Temporal variations of trends in the Central England Temperature series

Abstract

Variations in trend rates of annual values of the Central England Temperature series (CET) over the period1659-2017 were analysed using moving windows of different length, to identify the minimum period in which the trend expresses a climate signal not hidden by the noise produced by natural variability. Trend rates exhibit high variability and irregular shifting from positive to negative values unless very long window lengths (of 100 years or more) are used. In general, as the duration of the length of the temporal window analysed increases, the absolute range of the trend rates decreases and the signal-to-noise (S/N) ratio increases. The relationship between the S/N ratio and the window length also depended on the total length of the series, so high S/N values are achieved faster when shorter time series are considered. This prevents suggesting a minimum window length for undertaking trend analyses.A comparison between CET and the average continental series in the Berkeley Earth Surface Temperature (BEST) database in their common period (1753-2017) repeats the patterns described for 1659-2017, although the average values of the rates, ranges and the "threshold period" in years change, and are more variable in CET than in BEST.Analysis of both series suggests that the recent warming started early and can be linked to the recovery of temperatures after the Little Ice Age. This process has characterised by progressively increasing trend rates, but also includes periods of deceleration or even negative trends spanning less than 50 years. The behaviour of the two long-term temperature records analysed agrees with a long-term persistence (LTP) process. We estimated the Hurst exponent of the CET series to be around 0.72 and 0.8, which reinforces the LTP hypothesis. This implies that the currently widespread statistical framework assuming a stationary, short-memory process in which departures from the norm can be easily assessed by monotonic trend analysis should not be accepted for long climatic series. In brief, relevant questions relative to the recent evolution of temperatures such as the distinction between natural variability and departures from stationarity; attribution of the causes of variability at different time scales; determination of the shortest window length to detect a trend; and other similar ones have still not been answered and may require adoption of an alternative analytical framework.