Using Self-Organising Maps (SOMs) to assess synchronies: an application to historical eucalypt flowering records

Abstract

Self-Organising Map (SOM) clustering methods applied to the monthly and seasonal averaged flowering intensity records of eight Eucalypt species are shown to successfully quantify, visualise and model synchronisation of multivariate time series. The SOM algorithm converts complex, nonlinear relationships between high-dimensional data into simple networks and a map based on the most likely patterns in the multiplicity of time series that it trains. Monthly- and seasonal-based SOMs identified three synchronous species groups (clusters): E. camaldulensis, E. melliodora, E. polyanthemos; E. goniocalyx, E. microcarpa, E. macrorhyncha; and E. leucoxylon, E. tricarpa. The main factor in synchronisation (clustering) appears to be the season in which flowering commences. SOMs also identified the asynchronous relationship among the eight species. Hence, the likelihood of the production, or not, of hybrids between sympatric species is also identified. The SOM pattern-based correlation values mirror earlier synchrony statistics gleaned from Moran correlations obtained from the raw flowering records. Synchronisation of flowering is shown to be a complex mechanism that incorporates all the flowering characteristics: flowering duration, timing of peak flowering, of start and finishing of flowering, as well as possibly specific climate drivers for flowering. SOMs can accommodate for all this complexity and we advocate their use by phenologists and ecologists as a powerful, accessible and interpretable tool for visualisation and clustering of multivariate time series and for synchrony studies.

Wavelet Signatures of Climate and Flowering: Identification of Species Groupings

“…Indeed the members within each of these groupings were recently shown to flowering synchronously in a study using self organising map (SOM) methods for time series clustering . The resultant SOM correlations, which were based on clustering the patterns underlying the time series records in Hudson et al (2011a), also showed similar trends to the correlations gleaned from Moran based correlations obtained from the raw flowering records (see Kim et al, 2008). In agreement with the SOM clustering results, the wavelet based findings of this chapter (of the same eight species) showed via similarities in the wavelet correlation, wavelet crosscorrelation signatures and in the identification of the main species specific climatic drivers of flowering: Eucalyptus microcarpa to be asynchronising with E. leucoxylon and E. polyanthemos.…”

Visual Discovery of Synchronisation in Weather Data at Multiple Temporal Resolutions

“…Thus, discovering and analysing spatio-temporal weather patterns is essential for understanding the Earth system. One particularly interesting pattern is synchronisation, which refers to the degree of temporal similarity between two or more time series (Hudson et al, 2011). The identification of weather synchronous regions or years helps to understand the impact of climate change on living organisms.…”

“…The MTUP is, however, not new. Hudson et al (2011) identified differences in synchronized species at seasonal and monthly resolutions and Jong and Bruin (2012) illustrated possible impacts of operating at different temporal resolutions when analysing remotely-sensed vegetation indices.…”

Wavelet Signatures of Climate and Flowering: Identification of Species Groupings

“…Eucalyptus leucoxylon was demonstrated to be synchronous with both E. polyanthemos and E. tricarpa, and asynchronous to all remaining species. Hudson et al (2011a) demonstrated that the main influence for grouping or clustering (synchronisation of flowering) was the season in which flowering commences, however, other flowering characteristics such as the timing of peak flowering and start and cessation of flowering were also contributing factors to synchronisation. In the wavelets based WCORR and WCCORR analyses in this chapter timing of peak intensity was also shown to be an important factor, as reflected by the common seasonal change points for synchronous species groupings (Table 4).…”

“…Eight eucalypt species' median flowering duration and months in which the main phenophases occur. Terms describing flowering intensity and their assigned value (Keatley 1999) (Table sourced from Hudson et al, 2011a).…”