Data-driven clustering of rain events: microphysics information derived from macro-scale observations

Djallel Dilmi (1) , Cécile Mallet (1) , Laurent Barthès (1) , Aymeric Chazottes (1)

1 SPACE - LATMOS

The study of rain time series records is mainly carried out using rainfall rate or rain accumulation parameters estimated on a fixed duration (typically 1 min, 1 hour or 1 day). In this paper we used the concept of rain event. Among the numerous existing variables dedicated to the characterisation of rain events, the first part of this paper aims to obtain a parsimonious characterisation of these events using a minimal set of variables. In this context an algorithm based on Genetic Algorithm (GA) and Self Organising Maps (SOM) is proposed. The use of SOM is justify by the fact that it allows to maps a high dimensional data space to a two dimensional space while preserving as much as possible the initial space topology in an unsupervised way. The obtained 2D maps allow to provide the dependencies between variables and consequently to remove redundant variables leading to a minimal subset of variables. The ability of the obtained 2D map to deduce all events characteristics from only five features (the event duration, the rain rate peak, the rain event depth, the event rain rate standard deviation and the absolute rain rate variation of order 0.5) is verified. From this minimal subset of variables hierarchical cluster analysis were conducted. We show that a clustering in two classes allows finding the classic convective and stratiform classes while a classification in five classes allows refining this convective/stratiform classification. Finally, the last objective of this paper is to study the possible relationship between these five classes and their associated rain event microphysics. Some relationship between these classes and microphysics parameters are highlighted.

Domains: Meteorology; Ocean, Atmosphere

Djallel Dilmi, Cécile Mallet, Laurent Barthès, Aymeric Chazottes. Data-driven clustering of rain events: microphysics information derived from macro-scale observations. Atmospheric Measurement Techniques, 2017, 10 (4), pp.1557-1574. ⟨10.5194/amt-10-1557-2017⟩. ⟨insu-01410090⟩