Application of the singular spectrum analysis technique to study the recent hiatus on the global surface temperature record

Abstract

Global surface temperature has been increasing since the beginning of the 20th century but with a highly variable warming rate, and the alternation of rapid warming periods with 'hiatus' decades is a constant throughout the series. The superimposition of a secular warming trend with natural multidecadal variability is the most accepted explanation for such a pattern. Since the start of the 21st century, the surface global mean temperature has not risen at the same rate as the top-of-atmosphere radiative energy input or greenhouse gas emissions, provoking scientific and social interest in determining the causes of this apparent discrepancy. Multidecadal natural variability is the most commonly proposed cause for the present hiatus period. Here, we analyze the HadCRUT4 surface temperature database with spectral techniques to separate a multidecadal oscillation (MDV) from a secular trend (ST). Both signals combined account for nearly 88% of the total variability of the temperature series showing the main acceleration/deceleration periods already described elsewhere. Three stalling periods with very little warming could be found within the series, from 1878 to 1907, from 1945 to 1969 and from 2001 to the end of the series, all of them coincided with a cooling phase of the MDV. Henceforth, MDV seems to be the main cause of the different hiatus periods shown by the global surface temperature records. However, and contrary to the two previous events, during the current hiatus period, the ST shows a strong fluctuation on the warming rate, with a large acceleration (0.0085°C year(-1) to 0.017°C year(-1)) during 1992-2001 and a sharp deceleration (0.017°C year(-1) to 0.003°C year(-1)) from 2002 onwards. This is the first time in the observational record that the ST shows such variability, so determining the causes and consequences of this change of behavior needs to be addressed by the scientific community.

Figure 1. SSA reconstructed signals from HadCRUT4 global surface temperature anomalies.

The annual surface temperature (gray line), multidecadal variability (MDV, blue line), secular trend (ST, red line) and reconstructed signal (MDV+ST, black line) are indicated. ST represents 78.8% of the total energy of the series; MDV accounts for 8.8% of the energy and the reconstructed signal for 88%. The dashed thin red lines indicate the range of variability of the ST obtained by applying SSA to the temperature time series obtained for each individual month.

Figure 2. Singular spectrum analysis of the GMTA.

a) Energy diagram of the first 20 eigenvectors (EV) obtained from the SSA. The horizontal dashed line indicates the 1% energy limit. b) Residuals obtained by removing the signals associated with the first four EV from the original GMTA record. The black bold line represents the linear fit (not significant, p>0.05). c) Autocorrelation analysis of the residuals. The horizontal dashed lines indicate significant bounds.

Figure 3. Global warming rate analysis.

a) Warming rates (°C year⁻¹) obtained from the different signals identified in the SSA: ST (red line), MDV (blue line) and reconstructed signal (black line). The dashed thin red lines are the confidence intervals for the warming rate associated with the ST obtained from each individual month’s time series. b) Zoom on the last 25 years of the time series.

Figure 4. SSA reconstructed signals for Northern Hemisphere surface temperature.

a) HadCRUT4 annual surface temperature (gray line), multidecadal variability (MDV, blue line), secular trend (ST, red line) and reconstructed signal (MDV+ST, black line). The dashed thin red lines indicate the range of variability of the ST obtained by applying SSA to the temperature time series obtained for each individual month. b) Warming rates (°C year⁻¹) obtained from the different signals identified in the SSA for the Northern Hemisphere. The dashed thin red lines are the confidence intervals for the warming rate associated with the ST obtained from each individual month’s time series. c) Zoom on the last 25 years of the warming rate time series.

Figure 5. SSA reconstructed signals for Southern Hemisphere surface temperature.

a) HadCRUT4 annual surface temperature (gray line), multidecadal variability (MDV, blue line), secular trend (ST, red line) and reconstructed signal (MDV+ST, black line). The dashed thin red lines indicate the range of variability of the ST obtained by applying SSA to the temperature time series obtained for each individual month. b) Warming rates (°C year⁻¹) obtained from the different signals identified in the SSA for the Southern Hemisphere. The dashed thin red lines are the confidence intervals for the warming rate associated with the ST obtained from each individual month’s time series. c) Zoom on the last 25 years of the warming rate time series.