The author presents an iterative approach to describing a data set, such
as a paleoclimatic proxy or a model output, in terms of automata of
successively higher order. The automata reflect dynamics acting on
successively longer timescales and larger spatial scales. The method
uses the computed probability density function from reconstructed state
space portraits, over successive overlapping windows in time, of the
record of magnetic susceptibility of loess and paleosols at Luochuan,
central China. Areas of consistently high probability across several
time windows represent areas of quasistability, which are used as the
predictive and successor states of a succession of Markov Chains that
characterize the variability of the strength of the East Asian
paleomonsoon at different time scales. Seven metastable states are thus
identified, forming four Markov Chains, which show a marked increase in
complexity of behavior of the paleomonsoon system throughout the
Quaternary. A higher-order automaton is suggested by the sequence of
Markov Chains, suggesting differing cycles of dynamic behaviour in the
Early and Late Quaternary.