What the sky is doing, and what it is not

Implications

The reconstruction says the western Mediterranean has left Little Ice Age cloudiness and is now as clear as, or clearer than, the medieval warm period — for different reasons, with a different energy budget, and with people in the way.

Radiation

Dropping from a LIA mean of 49.1% to a 1990–2022 mean of 36.8% is on the order of several W m⁻² extra sunlight at the surface. That is not a rounding error next to greenhouse forcing. Clearer skies are an amplifier of the heat the basin is already famous for.

Drought loop

PDSI is in the model because wet soils make clouds. The reverse is a trap: dry soils → fewer clouds → more sun → more evaporative demand → drier soils. Headlines that recent western-Med droughts rank among the worst in 500 years are not a clean climate comparison. They set a pre-modern catchment, where rivers still reached their floodplains, against land after dams, inter-basin transfers, groundwater mining, and wetland drainage moved the water. A tree-ring scPDSI can still say the rain-minus-evaporation balance is extreme. It cannot say the managed landscape is the same kind of drought a 16th-century basin lived through. The cloud loop uses the climate balance, not the re-plumbed map.

Atlantic hinge

A positive AMO is a clearer Mediterranean in this reconstruction. That sign is only as good as the index. Mann’s series runs warmer than a detrended AMO in the late 20th century, which pushes the modern sky clearer than the instrumental Atlantic index would. Switching the buttons does not restore the Little Ice Age; it lifts the recent decades by about one to two percentage points of cloud cover.

Double-counted AMOLeft inThis atlas copies Mann’s series into soil moisture and temperature as well as the AMO term. Right removes those copies.Detrended index, from 1880MannLeft keeps Mann’s values, which retain global warming. Right uses the detrended ERSST AMO and drops the +0.34 °C splice.Hand-placed millennial shapeAs drawnA centennial nudge, about ±2 points, was added so the curve would match the paper’s rise to 1600 and modern low. Left takes that nudge off.Imposed PDSI drying after 1970Full rampThe late-century soil-moisture crash is a prescribed envelope, not the drought atlas. It also sits on land whose rivers were re-routed. Left turns the ramp off.AMO inside its 2σ bandArchiveThe archived uncertainty covers zero in most years. This slides the index from the low edge to the high edge. The sky barely moves — the band is wide, the cloud weight is not.

1990–2022 mean 36.9% · +0.1 points vs the curve as used

Medieval skies are not a precedent for comfort

MCA mean TCC is 43.2%; MWE is 43.4%; the last three decades are 36.8%. The paper’s claim that modern cloud patterns have come back toward the pre-1250 medieval state is directionally right and easy to over-read. Medieval clarity rode a warm Atlantic and a relatively dry land surface under a strong Sun, in a world with pre-industrial CO₂. Modern clarity rides those natural pieces plus greenhouse warming, aerosol cleanup after the late-20th-century “global dimming,” and a land surface that is drier in the climate indices. Those indices are not a before-and-after of the same ground: the modern basin is one where people re-routed the water. Same cloud fraction, not the same climate, and not the same hydrology.

Tambora was a hinge, not a cause of the modern sky

1815 is the date the authors flag, with Samalas-class ice-core sulfate, a following positive AMO, and rising solar. Superposed-epoch analysis (Figures, Added E) shows eruptions add cloud for a few years. The long decline after ~1600, and especially after ~1850, is AMO + solar + temperature + drying, not a two-century volcanic winter. Using Tambora as a narrative full stop is rhetorically clean and mechanistically incomplete — which the paper’s own multi-factor sentence already admits.

Who should care

Hydrology and agriculture. Fewer clouds, more PET, less summer convection: irrigation demand and fire weather, not just mean rainfall. Irrigation demand is partly circular here — a large share of summer water use is water that was stored or transferred, so the “dry land” the model sees is not the field the canal reaches.
CMIP cloud bias. Mediterranean models disagree on low-cloud feedback. A millennial observational constraint is useful if — and only if — the reconstruction’s AMO and PDSI levers are the real ones.
Attribution of heat extremes. 2022–2023 western-Med heat is usually framed as greenhouse plus soil drought. Add a millennial-low cloud shield and the energy budget is harsher.
Solar geoengineering debates. This record is not an analog for SAI, but it is a reminder that Mediterranean cloud cover is not a fixed climatology.

Limitations the paper underplays

No millennial cloud proxy. TCC before 1935 is entirely inferred. The Carcaixent diary is a 40-year sky log, not a thousand-year one. Tree rings see drought, not oktas.
The 500-year drought ranking mixes two landscapes. Tree-ring scPDSI ranks climate moisture. Gauged rivers, reservoirs, and water tables rank what is left after the Ebro, Guadalquivir, Tagus–Segura and Maghreb systems moved water from one basin to another. Calling both “the worst drought in 500 years” compares land before and after that re-routing. The post-1970 drying in this atlas is a prescribed envelope on top of that ambiguity, not a measurement that separated climate from canals.
PDSI already contains temperature. Feeding PDSI and GTA into one regression double-counts warmth. Some of the “soil moisture” lever is just heat in disguise.
NAO is missing. The paper’s introduction names NAO, PDO, ENSO. Eq. 2 does not. For the western Mediterranean, winter NAO is the textbook rainfall mode. Omitting it means AMO may be standing in for a bundle of Atlantic circulation.
GTA 25th percentile is opaque. A cold-tail statistic is not a standard cloud predictor. Without a physical derivation it reads as a term that helped the fit.
Uncertainty is a moving SD. That tracks decadal jitter, not proxy error, not coefficient error, not CRU inhomogeneity. The envelope around Fig. 6H is too tight.
Calibration on the decline. 1935–2018 is exactly the period of documented Mediterranean brightening. A model trained there will reconstruct a LIA that is simply “the opposite of now.” Independent documentary cloud series (Spain 1886–1960 increase then decrease; Lisbon; western Italy 1951–2018) support the sign of the 20th century, not the medieval amplitude.
Aerosols of the Industrial Revolution are in the paintings (Monet, Turner) and not in Eq. 2. European sulfur brightened clouds in the mid-20th century (global dimming) then faded (brightening). Some of the modern TCC drop is aerosol cleanup, which a millennial AMO–PDSI model will misattribute.

What we would do next

Put NAO (Trouet / Ortega reconstructions) in a nested model and test whether AMO still survives.
Replace GTA₂₅ with Mediterranean SAT and with 500-hPa geopotential, separately.
Use OWDA grid PDSI as the actual predictor, not a regionally invented cousin.
Verify against ICOADS ship cloud cover, Spanish/Portuguese observer series, and CLARA-A3 independently of CRU TS.
Run the same Eq. 2 on CMIP6 last-millennium and historical members — a true out-of-sample test.
Publish the annual series and code. A millennial reconstruction without a CSV is a figure, not a dataset.

Bottom line, which is also the authors’: Atlantic warmth and a dry land surface clear the western Mediterranean sky, and the modern combination of those two with a bright Sun is outside Little Ice Age experience. The numbers in this atlas recreate that argument with the published equation and the public AMO and solar series. They should be read as a working reconstruction, not as the last word on how many oktas hung over Córdoba in 1085.