Clouds dissipate quickly during solar eclipses as the land surface cools

Reducing the solar radiation incident on the Earth's lower atmosphere and surface has been proposed as a strategy to counteract global temperature rise when mitigation is insufficient. Solar geoengineering concepts include sunshades or reflective particles in space and stratospheric aerosol injection. General circulation models suggest that a 3.5–5.0% insolation reduction could offset warming and hydrological intensification under quadrupled CO2, but responses vary by latitude and may reduce tropical precipitation. Although clouds are central to Earth’s radiation balance, the impact of solar dimming on clouds is poorly understood; GCMs use idealized, long-term scenarios and parameterize short-term, small-scale processes like cloud formation, leading to uncertainties. Solar eclipses provide natural experiments with gradual insolation reductions over hours. Ground-based observations have noted rapid drops in temperature, winds, and turbulence, and anecdotal low-level cumulus dissipation before totality while mid/high clouds persist; however, quantifying timing is difficult due to cumulus variability and lack of a non-eclipse baseline. Satellite geostationary observations can continuously monitor large areas, but during eclipses, images are darkened and standard cloud retrievals are biased because reduced irradiance is not accounted for, leaving unknown how quickly clouds respond to varying obscuration. The study aims to correct satellite cloud measurements for eclipse-induced irradiance reduction and, together with large-eddy simulations, quantify the timing and magnitude of shallow cumulus response to solar obscuration.

Prior work on solar geoengineering indicates potential to offset warming with modest global insolation reductions, but highlights uncertainties in regional climate responses and precipitation changes (e.g., GeoMIP studies). The role of clouds in radiative forcing is substantial, yet cloud responses to dimmed sunlight remain uncertain in models that emphasize long-term equilibria and parameterize boundary-layer cloud processes. Ground-based eclipse studies documented rapid meteorological changes and anecdotal cumulus dissipation near totality, while modeling studies of eclipses have not analyzed cumulus sensitivity in detail. Geostationary satellites have hinted at cumulus reductions by comparing pre/post eclipse images, but retrieval biases under partial lunar shadow confound quantitative assessments. Corrections for eclipse-induced irradiance changes in satellite reflectances have been validated for other instruments (e.g., TROPOMI, DSCOVR/EPIC), motivating their application to geostationary cloud products to enable robust cloud evolution analysis during eclipses.

Satellite observations: The study used SEVIRI instruments aboard Meteosat-8 and Meteosat-10 (MSG) to obtain 15-minute TOA radiances across visible/near-IR channels, including the high-resolution HRV channel (1 km) and 11 narrowband channels (3 km). Shortwave reflectances were derived from radiances and calibrated (MODIS inter-calibration); longwave channels used EUMETSAT calibration. Cloud masks at low resolution were produced with the EUMETSAT NWC SAF v2021 algorithm, which also provided cloud top height. For high-resolution cloud analysis, the low-resolution mask was refined using HRV reflectances against a 16-day HRV climatology. Paired 0.6 and 1.6 µm channels retrieved cloud optical thickness (COT) and droplet effective radius, downscaled to HRV resolution. Non-cloudy pixels were assigned COT=0; sunglint-affected river pixels were filtered in the study area. Eclipse correction: TOA reflectances were corrected for eclipse darkening by dividing by (1 − f_o), where f_o is the pixel-specific solar obscuration fraction computed using Besselian elements, accounting for surface height, time, latitude/longitude, and wavelength-dependent limb darkening (coefficients at 0.635, 0.81, 1.64 µm, and 0.7 µm for HRV). The approach was previously validated up to f_o=0.92 with TROPOMI and applied similarly to DSCOVR/EPIC. Cloud algorithms were applied to corrected reflectances to yield corrected cloud mask and COT. Surface temperatures: Land surface temperature (LST) came from LSA SAF using SEVIRI 10.8/12 µm split-window retrievals (1–2 K uncertainty). Spatial averages over the study area were computed every 15 minutes, replacing cloudy pixels via nearest-neighbor interpolation using the corrected cloud mask. The eclipse-induced LST drop was estimated relative to the mean of comparable days. Sea surface temperature (SST) was from OSI SAF hourly products (uncertainty <1 K). Comparable days: From Sep–Oct 2004–2006, 100 days with smallest differences in total-day ERA5 sensible and latent heat flux sums (non-eclipse) were preselected, then refined by removing days whose mean absolute high-resolution cloud cover difference vs. the study day (06:00–09:00 UTC) exceeded 0.1. Eleven comparable days remained and were visually confirmed to have similar shallow cumulus with low COT (<5) before eclipse onset. Large-eddy simulations (DALES): DALES was configured on a 50×50 km domain, 100×100 m horizontal resolution, 209 vertical layers up to 14 km (20 m lowest layer, stretched by 1.01). Surface was flat. Initial horizontally homogeneous profiles (liquid potential temperature, total water specific humidity, winds) were from ERA5 at 02:00 UTC for the study area; pressure via gas law and hydrostatic balance. Time/height-dependent advective tendencies and geostrophic winds from ERA5 were prescribed; large-scale subsidence imposed at top boundary. Horizontal inflow/outflow of eclipse-induced mesoscale changes at domain boundaries was neglected. Sensible and latent heat fluxes were computed from near-surface gradients with stability-dependent aerodynamic resistances (Monin–Obukhov), calibrated via constant factors for roughness and soil moisture to match ECMWF fluxes in the reference case. Surface temperature T_sc was prescribed from SEVIRI LST (shifted by +2.0274 K to match ERA5 skin temperature at 02:00 UTC). In the no-eclipse reference, T_sc matched the eclipse case until 09:00 UTC and afterward used the average LST of comparable days (shifted similarly). Diagnostics: DALES liquid water specific humidity fields were regridded to 1 km to compare with SEVIRI. Column COT was computed by integrating droplet cross-section with assumed constant effective radius r_eff=10 µm (1×10⁻⁵ m), flagging columns with COT>1 as cloudy; cloud cover was the fraction of cloudy columns; mean COT averaged over cloudy columns. LCL was estimated from 10 m layer T and RH using an empirical formula; level of minimum buoyancy flux (LMBF) determined from vertical profiles of virtual potential temperature flux. Updraft statistics identified the 1–5 percentile fastest updraft velocities; travel times Δt to LCL and cloud top were integrated from vertical profiles of percentile updraft velocities. Radiative transfer: RRTMG computed TOA shortwave and longwave fluxes; in the eclipse case, incident TOA SW was multiplied by (1 − f_s), where f_s is the area-mean obscuration at 0.635 µm. MODIS white-sky albedo and ERA5 ozone profiles were used. Cases: Three eclipses (2005–2016) were analyzed with corrected satellite data; detailed results are shown for the 3 October 2005 annular eclipse over East Africa with a defined 3–7°N, 27–31°E land study area.

• Corrected geostationary satellite observations show that over cooling land surfaces, shallow boundary-layer cumulus clouds began dissipating at low solar obscuration (~0.15) and were not evident without eclipse correction due to lunar shadow darkening.
• For the 3 Oct 2005 case, morning cloud cover increase halted around ~09:30 UTC at ~15% obscuration; clouds substantially diminished and only returned about ~50 minutes after maximum obscuration (max at 10:52 UTC), evidencing a delayed response.
• Mean COT of cloudy pixels decreased during the period of rapid cloud cover recovery (12:00–12:30 UTC), consistent with newly formed shallow clouds; such behavior was absent on comparable non-eclipse days.
• Over adjacent ocean, cumulus clouds did not disappear and SST showed no eclipse-related drop, attributed to large heat capacity and efficient mixing in the ocean.
• LST over land dropped rapidly with obscuration, with an estimated maximum eclipse-induced decrease of 5.8 K at ~11:00 UTC; no detectable LST time lag relative to mid-eclipse was observed at 15-minute sampling.
• DALES simulations driven by measured LST reproduced a substantial cloud cover decrease relative to a no-eclipse reference, with divergence evident ~15–20 minutes after eclipse onset when obscuration was <10%.
• Simulated updrafts and parcel travel times explain the delay: fastest parcel travel times from surface to cloud top/LCL were ~16–24 minutes around 09:00 UTC and ~13–19 minutes to LCL around 11:10 UTC, implying cloud response was initiated at even smaller obscurations than observed.
• Eclipse suppressed surface sensible and latent heat fluxes, reducing buoyancy flux, weakening updrafts, lowering the probability that parcels reach LCL, and thus diminishing shallow cumulus formation.
• Radiative feedback: clearing skies reduced reflected shortwave at TOA, partially offsetting the direct dimming. Neglecting eclipse-induced cloud disappearance led to an overestimation of the eclipse-related reduction of net incoming shortwave radiation at TOA by about 20 W m⁻² at 11:22 UTC.
• Surviving clouds grew in size and COT during the eclipse period; shallow cumulus returned across the study area only in the final stages of the event.

The study directly addresses how shallow cumulus clouds respond to transient reductions in solar radiation by correcting satellite retrievals for eclipse-induced irradiance changes and coupling them with process-resolving LES. It demonstrates that shallow cumulus over land are highly sensitive, beginning to dissipate at small obscuration fractions, with a delayed response governed by parcel ascent times within the boundary layer. The findings imply that short-term, spatially varying solar dimming—analogous to potential solar geoengineering implementations—can reduce shallow cumulus coverage, thereby diminishing reflected solar radiation and partially counteracting the intended reduction in net incoming shortwave flux. This could also inhibit growth into deeper convective, possibly precipitating clouds, with consequences for local precipitation patterns. The observed and modeled timing underscores the importance of surface-atmosphere coupling and boundary-layer turbulence in cloud sensitivity to insolation changes. These results provide an observationally constrained benchmark for evaluating cloud responses in models under scenarios of sunlight interception, where non-uniform, time-varying obscuration patterns (e.g., space-based sunshades or spatiotemporally variable stratospheric aerosols) could induce local obscurations well above the small thresholds found to impact clouds.

By applying an eclipse-aware correction to geostationary satellite cloud products and conducting LES constrained by observed surface temperatures, the study reveals that shallow cumulus over land dissipate rapidly starting at roughly 15% solar obscuration, with responses initiated at even smaller obscurations due to boundary-layer parcel ascent times. The eclipse-induced cloud disappearance materially alters TOA radiative fluxes, such that omitting this effect would overestimate the eclipse-related reduction in net incoming shortwave radiation by about 20 W m⁻² at peak impact. These insights highlight the rapid sensitivity of land-based shallow cumulus to modest, transient solar dimming and underscore the importance of explicitly representing such processes in assessments of solar geoengineering techniques. Future work should include targeted cloud model simulations across diverse land surface types, climates, and obscuration geometries/durations, and leverage the presented satellite-corrected eclipse datasets for model validation to improve predictions of cloud behavior under natural eclipses and engineered solar dimming.

• Temporal resolution: LST was available at 15-minute intervals, insufficient to resolve minute-scale lags (~1.5 min) reported elsewhere; thus, precise timing of LST minimum relative to obscuration may be unresolved.
• Retrieval artifacts: COT retrievals had missing observations between 09:00–10:00 UTC due to cloud bow scattering geometry; river sunglint required masking.
• Case selection: Detailed analysis centers on a single annular eclipse case over East Africa (with two additional examples provided in supplements); generalization across climates and seasons may require more cases.
• Modeling assumptions: DALES used horizontally homogeneous initial conditions, flat terrain, prescribed LST (with calibration factors for surface flux resistances), and neglected horizontal inflow/outflow of eclipse-induced mesoscale perturbations; these simplifications may affect representativeness of mesoscale dynamics.
• Discrepancies: Simulated post-eclipse increases in COT contrasted with observations, likely due to horizontally averaged input settings and absence of deeper convective clouds seen in the southern study area in reality; larger post-eclipse cloud cover vs. reference is difficult to confirm observationally due to afternoon variability.
• Ocean response: Lack of observed SST response is consistent with physics but limits inference about marine boundary-layer cloud responses under different oceanic conditions.
• Affiliation of results: The sensitivity described is strongest over land where surface temperature adjusts quickly; responses over heterogeneous surfaces or urban areas were not separately analyzed.