Sea level modelling and forecasting in the Northern Adriatic

Transcription

1 Sea level modelling and forecasting in the Northern Adriatic F. Raicich/*) S. Piacsek/*) R. Purini,^ L. Perini<*> WCNR, Istituto Sperimentale Talassografico, viale R. Gessi 2, Trieste, Italy (^ Naval Research Laboratory, Code 73, Stennis Space Center, MS , USA ^ Ufficio Centrale di Ecologia Agraria, via del Caravita 7/a, Roma, Italy Abstract This paper presents some results on modelling and forecasting residual sea level by means of a one-layer free-surface barotropic ocean model, forced by surface wind stress and atmospheric pressure fields obtained from an atmospheric Limited Area Model (LAM). As a case study, the model is applied to a marked storm surge event that occurred in the Northern Adriatic Sea in December 1997 and was followed by seiche oscillations for several days. Model outputs are compared to the residual sea level deduced from observations made at various tide gauge stations along the eastern coast of the Adriatic Sea. Overall, the model simulates the main features of the observed time series reasonably well, but details are not always reproduced correctly. 1 Introduction The Adriatic Sea is a semienclosed basin (Fig. 1) and the activities which take place along its coasts can be significantly affected by several processes related to sea dynamics, as, for instance, pollutant dispersion, coastal erosion and marked sea level fluctuations.

2 11 Coastal Engineering and Marina Developments % #43 a ~ longitude E Figure 1: Map of the Adriatic Sea with the tide-gauge stations. In this paper we focus on the fluctuations of residual sea level, that is the component representing the difference between the observed sea level and the astronomic tide. Such fluctuations are connected with the atmospheric forcing and occur on time scales between few hours and several days. Forecasting the residual sea level can help planning routine activities, as for instance those related to coastal and harbour maintenance, and limiting the impact of potentially harmful events, as storm surges. Storm surges may occur rather frequently in the Northern Adriatic, particularly in autumn and winter, and are connected with the combined action of atmospheric pressure lows and related southerly winds along the Adriatic basin. During severe storm surges the northern coastal areas can be flooded, but even minor events may produce a significant limitation of river discharge into the sea, with in turn may result in floods affecting inland areas. Note also that the meteorological situations favouring storm surges frequently determine heavy precipitation over the drainage basins of the Northern Adriatic rivers. Numerical analysis and prediction of storm surges in the Northern Adriatic Sea dates back to the 197's. Since then several authors have been involved in such studies (e.g. Accerboni & Manca\ Robinson et al?, Stravisr*,

3 Coastal Engineering and Marina Developments 111 Cerovecki & Orlic*). At present, numerical and statistical sea level predictions are routinely performed to provide warnings to the population living within and close to the Venice Lagoon (Cecconi et a/.^), but a regular service for the whole Northern Adriatic is not available. Scientific institutions from Italy, Slovenia and Croatia are presently involved in the design and implementation of an oceanographic operational system in the Adriatic Sea region, in the framework of the project named Coordinated Adriatic Observing System (CAOS). The final aim is to provide the public and the scientific community with observations and forecasts concerning the marine environment, in order to keep the basin under control and manage potentially dangerous events. Here we present some results on residual sea level modelling and forecasting by means of a one-layerfree-surfacebarotropic ocean model, forced by surface wind stress and atmospheric pressure obtained from an atmospheric LAM. A short description of the models will be given in next section. We will then discuss the results of its application to the storm surge observed in December 1997, and compare them with experimental data of residual sea levels. 2 The models The ocean model is barotropic and based on the nonlinear shallow water equations on a -plane (see Hulburt & Thompson^ for a detailed description). It essentially represents the vertically averaged component of the flow with a one-layer isopycnal fluid of constant density with a free surface. The time integration uses a leapfrog scheme and the spatial derivatives are second-order accurate centred differences. The grid spacing is ~ both in latitude and longitude and bottom topography is mainly extracted from the U.S. Navy Digital Bathymetric Data Base 5; since the bathymetry of the shallower northern region is not very accurate, it has been corrected to match the observations. The use of a barotropic model is justified by two reasons: a) relatively short time intervals (few days) are to be studied, during which only small changes in the vertical structure of the water column could be expected; b) the events to be simulated usually take place in autumn and winter, when the water column is almost homogeneous. The prognostic variables are the x- and y-components of vertically integrated velocity (i.e. transport) and the free surface elevation, and are obtained from the following equations. drj 3U dv "^7 + ~Z~ + ~^~~ = (3) ot ox oy ^ '

4 112 Coastal Engineering and Marina Developments Here -^ is the total derivative operator, C? = ( 7, V) is mass transport, rj is free surface elevation, rja is the free surface elevation component due to atmospheric pressure, / is Coriolis parameter, g is gravity acceleration, H(x,y) is bottom topography, p is water density, f = (TX,TJ,) is surface (wind) stress, a = 1~ *~* is bottom friction coefficient, A = SOm^s"* is lateral friction coefficient and V = (^, ^). The bottom stress components are parameterized by the linear functions of mass transport au and crv, respectively. The coastal boundaries are rigid and no-slip, that is both the tangential and normal velocity components are set to zero. The only open boundary is located to the South across the Otranto Channel at N. The open boundary conditions are obtained by running an analogous barotropic model for the whole Mediterranean Sea on a coarser grid ( x grid spacing). The ocean models are forced by wind stress and atmospheric pressure. In the present configuration tidal forcing is not included. The forcing fields for the Adriatic model come from the outputs of the Data Assimilation Limited Area Model (DALAM), interpolated onto the ocean model grid. The Mediterranean model is forced by the atmosphericfieldstaken from the reanalyses produced by the National Centers for Environmental Predictions (NCEP). The data are provided every 6 hours on a 2.5 x2.5 grid, and interpolated onto the model grid. The model domain includes a buffer zone in the Atlantic Ocean outside the Gibraltar Strait. DALAM is the atmospheric model currently operated by the Ufficio Centrale di Ecologia Agraria (Central Office of Agricultural Ecology) in Rome. Its domain covers the geographical area 3-56 N, 5 W-26 E with grid spacing of approximately.28 in latitude and.39 in longitude; vertical discretization consists of levels. Initial and boundary conditions are provided by the analyses made by the European Centre for Medium-range Weather Forecast. In the present configuration the DALAM is initialized every day at UTC and provides forecasts for the following 72-hour period with 3-hour time step. 3 The case study The ocean model is tested by simulating the storm surge event whose maximum development was observed on December This event was the result of the combination of local atmospheric pressure decrease and southeasterly winds induced by a marked zonal pressure gradient over the Adriatic region, due to the presence of a cyclonic circulation over the Western Mediterranean and an anticyclone over the Balkans. This pattern produced the maximum effect on the basin on 19 and December, while on the following days the pressure gradient weakened and the wind was prevailingly westerly. Figure 2 shows the synoptic weather situation obtained from NCEP reanalyses of wind speed and atmospheric pressure at 12 UTC on December.

5 Coastal Engineering and Marina Developments N 5N- 45N 5W E 25E 3E 1 Figure 2: Atmospheric pressure (hpa) and wind (ms~*)fieldson December 1997 at 12 UTC from NCEP reanalyses. As a result, the residual sea level exhibited the evolutions shown in Fig. 3 (dotted lines), which presents the hourly observations made at the stations shown in Fig. 1. The data for Trieste are provided by Istituto Sperimentale Talassografico of CNR, in Trieste, those for Dakar by the Geophysical Institute of the University of Zagreb, and those for Rovinj, Zadar, Split and Dubrovnik by the State Hydrographic Institute in Split. The surge reached its peak between 12 and 15 UTC on December, depending on the location; by looking at hourly time series of atmospheric pressure and wind speed observed at the same locations (not shown), it can be seen that the peaks occurred in connection with a local pressure minimum and a sudden decrease of wind speed and change of wind direction. The sudden change in the atmospheric forcing was particularly abrupt in the Northern Adriatic and enabled a well developed seiche to start. This seiche represents the main free oscillation mode of the Adriatic Sea and has a period of about 21 h (Cerovecki et o/j, Raicich et al*). Residual sea level is modelled for the period December 1997 by running the ocean model with the DALAM atmospheric forcingfields.the model spinup lasts days and is performed by means of constant forcing, namely that of 15 December at UTC. With these initial conditions, runs

6 114 Coastal Engineering and Marina Developments - Figure 3: Trieste Bakar Split Rovinj Zadar Dubrovnik day day Comparison of observed residual sea level (dotted lines) with model simulations (solid lines). are performed with 3- and 6-hourly forcings. These sets of atmospheric forcing are composed of daily analyses at UTC and forecastedfields (at UTC + 3, + 6, up to + 21 h) in between. We checked that the forecasts for UTC + 24 h do not differ significantly from the analyses at the same time. The boundary conditions are prescribed for U and TJ with the same frequency as the atmospheric forcing. The Mediterranean model is spun up for 22 days with the forcing of 1 November at UTC, and then run through the end of December. The NCEP atmosphericfieldsare found to be not very accurate over the Adriatic region, probably because of their too coarse resolution, therefore we built new atmospheric data sets by replacing NCEP data with the DALAM data in the domain where the latter are available, with a smooth transition at the interface. The model performance is not very sensitive to the choice of 3- or 6- hourly forcings. By contrast, some tests show that a less frequent forcing (e.g. 12-hourly) is too coarse to resolve the surge buildup and seiche peaks. Figure 3 shows the comparisons between observations and model simulations with 3-hourly forcing. Although our work is focussed on the Northern Adriatic, where the storm surge is more marked, we display also the results

7 Coastal Engineering and Marina Developments Trieste Bakar Split day Figure 4: Comparison of observed residual sea level (dotted lines) with forecasts (solid lines) initialized on 18, 19 and December (from left to right) for Trieste, Bakar and Split (from top to bottom). obtained for Split and Dubrovnik (Middle and Southern Adriatic). Overall, the model performance can be considered satisfactory. The storm surge peak is reasonably well simulated, particularly the time of its occurrence, while the details are sometimes missed during both the buildup phase and the oscillatory phase. We made experiments in which a seiche was induced by switching off a spatially uniform wind stressfield,which had been active for 1 days. We could see the Adriatic model could reproduce the seiche period correctly, therefore a possible reason for the differences is an inadequate choice of the atmospheric forcing used by the Mediterranean model (for instance due to its coarseness), which in turn affects the boundary conditions for the Adriatic model. This is a delicate point which will be examined in more details in future experiments. Forecasting experiments are performed by forcing the Adriatic model for 72 hours with the 3-hourly forecasted wind stress and pressure fields of DALAM. The model initial conditions are prescribed at UTC every day from 17 to 22 December and are taken from the simulations described above. The open boundary conditions are extracted from the Mediterranean

8 116 Coastal Engineering and Marina Developments hindcast run. In the forecast experiments we aim to obtain predictions of the sea level deviation from the initial height at the beginning of the forecast run. Fig. 4 shows the residual sea level forecasts for Trieste, Dakar and Split with initializations made on 18, 19 and December at UTC. The model produces reasonably good results, but, again, some discrepancies can be seen. 4 Conclusions We presented some results of residual sea level simulations and forecasts connected with a storm surge event of December 1997 in the Adriatic Sea. The barotropic ocean model is intended to be a part of a future oceanographic operational system to be implemented in the Adriatic Sea in the framework of project CAOS. Due to its simplicity, the model consumes little time. On an SGI workstation one day of simulation takes approximately seconds for the Adriatic run and 8 minutes for the Mediterranean run. A key issue is represented by the atmospheric forcing: Using the atmospheric fields produced by a LAM is advisable because of the basin size (approximately km long and 25 km wide) and the complex orography surrounding it. Further studies are required to investigate the impact of the atmospheric forcing on the model performance and possibly explain the inaccuracies in the results. Acknowledgements The authors wish to thank Prof. M. Orlic (University of Zagreb, Croatia) and Dr. I. Vilibic (State Hydrographic Institute, Split, Croatia) for kindly providing the sea level data for the Croatian stations. This work has been partly funded by Agenzia Spaziale Italiana under contract ASI ARS References 1. Accerboni, E. & Manca, B. Storm surges forecasting in the Adriatic Sea by means of a two-dimensional hydrodynamical numerical model, Boll Geof. Teor. Appl, 15, pp. 3-22, Robinson, A.R., Tomasin, A. & Artegiani, A. Flooding of Venice: Phenomenology and prediction of the Adriatic storm surge. Quart. J. Royal Met. Soc., 99, pp , Stravisi, F. Analysis of a storm surge in the Adriatic Sea by means of a two-dimensional linear model, Accad. Naz. Lincei, Rend. Cl Sc. Fis., Mat., Nat., 54/2, pp , Cerovecki, I. & Orlic, M. Modeliranje rezidualnih vodostaja Bakarskog zaljeva, Geofizika, 6, pp , 1989.

9 Coastal Engineering and Marina Developments Cecconi, G., Canestrelli, P., Di Donato, M. & Cecchinato, M. I nuovi sistemi per la previsione dell'acqua alta a Venezia, Quaderni Trimestrali del Consorzio Venezia Nuova, 5, pp , Hulburt, H.E. & Thompson, J.D. A Numerical Study of Loop Current Intrusions and Eddy Shedding, J. Phys. Oceanogr., 1, pp , Cerovedd, I., Orlic, M. & Hendershott, M.C. Adriatic seiche decay and energy loss to the Medterranean, Deep-Sea Res., 44, pp. 7-29, Raicich, F., Crisciani, F., Malaac, V., Orlic, M. & Vilibic, I. On the seiche event in the Adriatic Sea on December 1997, Oceanography, 11, SuppL, p. 38, 1998.