ON THE ACCURACY OF ATMOSPHERIC FORCING FOR EXTRA-TROPICAL STORM SURGE PREDICTION PAULA ETALA Department of Meteorology, Naval Hydrographical Service Comodoro Py 2055, 1104 Buenos Aires, Argentina e-mail: [email protected] The wave/surge numerical prediction system at the Naval Meteorological Service (SMARA) comprises the SMARA/WAM wave model and a storm surge model applied to nested grids covering the de la Plata River and the shelf sea, respectively, as well as a coarser wave model that provides the ocean wave input. The atmospheric forecasts for the two finer mesh models are provided by the Eta model at the National Weather Service (SMN). The ability of the system to reproduce local effects in estuarine and coastal winds was recently improved by considering one-way coupling of the air-sea momentum exchange through the wave stress, and best forecasting practices for downscaling. Simultaneously, an ad-hoc review of the surface wind calculation algorithms in the operational version of Eta took place at SMN and wind fields from the higher resolution model Pampa were provided for assessment. The inclusion of long period atmospheric pressure forcing in tide and tide/surge calculations corrected a systematic error in the surge, produced by the South Atlantic Ocean quasistationary pressure patterns. The overall consideration of these aspects in the atmospheric forcing makes the deterministic forecast from the numerical models to slightly improve within the 24–hour range the best available prediction at Buenos Aires, i.e. the water level nowcast based on the closest water level observations within a 6-hour foresight. The maximum forecast range provided by the realtime use of observations is approximately 12 hours, the maximum available prevision until the numerical prediction system implementation. Although the numerical forecast accuracy degrades after the first 48 hours, the improvement to the full range observation-based prediction is maintained at the inner de la Plata River area and extends to the first 3 days at the intermediate navigation channels. Keywords: verification.

storm surge, prediction, modeling, surface stress, coupling, atmospheric forcing,

1. INTRODUCTION

The water level rising due to the action of storm surges is a cause of major damage on the southern coastal zones of the de la Plata River and in the upstream delta area. On the other hand, the combined effect of surge and waves at the Uruguayan northern shore, more exposed to open sea, as well as at the adjacent Atlantic coast, can be devastating. The Center for Prevention of the de la Plata River Rising at the Naval Hydrographical Service (SHN) receives real-time water level data from tide gauges along the de la Plata River south-western coast and the Atlantic coast, which are shown in Figure 1. This information, in conjunction with forecasted winds, provides the input to an empirical model which estimates the storm surge within a 12-hour foresight that may be increased to 24 hours in the case of remote surges generated along the open ocean coasts, which can be detected at Mar del Plata. Warning thresholds for a Rising Sea Level Warning depend on the position along the coast and vary from 2.80 m above reference level at Buenos Aires to 2.40 m at Mar del Plata and/or Santa Teresita.

FIGURE 1. Right: Area of study. Left: Real-time stations for water levels in the de la Plata

River and adjacent Atlantic coast; the grid corresponds to the Eta SMN atmospheric model. The amplitudes of atmospherically-induced long period tidal constituents are relevant in the entire region, as it comes out from SHN tidal analyses. D’Onofrio et al. (1999) identify the solar annual Sa among the largest contributions to tidal range at the upper de la Plata River, only overcome by the dominant M2 and N2, O1, K1, and S1, which produce the characteristic inequality semidiurnal tidal regime in the area. To a minor extent, the solar semiannual Ssa is also relevant and its amplitude compares to those of the most significant shallow water constituents only in the extremely shallow head of the de la Plata River. Here, it will be shown that quasi-stationary pressure patterns play an important role in the determination of the total water level at Southern Hemisphere mid-latitudes and that their effect on the tides should be considered in the storm surge prediction. The dependency of sea surface roughness with the wave state of development in wind seas is widely accepted. In a coupled model study, Weber (1994) identified the southeastern South American shelf sea as one of the areas with a largest degree of interaction in the world. This is due to the dominant generation and young seas at this zone, located lee of the South American continent. For simulations of selected storm events in the de la Plata River area (Etala, 2001) the wave stress represented over 70% of the total stress during the developing stage. Next section describes improvements introduced in the formulation of the atmospheric forcing to the storm surge models and their application in the forecasting system. Section 3 presents some considerations on the atmospheric models applied and the predicted wind fields. Section 4 compiles the latest results of quasi real-time storm surge forecast verification and section 5 summarizes the experience gained in the application of the storm surge forecasting system, with emphasis on the atmospheric forcing. 2. ATMOSPHERIC FORCING IN THE STORM SURGE MODELS

A set of two nested models is applied to tide and surge calculations at the continental shelf sea and the de la Plata River (RP, hereafter). The storm surge models are based on the depth-averaged shallow water equations (Etala, this symposium). The model resolution on the shelf is 1/3º lat. x 1/3º lon. and 1/20º lat. x 1/20º lon. for RP. Here, only specific implementation details of the surface wind stress and the atmospheric pressure forcing are discussed.

2.1 WAVE AND SURGE MODELS ONE–WAY COUPLING

In a previous decoupled version, the surface wind stress was related to 10-m wind through the extensively used quadratic law r r r Τs = ρ a C D W | W | r where ρa is the air density, W is 10-m wind and CD is the drag coefficient, varying quasi r linearly with the wind speed | W |. In the current version, the air-sea momentum exchange depends on the wave state of development, as it is briefly described below. The wave stress modifies the total surface wind stress, producing relatively larger values during the initiation of the most intense stage of the storm, when wave growth occurs. This interaction has significant consequences for essentially transient phenomena, such as the storm surges. One-way coupling between a wave and a storm surge model is described by Burgers et al. (in Komen et al., 1994). In Etala (2001) it is shown that this coupling reproduces better the observed values of an extreme event in the de la Plata River. The same scheme was tested at a pre-operational stage of the current implementation, which also revealed the need of including the coupling effect. The SMARA/WAM wave model is based on the third generation model WAM4 (Komen al., 1994), including different implementation details on three nested grids. The most extensive and coarse grid covers the Southwestern South Atlantic and provides boundary information to the shelf wave model, which is set at ¼º x ¼º lat./lon., in a domain that is approximately coincident with that of the shelf sea storm surge model. At the RP, the wave and surge models have identical resolutions. The two latter wave models determine the air-sea momentum exchange through the modification of the roughness length (hence, of the friction velocity and drag coefficient), over the wave field. The surface wind stress is then introduced FIGURE 2. Storm surge (m) in Buenos Aires at the end of July in the storm surge model as 2006. Red: decoupled. Τs = ρau*2 Blue:coupled. Black dots: with friction velocity u* as provided by SMARA/WAM. hourly observed values. Coupling the wave and surge models introduced a significant and systematic improvement in the forecasting practice. A negative surge in July 2006 at Buenos Aires produced by northwesterly winds and the sudden rise that followed, are shown in Figure 2. The decoupled version (red line) could not reproduce the event correctly. The off-shore wind (alongshore in RP) provided a favorable scenario to identify the effect of the coupling on the surge. The coupled run (blue line) produced the most accurate results for the minimum, as well as for the subsequent rise. 2.2 LONG-TERM ATMOSPHERIC PRESSURE FORCING

The open boundary condition as proposed by Davies and Flather (1978) is applied to the storm surge models of the shelf sea and RP. The tide and surge inputs are prescribed, so that the outgoing current follows the well-known expression

where q is the outward depth-averaged current normal to the boundary; D is depth; is the phase speed of outgoing long gravity waves, and are the prescribed tidal water level perturbation and current at the boundary, respectively; h is the water level calculated at an adjacent inner point and finally, hM and qM are the incoming surge level and current, respectively. The latter are provided either by a coarser model (nesting) or by some previous knowledge. At the shelf model open boundary, the meteorological input is given by the with and p the mean and instantaneous atmospheric inverse barometer relation pressure, respectively. Surge current at the boundary is considered null. An important negative bias was noted in water levels at the first implementation of the storm surge models, in both coupled and decoupled cases, which was identified as being produced by the quasi-permanent atmospheric pressure gradient. Evidence was found that the quasi-stationary pressure patterns, which are typical at Southern Hemisphere mid-latitudes, produced a systematic depression in water levels at the zone of study of Figure 1. We should note here that, as in many other applications elsewhere, no long period atmosphericallyinduced tidal constituents were included in the runs and only the principal diurnal and semidiurnal constituents of the tide were selected for the tidal forcing. In order to account for that, the classical open boundary condition was adapted to provide the effect of climatological pressure fields in the tidal model runs. The inverse barometer condition is applied at the shelf open boundaries also for the tidal runs as p − p mon hM = a ρg where hM is the meteorological contribution to the water level at the boundary, ͞pa and p͞ mon are the annual and monthly means of the sea level pressure (SLP), respectively. To avoid input discontinuities at input times, monthly means are interpolated to a daily value.

FIGURE 3. Blue: Hindcasted storm surge 1- 8 July 2007. Red: climatological hydrostatic forcing at open boundaries for tidal run, removed. Black dots: hourly observed surge. Left panel: Mar del Plata (open sea coast). Right panel: Buenos Aires (inner RP).

The bias correction due to the extra pressure forcing introduced in tidal runs for the first 8 days of July 2007 hindcast is shown in Figure 3. For clarity, running means have been applied to remove 3-hour disturbances produced by the atmospheric input to the model. The negative systematic error is pronounced at Mar del Plata, on the open sea coast (left panel). Although the surface wind stress is the most important forcing within the RP, the effect of the open boundary condition off the shelf edge is evidenced even at Buenos Aires (right panel), located at its innermost area. 3. ATMOSPHERIC MODELS AND THE PREDICTED SURFACE WINDS

Ocean wave hindcasts are driven by the National Centers for Environmental Prediction (NCEP) global 10-m wind fields acquired four times daily at an approximate resolution of 35 km on a regular grid. The surface stress provided by the wave models drives the storm surge models, as described in previous section. The shelf and RP 96-hour wave and surge forecasts are driven twice daily by the 10-m wind and SLP fields from mesoscale Eta at the National Weather Service (SMN) at a 1/3º x 1/3º lat./lon., although its full resolution is 25 km. The current version of the Eta SMN is based on the ICTP 2005 upgrade (http://www.cptec.inpe.br/etaweb/). An experimental higher resolution implementation (Pampa) is applied at 10 km x 10 km to a smaller domain that includes the RP area. The RP wave and surge models are run experimentally once daily, driven by Pampa 36-hour forecasts of 10-m wind and SLP fields. Summarizing, the resolutions of the regional atmospheric model and global model used for hindcasts are roughly those of the wave and surge models on the shelf, while the RP models double the resolution of the nested Pampa atmospheric model.

FIGURE 4. 10-m wind fields (knots) 12-hour forecasts or 27 July 2007 12 Z from Eta SMN

(left) and Pampa (right). Pilote Norden buoy (Riovia S.A.), Aeroparque J. Newery (Argentina) and Carrasco Airport (Uruguay) stations. The 2005 Eta upgrade includes an averaging of four wind neighboring points as an improvement to lower layers wind fields over topography. This resulted in a negative impact on the wind fields from the coarse Eta SMN over RP, evidenced as the calculated storm surge missed every significant peak at the stations of reference (not shown). It is seen in Figure 1, where the coarse grid is over imposed, that very few points actually correspond to wet points in RP. The averaging had an unrealistic smoothing effect of the land-sea contrast on the wind field. On the contrary, this effect was not seen in the high resolution Pampa model results. As

an outcome of those tests, this aspect of the upgrade from Eta 2002 to 2005 was annulated in the coarse (operational) Eta SMN model, but maintained in Pampa (M. Suaya, personal communication). It is well known that mesoscale models develop rapidly the local physics, providing benefit even when they lack of initialization at an adequate scale. It will be shown in next section that the performance of the storm surge hindcast was not degraded in the first forecast hours. Even though, improvements based on local knowledge (downscaling) still produced further benefit as increased resolution had proved not to be enough to account for some relevant local effects. These results are in agreement with the ordinary forecasting practice in the small scale RP. We show here a sample case of a rapid sea level rise for which an additional wind intensity enhancement was applied to the forecast, despite the evident benefit of increased resolution. The 12-hour 10-m wind forecast for 27 July 2007 12 Z from Eta SMN and Pampa models are shown in Figure 4 left and right panels, respectively. Pampa model correctly developed the wind pattern over the inner RP but wind intensity was significantly underestimated. The monitoring buoy located in one of the major navigation channels, Pilote Norden (Riovia S.A.), reported sustained SW 20 kts for the period. Modeled intensities shown in the picture, even those produced by the Pampa model, were in concordance with coastal land observations at the airports J. Newery (Argentina) and Carrasco (Uruguay). These differences are sustained and led to the implementation of empirical corrections to the long-shore winds intensity. It will be seen in the storm surge verification plots in next section that an accurate forecast of the storm surge was nevertheless achieved by the standard FIGURE 5. The 10- m wind intensity bias downscaling applied and coupling. (m/s) for the Eta SMN 96-hour forecasts The regional scale wind intensity from from November 2006 to July 2007, as the operational Eta SMN 24 to 96-hour forecasts compared to Jason-1 altimeter wind data. is routinely verified against Jason1 altimeter wind data. A negative bias was initially noted with increasing range (Etala et al., 2005). A fine tuning that was applied to improve the storm surge forecasts on open sea stations (Mar del Plata and Santa Teresita) corrected this systematic error and an apparent random pattern is found now on the wind intensity mean error. The Eta SMN 96-hour forecast mean intensity error for the period November 2006-July 2007 is shown in Figure 5, where still the effect of individual storms is noted. Quality control to altimeter quasi real-time data is applied as in Etala et al. (2005). 4. STORM SURGE FORECAST VERIFICATION

The storm surge and total water level forecasts verification is performed in quasi-real time on a routine basis. Forecast ranges from 12 to 96 hours are arranged in ± 6 hour windows for validation. Hourly tide gauge data and high and low water forecasts are obtained once daily from the Center for Prevention of the de la Plata River Rising. Tidal analyses at the selected stations in Figure 1 are provided by the Tides Section at SHN. Hourly plots of storm surge hindcasts and forecasts for July 2007 at the reference stations Mar del Plata, Santa Teresita,

Oyarvide Tower, La Plata, and Buenos Aires are shown in Figure 6. In this month, most events were reflected from the open sea coast at Mar del Plata to the inner RP at Buenos Aires. The 7-8 July event produced by along-shore winds on the shelf was well forecasted at all ranges. At 23-24 July a slightly misplacing of the extra tropical cyclone, varying at every forecast run, and finally centered close to the RP mouth was reflected in the dispersion of the surge forecast errors at the inner RP. Finally, the 27-28 storm surge (the largest at open sea coast) was only well predicted in the short term (24 hours).

Panel A. Mar del Plata.

Panel B. Santa Teresita. FIGURE 6. July 2007 hourly observed, hindcasted and forecasted storm surge at the stations of reference.

Panel C. Oyarvide Tower.

Panel D. La Plata.

Panel E. Buenos Aires.

FIGURE 6. Continued.

FIGURE 7. Total high water level error statistics corresponding to selected storm surge events during May, June, and July 2007. Upper left: Oyarvide Tower. Upper right: Buenos Aires. Bottom left: Mar del Plata. Bottom right: Santa Teresita. Significant surge events are included monthly in the overall statistics that are presented above. For the three stations along the RP (Oyarvide, La Plata and Buenos Aires) and Santa Teresita, those events producing positive surges above 1 m at Buenos Aires are included in the statistics. Open sea events producing surges above 0.5 m at Mar del Plata are considered for Santa Teresita and Mar del Plata analyses. The maximum surge level for the event is used for surge verification while every high water level during the event is included in the total level validation. In the latter, the high water levels predicted up to 12 hours ahead by the empirical method are included for assessment (labeled as forecast), as well as their shorter term updates (nowcast). The total high water level forecast validation merges the surge magnitude and timing errors. Its statistics for May, June and July 2007 are shown in Figure 7. Upper panels correspond to the validation of the RP model predictions at the Oyarvide Tower and Buenos Aires (left and right panels, respectively). The dashed part of the series lines refers to model hindcast and to the manual forecast and nowcast; full lines correspond to the

model forecast. Bottom panels show the verification of the shelf sea model predictions for Mar del Plata and Santa Teresita (left and right, respectively). No manual prediction methods are available for the latter open sea stations. It should be noted that a systematic error correction, constant with forecast range, is applied to the Santa Teresita model output. The period included 14 cases in Buenos Aires for which only 8 forecast updates or nowcasts were provided. The total number of cases for Oyarvide and Santa Teresita were 9, while Mar del Plata presented a very active period, with 22 high water occurrences during surge events according to the criteria described above. The model performance was regular within the first 48 hours, even improving the manual predictions in RP. No reference forecast could be adopted for outer stations. The positive surges in the relevant cases selected were systematically under-predicted although the bias was low in the short ranges. Results are summarized in next section.

1.8 1.6 1.4 1.2 1 0.8 0.6

) m ( 0.4 l e v 0.2 e l r e 0 t a W -0.2 -0.4 -0.6 -0.8 -1

hindcast

36-hr fcst.

24-hr fcst.

12-hr fcst.

obs.

FIGURE 8. Storm surge at Buenos Aires for July 2007 from Pampa atmospheric model input.

Finally, RP short-term surge predictions based on the Pampa atmospheric model forecasts are presented in Figure 8. The 36-hour forecast range is related to the limited area of the high resolution atmospheric model. Although winds were slightly stronger than in the coarser model, as it was shown in section 3, the enhanced resolution did not provide an evident improvement in the accuracy of the storm surge prediction. Further case studies in very local wind events are needed to assess this conclusion. It should be noted that a systematic 5 % increase in wind intensity over the water that is made on the coarser model wind fields was avoided in these runs. Moreover, the algorithm introduced to the Eta 2005 model upgrade could be maintained in this resolution. 5. SUMMARY OF RESULTS AND CONCLUSIONS

The benefit of the enhanced quality boundary layer formulation in mesoscale atmospheric models to marine forecasting is undoubted. However, the concretion of those improved physics in accurate storm surge predictions is not always straightforward. Land-sea contrasts may not be accurately reproduced in small wet models domains if the atmospheric model grid is relatively coarse. In this case, some smoothing in the surface wind fields introduced in the Eta model for a better representation of low layers winds over complex topography had to be suppressed in the coarser version as it produced a severe underestimation of surge peaks in the de la Plata River. An improvement in the storm surge forecast accuracy by a higher resolution atmospheric input could not be assessed in RP for the 3-month period from May to July 2007. Even though some small scale atmospheric circulation features are better reproduced by the high resolution model, the surface wind intensity over the water is still underestimated as compared to observations. These tests’ results suggest that, as far as the atmospheric model physics are not improved, the isolated effect of enhanced resolution is not enough to upgrade the storm surge forecasting system. The longer term in the regional Eta SMN forecasts still provides a benefit when compared to the application of limited area models. Confirming preliminary results, the surge and wave models one-way coupling through the wave stress demonstrated to improve significantly the storm surge on a regular basis and was introduced in operations. Long term atmospheric pressure forcing due to the seasonal cycle was also incorporated in the operational system. This modification eliminated a severe bias in the surge that otherwise occurred if long term atmospheric tidal forcing were not considered. The accuracy required in applications such as land flooding is usually in the order of one to a few tenths centimeters. At SHN, this goal has been locally best achieved within very short forecast ranges by empirical nowcasting methods that consider nearby real-time water level observations. Model guidance introduces a clear benefit in the existing forecasting system, moreover by providing accurate predictions at the open sea coast, where there were no available forecasts so far. More extensive verification is needed to include negative and dropping surges. For surges that exceeded 1 m at the head of the RP, the numerical model improved the performance of the very short range forecasts within its first 24 hours. The intermediate navigation channels are the most benefited by the model predictions. On the other hand, the numerical model forecasts improved the empirical method 12-hour forecasts within the third and second day range for the intermediate and inner de la Plata River, respectively. For consistency with the pre-established methods, these are considered to be the limits for usefulness of the model forecasts. The possible extra benefit of further foresight, with some loss in the current accuracy of predictions, has not been assessed among users yet. Neither have the needs for accuracy of water level forecasts at the open sea coasts, as there is

no previous experience on storm surge predictions. These requirements and other user outreach aspects are subject of future extensive assessment. Acknowledgements Martina Suaya (SMN) reviewed the surface wind interpolation scheme in Eta and provides the experimental high resolution winds. Mauricio Gatto supports wet models’ operations at SMN. Global meteorological fields are acquired from National Centers for Environmental Prediction (NCEP) and National Weather Service (NWS) ftp servers. Climatological MSL pressure fields were obtained from NCEP Reanalysis Derived data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/. The Jason-1 Operational Sensor Data Record (OSDR) altimeter 10-m wind data were obtained from the Physical Oceanography Distributed Active Archive Center (PO.DAAC) at the NASA Jet Propulsion Laboratory, Pasadena, CA. http://podaac.jpl.nasa.gov. Claudia Romero performs the on-going storm surge forecast verification. Stella Maris Alonso provided programming support to this work. REFERENCES

1. D’Onofrio, E. E., M. M. E. Fiore, and S.I. Romero, 1999: Return periods of extreme water levels estimated for some vulnerable areas of Buenos Aires, Cont. Shelf Res. 19, 16811693. 2. Weber, S. L., 1994: Statistics of the Air-Sea Fluxes of Momentum and Mechanical Energy in a Coupled Wave-Atmosphere Model. J. Phys. Oceanogr., 24, 1388-1398. 3. Etala, M. P., 2001: An investigation of the role of wave-induced stress in the ocean atmosphere models coupling (in Spanish). CLIMET IX-CONGREMET VIII, 165, 7 pp. 4. Komen, G. J., L. Cavalieri, M. Donelan, K. Hasselmann, S. Hasselmann, P. A. E. M. Janssen, 1994: Dynamics and Modelling of Ocean Waves, Cambridge University Press, 532 pp. 5. Davies, A. M. and R. A. Flather, 1978: Application of numerical models of the north-west European continental shelf and the North Sea to the computation of the storm surges of November to December, 1973. Dtsch. Hydrogr. Z., 14, p. 72. 6. Etala M. P., S. M. Alonso, and K. López Cristaldo, 2005: Pre-operational stage of a wave forecasting system for the SW South Atlantic and the Argentine Sea (in Spanish). CONGREMET IX, MAO-16, 9 pp.