ERA5-Land climate pattern detection pipeline: requesting methodology critique
I have built a pattern detection pipeline analyzing 75 years of ERA5-Land reanalysis data across four regions: Western Europe, Central US, Costa Rica, and New Zealand. Phase 1 detects geographic patterns in temperature and precipitation trends from 1950 to present. The approach is deliberately non-causal. The goal is to show what 75 years of verified observational data says at zone resolution and let the findings drive the questions, not the other way around.
Key methodology decisions I would welcome scrutiny on:
Temperature derived from t2m via annual resample max and min rather than mx2t24. Precipitation converted from tp in meters to mm annually. Trends calculated per grid cell using linear regression across three time windows: 1950 present, 1970 present, 1990 present. Statistical significance filtered at p less than 0.05. Precipitation box plots use significant cells only because median collapses to zero across all bands when all cells included.
Cross-regional finding worth flagging: nighttime low temperatures warming faster than daytime highs across all four regions, both hemispheres, all elevations, all ocean distances. Consistent enough that it points toward something operating above regional scale. Cloud cover as a mechanism is on the Phase 2 list but not yet tested.
Specific questions: Is the t2m resample approach for tmax and tmin defensible or is there a better ERA5-Land variable I should be using? Are there known ERA5-Land artifacts in the 1950-1970 period that would bias long period trends? Any other methodological red flags?