Skip to Main content Skip to Navigation

Introduction d'outils de l'intelligence artificielle dans la prévision de pluie par radar

Abstract : The work presented in this thesis aims at the development of a short-term radar rainfall forecasting system, that is adapted to use in urban hydrology. The system, named PROPHETIA, is based on rain cell tracking, in order to take individual advection into account. To forecast the rainfall at an instant tn, based on the series of radar images (I1…In), the algorithm of PROPHETIA contains four main parts: - identification and description of the rain cell echoes on image In - coupling of rain cells observed on the images I1…In, with the echoes defined on image In - characterization of the rain cells in the interval (t1, tn) - forecast of the rainfall by extrapolation of rain cell characteristics. A threshold technique is used for echo identification. The coupling algorithm is based on the application of a decision tree, that has been generated by machine learning from human defined examples. The high performance of this rule base is demonstrated by comparison with the human-defined couplings. In the first part of the study, the forecast is based on rain cell advection alone. The performance of the system is analyzed with respect to a new hydrological evaluation criteria, named TMP. It is shown that the error rate of forecasts by PROPHETIA is lower than that of other techniques, especially for convective rainfall events. The principal remaining forecast error is due to convective rain cell development during the forecast lead time. A model of this development is suggested, that allows for localizing the air masses alimenting the rain cells. Based on a statistical analysis of convective rain cells in terms of the model, a method is proposed, that takes rain cell growth/decay rates into account. It is shown, that a significant improvement of forecast quality can be achieved with this technique, if the development rates are known in advance. In the last part of the study, the possibilities of forecasting the rain cell development are explored. An examination of a sample set demonstrates the necessity of taking local variations of the surface conditions into account. Possible solutions to this problem are discussed.
Complete list of metadata

Cited literature [128 references]  Display  Hide  Download
Contributor : Ecole Des Ponts Paristech Connect in order to contact the contributor
Submitted on : Friday, September 24, 2010 - 12:57:13 PM
Last modification on : Monday, October 19, 2020 - 11:05:51 AM
Long-term archiving on: : Saturday, December 25, 2010 - 2:36:53 AM


Files produced by the author(s)


  • HAL Id : tel-00520834, version 1


Andreas Neumann. Introduction d'outils de l'intelligence artificielle dans la prévision de pluie par radar. Hydrologie. Ecole Nationale des Ponts et Chaussées, 1991. Français. ⟨tel-00520834⟩



Record views


Files downloads