Extraction of variability trends via record-breaking statistics

Record-breaking statistics are employed to examine natural variability in time-series. Specifically, we propose an index, α, which is based on counts of record-breaking highs and lows and reversibility in time. This technique is useful for extracting weak variance trends in multiple time-series and is generally distribution independent, thus removing a common challenge in time-series analysis. Here the method is applied to globally distributed monthly temperature time-series. For 15635 monthly temperature time-series from different geographical locations (Global Historical Climatology Network), each time-series about a century-long, < α > = -1.0, indicating decreasing variability. This value is an order of magnitude greater than the 3-σ value of stationary simulations. Using the conventional best-fit Gaussian temperature distribution, the trend is associated with a change of about -0.2°C per 106 years in the standard deviation of interannual monthly mean temperature distributions (about 10%).