Rainfall nowcasting for landslides early warning systems: an integrated
modeling approach.
Abstract
Effectiveness of floods and landslide early warning systems can be
clearly improved by reliable quantitative predictions of rainfall, which
represents the main precursor. With this aim, a methodology for
probabilistic rainfall nowcasting, based on a coupling between a
stochastic model and outputs provided by a Numerical Weather Prediction
(NWP) model is proposed in this paper. The coupling among different
types of models usually allows for improving the prediction, as the
positive aspects of all the model components are merged. In this paper,
the hybrid model, named PRAISE-ME (Prediction of Rainfall Amount Inside
Storm Events with MEteo), is proposed. This model allows improving the
rainfall prediction at hydrological scales, where only NWP models are
not so suitable and the simple use of stochastic models provides the
same forecast, regardless of weather forecasts as they depend only on
previous rainfall. PRAISE-ME provides probabilistic quantitative
predictions and it can be easily set up as input in other models for
Rainfall-Runoff or Landslide prediction, as in the application case here
illustrated. In this work, PRAISE-ME was used with the empirical FLaIR
model (Forecasting of Landslides Induced by Rain-fall, Capparelli and
Versace 2011) in order to obtain in real-time indications about
exceedance probabilities associated to specific thresholds. The
procedure was applied for a landslide case study, occurred in Montenero
di Bisaccia (Central Italy) in March 2006. The obtained results
encourage the use of this methodology as a component of early warning
systems.