”
This sentence contradicts the sentence on line 211, which (correctly) describes the equation on line 209 as contribution of low clouds to SW cloud-radiative effect. Please clarify.
Line 209:
(a) Since the reader does not know if daily or monthly FBCT data are used, it’s unclear if this equation refers to a monthly mean in a gridbox derived from daily FBCT values or a monthly or longer average of multiple gridboxes. Also, U overbar is not defined.
(b) If this equation were summed over all 42 cloud types, would that equal the overall gridbox CRE, in a manner similar to what one gets for all-sky TOA flux in a gridbox (i.e., sum of Rclr plus the sum of f(p, tau)*Rcld(p, tau) for all (p, tau) cloud types)? I wonder if substituting L with Ln means there is a lack of closure between the gridbox CRE and the sum of CREs from all individual cloud types. One can test this by checking if CRE – RSWlow equals RSWhigh calculated from individual cloud types in p=3 to 7, where CRE is the gridbox cloud radiative effect. If these are not consistent, what are the implications for how we interpret trends in RSWlow?
Lines 226-228: “Finally, we concatenate the Terra–Aqua data up to February 2023 with the modified NOAA-20 data from March 2023 to July 2024, to produce a single record from September 2002 to July 2024.”
Why transition to NOAA-20 in February 2023? The Loeb et al. (2024) paper makes the transition to NOAA-20 in April 22 to avoid any impact of Terra and Aqua orbit drift. How does the date of the transition impact the trend?
Line 230: “Ceppi et al. (2024), with the addition of an aerosol CCF following Wall et al. (2022).”
A more recent study by Park et al (2024; https://doi.org/10.5194/egusphere-2024-2547) argues that the activation rate of cloud droplet number concentration in response to variations in aerosol (e.g., sulfate aerosols) needs to be explicitly accounting for, otherwise the influence of aerosol is overestimated. At the very least, this possibility should be acknowledged.
Line 231: “SW low-cloud radiative anomalies”
Shouldn’t this be “SW low-cloud CRE anomalies” instead?
Lines 258-259: “This assumes that the rate of global warming is forced on this 21-year timescale, and that the CCF forced responses scale with global warming.”
Is there any justification for using this assumption?

Figure A3: Please indicate what “log(s)” is in the caption or main text (e.g., lines 180-182). It is only revealed that s is sulfate mass concentration on line 324, well after the figure is introduced.
Citation: https://doi.org/10.5194/egusphere-2025-5206-RC2

Paulo Ceppi, Sarah Wilson Kemsley, Hendrik Andersen, Timothy Andrews, Ryan J. Kramer, Peer Nowack, Casey J. Wall, and Mark D. Zelinka

Viewed

Total article views: 615 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
291	298	26	615	19	20

HTML: 291
PDF: 298
XML: 26
Total: 615
BibTeX: 19
EndNote: 20

Views and downloads (calculated since 24 Oct 2025)

Month	HTML	PDF	XML	Total
Oct 2025	83	34	4	121
Nov 2025	119	89	13	221
Dec 2025	89	175	9	273

Cumulative views and downloads (calculated since 24 Oct 2025)

Month	HTML	PDF	XML	Total
Oct 2025	83	34	4	121
Nov 2025	119	89	13	221
Dec 2025	89	175	9	273

Viewed (geographical distribution)

Total article views: 600 (including HTML, PDF, and XML) Thereof 600 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 23 Dec 2025

Short summary

Recent decades have seen a marked decrease in global low-level cloud cover, leading to more sunlight heating the Earth. This trend is poorly understood, raising the concern that clouds may amplify global warming more than previously thought. We show that the cloud decrease is mostly caused by human forcing on climate, and that it agrees with previous estimates of how clouds respond to decreasing aerosol pollution, increasing greenhouse gas concentration, and their effects on global temperature.


Total:	0
HTML:	0
PDF:	0
XML:	0