Abstract:
Flood forecasting in small watersheds is a complex problem, given the stringent time scales to convey
accurate alerts in due time and small spatial scales for both atmospheric and water basin domain prediction.
The traditional forecast approach, based on a chain of numerical models for meteorological, hydrological and
hydraulic processes is not sufficient, requiring the integration with tailored, real-time data to produce accurate
inundation maps and provide timely warnings.
Herein, we present a new methodology for flash flood forecasting, based on a two-step procedure and on the
use of WIFF, a generic forecast framework applied successfully in estuarine and coastal flood forecasting. In
this methodology, WIFF executes two procedures in parallel. First, a large-scale approach, based on
conventional numerical models, running continuously every day, to detect significant rain events. If a predicted
rain event crosses a warning threshold, a second approach is triggered, involving a small-scale data-based
model to predict flooding for the following hours, based on real time monitoring networks data and on the use
of high performance computing for machine learning-based simulations. For the first step, we are updating the
WIFF framework to integrate both hydrological and hydraulic models of the HEC model family (Brunner,
2021).
This methodology is being validated in the Ribeira das Vinhas basin, an area prone to torrential floods that
inundate the urban area of the city of Cascais, located at the Tagus estuary mouth.
