Modelli di preannuncio delle piene in piccoli bacini ed incertezze legate alla densita' della rete pluviometrica.

Small catchments with fast hydrological responding geomorphology are commonly prone to Flash Floods, which have to be predicted on the basis of meteorological forecast and radar nowcasting, rather than continuously monitoring river water levels. Moreover, real-time rainfall observations at high resolution in time and space are required to understand the watershed processes involved. However small catchments are generally scarcely instrumented and only a sparse rain gauge network is available. This work analyze different flood forecasting techniques and investigates their applicability on some small catchments located in the east side of southern central Sardinia, where only a small number of rain gauges is available. The first part of the work investigates the sensitivity and accuracy of hydrological processes simulation of flooding events, using rain gauge networks of different density with a statistical approach, by means of a long synthetic rainfall dataset simulated at high resolution in time and space. Analyses on a set of 12 basins of different sizes, ranging from 15 up to 1800 km2, are performed on the hydrological response of two simplified rainfall-runoff models: a lumped and a distributed model. Results highlight a strong dependence of model performance with the event severity, and show that even in very small basins and regardless of the model approach implemented, to guarantee satisfying hydrological simulations, more rain gauges than those generally available are required. A final comparison with a real case, although restricted to very few rain gauges, seems to confirm the outcomes of the synthetic approach. In the second part of the work, different Flash Flood forecasting techniques are tested on two small basins (sizing 121 and 53 km2) provided with long observations at the rain gauges and hydrometric stations. These techniques are classified in two main forecasting approaches: RTCM (Rainfall Thresholds based on Conceptual Models) e RFTDM (Runoff and Frequency Thresholds based on Distributed Modelling). RTCM are strictly deterministic and provide rainfall thresholds for the entire basin through simple operational curves, obtained by applying different event-based lumped models in inverse mode, and taking into account only initial soil moisture content and event duration. These techniques are operationally easy and could be rapidly transferred to other catchments. Meanwhile they provide fairly good forecasting performances when base flow is rather low, even if an high false alarm rate is usually exhibited. RFTDM rely on a physically based distributed model which simulates continuously all hydrological basin processes: in this work tRIBS (TIN based Real Time Integrated Basin Simulator) is applied for its efficiency and computational speed. In particular two different approaches are proposed: Direct method, in which the model is part of a forecasting chain running continuously in real-time and simulating directly maximum floods on the basis of meteorological forecasts; Statistic method, through flood frequency analyses (FFA) on observed and simulated discharges, provides probabilistic flood predictions comparing occurrence frequencies rather than discharges. Results highlight a significant reduction of false alarms with respect to the RTCM, preserving good prediction skills in different operational conditions. Meanwhile, compared with the direct method, the expected forecasting improvement using statistic method has not detected, regardless of the probability distribution chosen for FFA. It’s important to note that Statistic method, notwithstanding the laborious setting, allows to creating alert maps for flash flooding. Outcomes suggests to pay particular attention when using alert maps produced by few rain gauges due to distortions induced through rainfall field sampling.