Frequency-based representation of the slowest climate variability

These are the results of the experiment on frequency-based rotation of the slow climate variability reported in

• A. Ilin, H. Valpola and E. Oja. Exploratory Analysis of Climate Data Using Source Separation Methods.
  Neural Networks, Vol. 19, No. 2, pp. 155-167, March 2006.

The signals were extracted by first using the clarity-based analysis in order to find components with the most prominent variability in the slowest timescale. The linear temporal filter shown on the right was used for defining the timescale of interest. After that, the components were rotated using the frequency-based rotation in order to make their spectral contents as distinct as possible. This procedure is essentially equivalent to ICA (based on signal non-stationarities) in the frequency domain.
You can click on the maps to see larger images.

Time course

Surface temperature, °C

Sea level pressure, Pa

Precipitation, kg/m²