Prévision des précipitations par les modèles numériques de météorologie dynamique

Forecasting precipitation using dynamic meteorological numerical models

Ingénieur en Chef de la Météorologie, EERM-GMD

Abstract

Dynamic numerical meteorological models simulate atmospheric conditions by solving the equations for the trend of the different variables describing the atmosphere. Water is a major variable in the atmosphere; precipitation is one of the results yielded by these models. The forecast of precipitation involves the identification of large scale phenomena which can be explicitly processed by the model (with a resolution of the order of 150 x 150 km) and smaller scale phenomena which are below the calculation range (convective type rains). Explicit processing takes into account the water vapor transfer, condensation and the resulting heat discharge. A simple chart of precipitation, assuming a Marshall-Palmer-type distribution of rainfall, allows the volume of water reaching the ground to be calculated, taking account of the evaporation of raindrops in non-saturated layers. Convective effects are parametered in a chart which distributes thermal energy and humidity as a function of altitude when the model indicates the existence of cloud convention, following two criteria: conditionally unstable atmosphere and positive water vapor addition at the base by convergence and evaporation. Rainfall forecasts have been made for France over 25 basins for 12 hour stretches during the day, at night and the following day. A verification criterion: I=100ad - (b+c/2)2/(a+ b+c/2)(d+ b+c/2) has been defined, in which a, b, c and d are the coefficient of the contingency table of rain observed and forecast exceeding a certain threshold. The value 100 of the index corresponds to a perfect forecast, and the value 0 describes a meteorological forecast of no value. Quality deteriorates the further forward the forecasting and the greater the rate of precipitation. Nevertheless, the model is always helpful. In addition, it can be seen that summer rain is not forecast as well as winter rain, in particular for moderate and low rainfall. These criteria relating to direct forecasts from the model can be used as a basis for assessing the improvements which can possibly be obtained through dynamic or statistical forecasts through local adjustments.