A moisture-weighted typical-month selection for environmental exposure screening: design and evaluation of CEY20 at 14 locations

Wadim Okley 

Typical-year datasets condense multi-year hourly records into one year for engineering calculations. Established methods were designed for energy and solar applications, whereas moisture and thermal exposure screening needs a typical year that preserves humid-hour statistics. This study presents CEY20, a typical-month selection on the global 0.25° ERA5-Land land grid for 2006–2025. For each point and calendar month it selects the source year whose ten monthly temperature and humidity features are closest to their 20-year means, with 71 % of the weight on moisture features. The released version smooths month interfaces with a 6-hour excess-step ramp. Target checks were specified before target outcomes at 14 locations from tropical to subarctic climates; hourly diagnostics were defined subsequently, before their computation. The released version was evaluated under a later amendment with unchanged definitions, paired with the unsmoothed construction. For the released version, the composite year is closer to the pooled 20-year joint temperature–humidity distribution than every individual source year at 9 of 14 locations. Of 196 humid-spell and drying-gap statistics, 195 fall within the range of the actual years, as do 194 of 196 screening outputs; both exceptions are closer to the pooled climate than every actual year. Under strict tolerances, it meets ±5 % on annual exposure hours in 33 of 42 checks, ±0.5 °C on monthly mean temperature in 109 of 168 checks and ±2 percentage points on monthly mean humidity in 153 of 168 checks, with median deviations of 0.31 °C and 0.61 percentage points. Deviations beyond tolerance concentrate at arid and cold continental locations. Under the primary reference, boundary steps above the 99th percentile of natural hourly changes fall from 68 to 0 of 150 cross-year boundaries for temperature, while the largest change within the 12-hour smoothing window still exceeds it at 14 of 150. On the same data, all unsmoothed, CEY20 meets more annual exposure and monthly humidity checks than two Finkelstein–Schafer selections; a temperature-oriented variant meets more monthly temperature checks. At California weather stations, the departure of the selected year from its own source climatology is smaller than the source–station discrepancy at 37–39 of 40 stations. Relative to the unsmoothed construction, the annual tolerance counts are unchanged, one monthly temperature check enters tolerance and one humidity check leaves it. Results describe in-sample agreement for the tested points and quantities from one reanalysis at 0.25°. Screening is the intended application, and application-specific suitability remains to be evaluated.