Skip to Main content Skip to Navigation
Journal articles

Classification of rainfall radar images using the scattering transform

Abstract : The classification of rainfall fields has mainly focused on the split between convective and stratiform rainfall fields. In the present case study, the wavelet-based scattering transform is used to classify rainfall events observed by a weather radar. This very recent method has, to the best of the authors’ knowledge, not yet been applied for such a purpose. This method considers the spatial properties of rainfall radar images. This case study regroups 34 rainfall periods recorded over the Nantes region (western France) during 23 days in both 2009 and 2012. These periods display different characteristics in terms of duration and type of rainfall field. A reference configuration of the scattering transform has been evaluated and compared to various configurations in order to approximate the application conditions most appropriate to this case study. This evaluation is performed by a leave-one-out cross validation. A global accuracy of 93.5% of well classified images is obtained in the reference conditions which is an encouraging result. The temporal sampling of the rainfall fields is an important aspect of the classification process.
Document type :
Journal articles
Complete list of metadatas
Contributor : Mathieu Lagrange <>
Submitted on : Wednesday, November 16, 2016 - 11:28:30 AM
Last modification on : Tuesday, November 10, 2020 - 9:08:03 AM


  • HAL Id : hal-01397741, version 1


Mathieu Lagrange, Hervé Andrieu, Isabelle Emmanuel, Gerard Busquets. Classification of rainfall radar images using the scattering transform. Journal of Hydrology, Elsevier, 2016. ⟨hal-01397741⟩



Record views