Measured meteorological time series are frequently used to obtain information about climate dynamics. We use time series analysis and nonlinear system identification methods in order to assess outdoor-environment bioclimatic conditions starting from the analysis of long historical meteorological data records. We investigate and model the stochastic and deterministic properties of 117 years (1891-2007) of monthly measurements of air temperature, precipitation and sunshine duration by separating their slow and fast components of the dynamics. In particular, we reconstruct the trend behaviour at long terms by modelling its dynamics via a phase space dynamical systems approach. The long-term reconstruction method reveals that an underlying dynamical system would drive the trend behaviour of the meteorological variables and in turn of the calculated Universal Thermal Climatic Index (UTCI), as representative of bioclimatic conditions. At longer terms, the system would slowly be attracted to a limit cycle characterized by 50-60 years cycle fluctuations that is reminiscent of the Atlantic Multidecadal Oscillation (AMO). Because of lack of information about long historical wind speed data we performed a sensitivity analysis of the UTCI to three constant wind speed scenarios (i.e. 0.5, 1 and 5 m/s). This methodology may be transferred to model bioclimatic conditions of nearby regions lacking of measured data but experiencing similar climatic conditions.
Keywords: AMO; Machine learning; Outdoor environment; Time-series; UTCI.