Analysis of the Relationship between Banded Orographic Convection and Atmospheric Properties Using Factorial Discriminant Analysis and Neural Networks

• Published in: Journal of applied meteorology and climatology Vol. 49; no. 4; pp. 646 - 663
• Main Authors: Godart, A., Leblois, E., Anquetin, S., Freychet, N.
• Format: Journal Article
• Language: English
• Published: Boston, MA American Meteorological Society 01.04.2010
• Subjects: Analysis; Archives & records; Artificial neural networks; Atmospheric moisture; Atmospheric sciences; Atmospheric sounding; Atmospherics; Classification; Convection; Discriminant analysis; Discriminants; Earth, ocean, space; Environmental Sciences; Exact sciences and technology; External geophysics; Factorial analysis; Gauges; Geophysics. Techniques, methods, instrumentation and models; Hydrology; Mathematical models; Meteorology; Moisture effects; Neural networks; Neurons; Numerical simulations; Potential temperature; Precipitation; Radar; Radar data; Rain; Rain gauges; Rainfall; Satellite data; Soundings; Statistical methods; Studies; Water in the atmosphere (humidity, clouds, evaporation, precipitation); Wind shear; French Central Massif; France; Cevennes; Europe; Mediterranean Basin; Western Europe; Massif Central; factor analysis; rainfall; discriminant analysis; Orographic cloud; atmospheric precipitation; neural networks; Observation data; Convective cloud; Atmospheric sounding; statistical analysis; Orographic effects; Précipitation atmosphérique; Neural networks; Statistique; Convection
• ISSN: 1558-8424; 1558-8432
• DOI: 10.1175/2009JAMC2217.1

The relationship between banded orographic convection and atmospheric properties is investigated for a region in the south of France where the associated rainfall events are thought to represent a significant portion of the hydrologic input. The purpose is to develop a method capable of producing an extensive database of banded orographic convection rainfall events from atmospheric sounding data for this region where insufficient rain gauge data and little or no suitable radar or satellite data are available. Two statistical methods—discriminant factorial analysis (DFA) and neural networks (NNs)—are used to determine 16 so-called elaborated nonlinear variables that best identify rainfall events related to banded orographic convection from atmospheric soundings. The approach takes rainfall information into account indirectly because it ‘‘learns’’ from the results of a previous study that explored meteorological and available rainfall databases, even if incomplete. The new variables include wind shear, low-level moisture fluxes, and gradients of the potential temperature in the lower layers of the atmosphere, and they were used to create an extensive database of banded orographic convection events from the archive of atmospheric soundings. Results of numerical simulations using the nonhydrostatic mesoscale (Méso-NH) meteorological model validate this approach and offer interesting perspectives for the understanding of the physical processes associated with banded orographic convection. DFA proves to be useful to determine the most discriminant factors with a physical meaning. Neural networks provide better results, but they do not allow for physical interpretation. The best solution is therefore to use the two methods together.