1. A SEMI-EMPIRICAL APPROACH FOR EXTREME SEA-LEVEL EARLY WARNING FORECASTS ALONG THE ITALIAN COASTS A. Bonaduce (1), N. Pinardi (2), G. Coppini (1), G. Nardone (3), F. Lalli (3) (1)Euro-Mediterranean Center for Climate Change (CMCC), Italy (2) University of Bologna, Environmental Science, Italy (3) Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) 13th INTERNATIONAL WORKSHOP ON WAVE HINDCASTING AND 4th COASTAL HAZARDS SYMPOSIUM, 27 October – 1 November 2013, Banff, Alberta, Canada

2. Motivation Sea-level variability is characterized by multiple interacting factors that act over wide spectra of temporal and spatial scales. Locally lateral mass fluxes contribute to the mass budget and to the changes of mean sea level of the relevant coastal region (Pinardi et al., 2013, in-press). higher frequency motion, such as tides and wind waves, contribute to the increase of sea level and generate extreme sea level changes. Such extreme sea level events can generate coastal flooding (Nichols, 2004; Nicholls and Cazenave, 2010) An Early Warning (EW) system (Grasso and Singh, 2011) can prevent losses. Focus on monitoring and forecast activities to be included in an EW system. Objective Design a sea water level EW system for target coastal areas along the Italian coasts, using a semi-empirical model capable to use the numerical models data as forecasting components and observational data as calibration components.

3. 1.Forecasting Component Sea Water Level (SWL) signal components characterized using different numerical models: 2. Calibration Component Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) Observational Network (Bencivenga et al., 2012) 3. Forecasts Combination Combine different forecast realizations with available observations Methods: Sea Water Level (SWL) Forecasting System TIDES WAVES SSH + STERIC ATM. PRES (Cohelo et al., 2004)

4. Summary and Conclusions A semi-emphirical model for SWL forecasting has been develop, using numerical models and in-situ observations as forecasting and calibation component respectively. At this stage of development, the system do not considered the SWL signal due to wave-setup. The forecasting system developed show skill with respect to the observational data of the Italian tide-gauge network. A bayesian approach has been adopted in order to combine and calibrate forecasts with observations and to obtain different forecast realizations. The prediction obtained considering three ensemble members show the higher skill in terms of anomaly correlation coefficient and RMS In conclusion: the system developed can represent a monitoring and forecasting tool to be included in an Early Warning system for sea-level extreme events along the Italian coasts.

5. Semi-Empirical Model Description: Forecasting Component ηM = Sea-level mass variations component resolved by OGCM (incomp. and Bousinnessque) Sea-Surface Height (SSH) Mediterranean Forecasting System (Oddo et al., 2009) horiz resolution: 1/16 ; 72 vertical levels Geografical domain: -18.125 W – 36.25 E ; 30 N – 46 N ηIB = Inverse Barometer Effect (Dourandeau and Le Traon, 1999) European Center for Medium-Range Weather Forecasting Horiz. Resolution: 1/4 ηST = Sea-level steric comp. due to water column density variations (Mellor and Etzer 1995)

6. Semi-Empirical Model Description: Forecasting Component ηWA = sea-level and waves interactions in the coastal zones, due to the transfer of momentum at the wawe breaking to the water column (wave-setup) Wave Model: i.e. Wave Watch III (WWIII) (Tolman et al., 2009) Horiz. Res. : 1/16 ; Geografical domain: MFS Significand Wave Heigh (Hs) and Wave Peak Period (Tp) ηAT = Sea-level comp. due to astronomichal tides Oregon State University Tidal Prediction Software (OTPS; Egbert and Erofeeva, 2002) 8 tidal constituents Mediterranean solution: 1/30

7. Semi-Empirical Model Description: Forecasting Component ( ηWA) MFS model is coupled with the spectral wave model WaveWatchIII. Two ways coupling: - surface currents and SST computed by the MFS model are transferred to the wave model; - neutral drag coefficient computed by the WWIII passed to MFS model with 1-h frequency. WWIII forecast outputs available as hourly data ηW A signal implementation needs further investigations. the effect of ηW A component is not relevant if the models horizontal resolution is in the order of kilometers, since wave set-up is a processes that occurs in the surf zones that generally in the Mediterranean Sea has an extensions of few hundred meters also during storms (Ferrarin et al., 2012).

8. 2. Semi-Empirical Model Description: Calibration Component MFS OGCM domain and bathymetry ISPRA tide-gauge (Bencivenga et al., 2012) ISPRA wave-gauge Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) Observational Network Used as Reference system: - model and observations have a different reference systems (zero level). - anomalies are considered and observations mean values are added to the model outputs (Guarnieri et al., 2012) Calibrations parameters: multiple regression against in-situ SL signals filtered accordingly. i.e: Steric signal obtained from model data as f(T,S,P) againts a climatological value obtained considering 20 years of observed data

9. 2. Semi-Empirical Model Description: Output Monitoring and forecasting informations for 26 target loacations. Data available over a nine days time window. 7 days of observational data and 2 days of forecast data, as 3-h mean. Every day the time window considered shifted 24 hours ahead. A single run per day is performed. In the near future multiple runs/day in order to update the forecasts several time during the 24 hours.

10. Semi-Empirical Model : Output

11. Forecast Combination Forecasts Combination (Coelho et al., 2004; Stephenson et al., 2005) Three steps: 1.Choice of the prior distribution. 2. Modeling of the likelihood function. 3. Determination of the posterior distribution. Bayes Theorem = Observations distribution = Forecasts = Observations = Forecasts distribution

12. 1.STANDARD FORECAST Standard and Calibrated Forecasts STANDARD FORECAST (as described before) CALIBRATED FORECAST CALIBRATED ENSEMBLE FORECAST Each member due to different physical representation of the SL signals 2. = 3. =

13. Semi-Empirical Model : SKILL

15. Concluding Remarks Forecast Combination results as good way to improve the forecast SKILL Maximum Likelihood function (now linear) can be the key to go a step further. (neural networks) Using multiple models (nested) to consider each SL signal component could allow to create larger ensembles. Future work will be centered on: the implementation of the ηWA component in the SWL forecasting system the reasearch and analysis of new observational data-set, both in-siu and remote sensing, in order to test the calibration component with larger data-sets to extend the spatial domain of the SWL forecasting system.