Comparison of two methods to forecast a low predictable heavy rainfall case
Prediction of extreme weather events is one of the most difficult tasks attributed to operational meteorological centres. These last two decades, improvement of model physics associated with a continuous decrease of horizontal resolution and advanced initialization methods have helped forecasters to predict local intense weather phenomena such as heavy rainfall or high winds. However, deterministic numerical weather prediction is limited by the chaotic nature of the atmosphere and small errors in model parameterizations or initial conditions may lead to substantial forecast failures, especially at the mesoscale.
In this study, two methods are compared to predict a cyclone development and its associated precipitation field in the Western Mediterranean area. In the first method, a small Meso-NH ensemble is built by using a date shifting initialization technique wherein members differ by their lateral boundaries conditions. Half of the members use operational ARPEGE analysis whereas others are integrated using ECMWF analysis. The second method is based on assessment of upper-level dynamics, in initial conditions and model outputs, by comparison of potential vorticity (PV) fields, METEOSAT water vapour (WV) images and pseudo water vapour (PWV) images derived from a radiative transfert code. In case of mismatch between observations and numerical outputs, local PV corrections may be applied to model fields so as to obtain atmospheric states used to integrate new runs.
It is found that the predictability of the investigated case is low according to rainfall and surface pressure prediction. However, the combined use of PV diagnosis and WV imagery has proven its utility by improving the forecast of accumulated precipitation, cyclone development and main cloud systems.
HyMeX – Hydrological cycle in the Mediterranean Experiment 2010-2020