My map is better 😉

Agree!

I think you’d have to download the raw HadEX files directly. Berkeley Earth daily data may also be homogenized. In any case, since your plots reflect averages of large regions, they are probably less affected.

"Homogenization" issues become important when looking at trends in station observations over many decades. Datasets like GHCN-daily are not homogenized and may contain spurious trends. E.g. HadEX3 trends in annual minima/maxima have higher fidelity metoffice.gov.uk/hadobs/hade…

I think warm nights are the preferred metric here (Tmin > 20C/68F). Of course, ERA5 is a model product and does deviate from station observations (significantly, at times).

Yes this is a real issue for TXx in western Europe. Bias-adjusted models can still understate hot extremes, and products like ERA5 can underpredict station-observed TXx too.

I know these models change over time, but has anyone done a post-mortem on forecast skill at this lead time, particularly across the spring barrier, for recent events like '15/'16 and '23/'24? Curious how the June-initialized runs scored against the eventual peak.

Climate impacts research & practice should move to a GWL framework to not be as scenario dependent

Dunn et al. 2020,2024 (HadEX2/3) better handles the daily inhomogeneities. HadEX does show some regions with no warming of extreme T or declines in extreme T agupubs.onlinelibrary.wiley.…

There are statistically significant trends in large hail, but not in Wisconsin: nature.com/articles/s41561-0… And the presence of trend does not always mean "climate change"-caused. Climate change attribution is very challenging to do well (if it is even possible at all).

Dunn et al. 2020 also shows this and underpins ch 11 of AR6. agupubs.onlinelibrary.wiley.… See also Dunn et al. 2024

Mostly agree. But managing false positives matters too, and forecast skill for surface wind extremes is still much better than for flood impacts. That gets even harder when impacts depend on local factors like levees, drainage, and exposure.

Assumes stationarity in variance, which may be a poor assumption.

after LOESS detrending, the anomaly has about 0.02% annual chance of occurrence/exceedance (Weibull fit). 2022 anomaly was also remarkable.

I accounts for some aspects of inflation via the CPI, which measures out-of-pocket prices paid by consumers. This leaves out important factors like building labor and materials as well as the market value of structures.

I am waiting for someone to write one big paper comparing near-term CMIP projections with ERA5 trends to show discrepancies across many variables. The models just don't align at local scales in many places. This hugely impacts their utility as biz and insurance decision tools.

Makes sense. The recent uptick in Springer Nature's Discover "xyz" journal requests I get asking for reviews makes me wonder if publishers will take advantage of the increased volume

As GenAI erases the friction that once forced deep thinking, manuscript review becomes instant and effortless. We’ll publish ten times more papers, each with a tenth of the rigor. The academic literature risks becoming noise, not knowledge

Boggles my mind how earthquake scores don't appear on these sites despite the much greater risk

A few unlucky years and the cost of capital/labor has increased dramatically since '20. The NBER paper does not isolate any local climate change signal. As long as people keep buying policies at these prices, reinsurers will keep charging them.