Interdecadal surface temperature changes

Temperature changes can be decomposed into slow interdecadal components and faster intradecadal components. The slowest components of temperature changes have the largest ratio R of interdecadal variance to intradecadal variance; faster components of temperature changes have a smaller ratio R of interdecadal variance to intradecadal variance. By filtering out components with smaller variance ratios R, we can isolate the slow component of temperature changes.

The following animations show the slow component of changes since 1890 in annual mean temperatures and in seasonal mean temperatures (for the DJF, MAM, JJA, and SON seasons). The upper panel of the animations shows the slow component of temperature changes between 22.5°S and 67.5°N, an area with sufficient data coverage since 1890 to allow this multivariate analysis. The indicated temperature changes are changes relative to the period 1890–1909. The lower panel shows the time series of area-mean temperature changes (black) and the area-mean temperature change accounted for by the slow component in the main panel (red). Gray areas in the upper panel are areas with insufficient data coverage for this analysis. Each file (Quicktime format) is about 2 MB in size.

[Annual mean]  [DJF mean]  [MAM mean]  [JJA mean]  [SON mean] 

Several of the temperature changes are suggestive of human influences on climate. For example, the relatively uniform and steady warming of the ocean surfaces, the generally enhanced warming of continents, and the strong warming of high northern latitudes, particularly in the transition seasons, is consistent with expected effects of increases in greenhouse gas concentrations. The localized cooling between about 1950 and 1970 over industrial regions such as Europe and Southeast Asia, where anthropogenic sulphate aerosol loadings were high, is consistent with the expected cooling effect of sulphate aerosols. Also recognizable are numerous apparently natural climate variations, for example in the North Atlantic and central Pacific.

The animations are produced using the methods described in Schneider and Held (2001). As in the paper, the data are from the Climatic Research Unit at the University of East Anglia (dataset HadCRUT3v). The analysis in the paper has been extended to the annual mean and to all seasons, including data through 2006. The overlapping decadal data groups used to determine the slow temperature variance are defined similarly to the analysis in the paper but are centered on the years 1896, …, 2001, with 15 years between successive group centers. The temperature changes represented in the animations are those accounted for by all discriminants with a variance ratio R greater than one. There are 3 or 4 such discriminants in each animations (i.e., the temperature changes in the animations have 3 or 4 spatial degrees of freedom).