Temperature variability and early clustering of record-breaking events

Abstract

As the number of climatological studies using record-breaking statistics is growing rapidly, understanding the sensitivity of the chosen time period becomes essential. To that end, here we examine the evolving variability of monthly mean temperatures and its dependence on beginning and final year. Specifically, we use an index, α, based on record-breaking statistics and employing reversibility such that 〈α〉=0 indicates no trend in variability. Generally, 〈α〉 has decreased between 1900 and 2013, indicating decreasing variability relative to early decades for stations from the contiguous USA (United States Historical Climatology Network, version 2.5). We find, somewhat surprisingly, that the observed decrease is due to an early excess of records beginning in 1917 (record low value) and 1921 (record high value). While detailed results depend on whether the data is gridded, detrended, etc., the general finding appears remarkably robust and holds globally as well.