Data and code for: Narrowing of the daily temperature distribution biases threshold day estimates in the High Arctic (four Svalbard stations, 1979 to 2025)

Observational series, derived results, analysis code and figures supporting the study "Narrowing of the daily temperature distribution biases threshold day estimates in the High Arctic: evidence from four Svalbard stations, 1979 to 2025". Threshold day counts, such as the number of days above 0 degrees Celsius, are commonly estimated by displacing an observed daily temperature distribution by a projected change in the mean. That procedure is exact only if warming acts as a pure translation. Using daily records from four Norwegian Meteorological Institute stations at Svalbard over 1979 to 2025, this study shows that it does not. The fifth percentile of daily mean temperature warmed by 5.42 degrees Celsius against 1.38 at the ninety fifth, and the daily standard deviation fell by 16.6 per cent. Threshold counts obtained by translation are correspondingly biased, warm day counts being overestimated by 12.0 to 41.6 days per year at every station. The effect is confined to the cold season, and in the two records that reach back to 1922 and 1946 it emerges only after about 1980. Of six candidate corrections tested on withheld years, linear quantile regression more than halves the error of current practice, while a gradient boosted quantile model is indistinguishable from a trend extrapolated pure shift. CONTENTScode/ 21 Python scripts covering download, station screening, threshold indices, the pure shift decomposition, percentile shifts, bootstrap intervals, homogeneity screening, the long record analysis, all six candidate models and every figure.data/ the five daily station series as retrieved from the Frost API, plus all derived indices, decomposition results, percentile shifts, bootstrap intervals, homogeneity tests, model comparisons and the complete set of values printed in Tables 1 to 9.figures/the seven manuscript figures at 600 dpi, two graphical abstract versions and a plain language poster. Everything reported in the paper can be regenerated from this record. See README.md for the run order and requirements. The underlying daily observations are from the Frost application programming interface of the Norwegian Meteorological Institute (https://frost.met.no) and are made available by that institute under the Norwegian Licence for Open Government Data and Creative Commons Attribution 4.0.