Find the EGU on

Tag your tweets with #EGU17

SC6/NH10.1

How to apply and interpret the Fast Fourier Transform (FFT) for Time-Series Analysis (co-organized)
Convener: Robin Crockett 
Mon, 24 Apr, 19:00–20:00 / Room N2
Add this session to your Personal programme

Many natural hazards and other phenomena have cyclic/periodic behaviour, e.g. radon and soil gases, earthquakes (under some circumstances), annual/seasonal droughts and floods, whereas others are anomalous, recurring apparently randomly with regard to time, e.g. earthquakes and volcanic eruptions (under most circumstances). For the cyclic cases, an analysis of past time-series can yield an expectation and perhaps some degree of forecasting, even if only at the level of 'more probable' and 'less probable' times of occurrence.

The aim of this two-part short-course is to provide an introduction to and overview of the Fast (Discrete) Fourier Transform as a key technique of time-series analysis to identify and quantify periodic features, as distinct from its more conventional usage in digitisation and signal-analysis. It will take the key question β€œAre we looking for cyclic, recurrent or anomalous phenomena?” as its starting point.

In the first part, the aim will be to provide an introduction to and overview of the Fast (Discrete) Fourier Transform. The focus will be on the application of the Fast Fourier Transform, as implemented in many software packages, and interpretation of the output, including assessment of statistical effect-size. In the second part, the aim will be to consider techniques which are based on and/or allied to the Fast (Discrete) Fourier Transform, such as spectrograms (e.g. for varying frequencies) and and auto- and cross- correlation techniques (e.g. for recurrent rather than periodic features). Anticipating that most of those who attend will be 'technique users' rather than 'technique developers', coverage of the underlying mathematics will be kept to a necessary minimum to facilitate informed interpretation of the techniques considered.