Is it possible to fit extreme climate change indices together seamlessly in the era of accelerated warming?

Abstract This study examines the problematic impact of selecting a different base period (colder 1961–1990 vs. warmer 1988–2017), on the trend magnitude of widely used percentile‐based extreme temperature indices (e.g., warm/cold spells, warm/cold days and nights). The percentile‐based indices are part of a core set of indices (27 in total) that have become a common standard for monitoring climate change, as recommended by the Expert Team on Climate Change Detection and Indices (ETCCDI). The indices were designed to be comparable across regions provided that similar analyses are employed. Unfortunately, the use of different base periods and periods of interest to explore local and global climate change undermines the comparability of findings across regions. When utilizing “day‐count” indices with fixed thresholds, the use of different base/reference periods changes the intercept without influencing the slope (for a given period length). However, this assertion does not hold with percentile‐based indices. Our analyses show that percentile‐based temperature indices (e.g., days with temperature below the 10 th or above the 90 th percentiles) are particularly susceptible to the problematic use of different base periods. Hence, using percentile‐based indices may have adverse effects on researchers' conclusions. The current paper reports the results of a comparative study that used different base periods for the most commonly used percentile‐based extreme temperature indices. It was found that the (negative) trend magnitude of the cold percentile‐based indices (frequency of cold days and nights and cold spells) is strongly amplified while the (positive) trend magnitude of the warm indices (frequency of warm days and nights and warm spells) is dramatically diminished when percentiles were derived from a base period that included records from the last two decades (e.g., 1981–2010, 1988–2017). These features are even more pronounced when the study period covers only the last 30–40 years.