Classification of structures from atmospheric time series

Time series are a major source of information about atmospheric boundary layer characteristics. Under specific situations, such as in convective or canopy turbulence, time series are frequently characterized by coherent structures of mostly known origin. These include ramp-cliff convective patterns and inflection-point instability above canopies, and several techniques have been developed for their detection and understanding. Stable boundary layers are also characterized by a myriad of structures, but their characteristics and generating mechanisms are generally unknown. Common approaches for detecting coherent structures are of limited use under these conditions. Still, the importance of understanding the stable boundary layer structures cannot be overstated, considering the inadequate treatment of stable conditions in numerical models.

We present a new technique for detecting and classifying structures from time series. The main idea of the technique is in defining structures as time-series subsequences that are significantly different from noise. This switches the focus of the approach towards defining the characteristics of noise, which is in many situations an easier problem than defining a structure. For atmospheric time series, a natural characterization of the noise is red noise, which is a stationary AR(1) process. The method consists of two steps. In the first step, the structures are detected as non-noise time-series subsequences. In the second step, the detected structures are classified in groups with similar characteristics using hierarchical clustering. We show that the method successfully recognizes and clusters the usual coherent structures, and that it also classifies unknown structures in stable conditions in a way that is useful for their further study.