Sediment Dynamics in the Adriatic Sea Investigated with Coupled Models

Abstract : Several large research programs focused on the Adriatic Sea in winter 2002-2003, making it an exciting place for sediment dynamics modelers. Investigations of atmospheric forcing and oceanic response (including wave generation and propagation, water-mass formation, stratification, and circulation), suspended material, bottom boundary layer dynamics, bottom sediment, and small-scale stratigraphy were performed by European and North American researchers participating in several projects. The goal of EuroSTRATFORM researchers is to improve our ability to understand and simulate the physical processes that deliver sediment to the marine environment and generate stratigraphic signatures.

In many of the well-studied systems, the combination results in dispersal systems that combine advection and diffusion in varying amounts (Swift et al., 1972).In most cases, however, the locus of deposition is close to the source of sediment supply, and accumulation rates decrease with distance in a pattern that refl ects the long-term correlation among sediment concentrations and current velocities.Observations made during the STRATAFORM study of the Eel River margin offshore California revealed a paradox that challenged these traditional views.There, measurements during the fi fth largest fl ood in 93 years indicated that rapid deposition of fl ood sediments occurred beneath the river plume in shallow water, but long-term accumulation was centered in deeper water (Sommerfi eld and Nittrouer, 1999;Wheatcroft and Borgeld, 2000).Wave-induced remobilization of the muddy fl ood deposits and subsequent downslope density fl ows were key mechanisms of cross-shelf sediment transport (Traykovski et al., 2000) (Figure 2).However, the Eel margin has large waves and high rates of sediment supply.Another important objective of the EuroSTRATAFORM program was to determine whether wave-induced density fl ows occur in more moderate environments such as the Adriatic Sea.

SCALING FROM EVENTS IN A SINGLE SEASON TO GEOLOGICAL TIME SCALES
One of the big challenges in sediment modeling is scaling.We do not have the computer power to resolve processes at the fi nest scales and also cover the necessary temporal and spatial scales.Even if we could, uncertainty in boundary conditions and forcing would require us to consider even the most deterministic model results in a statistical sense.One response to this dilemma is to adopt a hierarchy of models that inform each other across a range of domains and scales.The Adriatic modeling work presented here is an example of this approach.We use multiple nested meteorological models to   (Fox et al., 2004;in press).

Waves
Waves were simulated using SWAN (Simulating WAves in the Nearshore), which models the evolution of wave en-Oceanography Vol.17, No.4, Dec. 2004 63 ergy imparted by wind as waves propagate across the sea; are transformed (in frequency and direction) by shoaling, refraction, and wave-wave interactions; and are dissipated by whitecapping and bottom friction (Booij et al., 1999).We

Meteorology and Wave Models
Wind fi elds and air-sea heat fl uxes derived from both meteorological models COAMPS and LAMI matched each other and fi eld measurements (e.g., winds measurements offshore Venice) in a broad sense (correlation ~0.65; error in amplitude response ~10 percent), but differed in the details (Cavaleri, 1999).
Compared to the coarser resolution runs, the highest-resolution COAMPS nest (4 km) demonstrated better skill in producing the spatial pattern of Bora winds, and this led to improvements in modeled wind-forced ocean currents (Pullen et al., 2003).These wind patterns also seem to be captured by LAMI.Some of the small-scale structure is deterministic and refl ects improved resolution of orographic effects, but some is random and can degrade single-point correlations between model results and measurements (Signell et al., in press)     driven by winds and buoyancy input, matches circulation described by earlier drifter studies (Poulain, 2001) and recent modeling efforts (e.g., Pullen et al., 2003).Detailed comparison during Bora events (Figure 5) indicates that the model captures complex circulation pat-Oceanography Vol.17, No.4, Dec. 2004 67  These sensitivity studies will help us understand the weak links in the overall modeling system.

SEDIMENTTR ANSPORT MECHANISMS
Measurements and model simulations of  Oceanography Vol.17, No.4, Dec. 2004 69 understanding of the mechanisms that form deposits on the margins of epicontinental seas.

Figure 1 .
Figure 1.(a) Suspended particulates help mark the western Adriatic coastal current (WACC) in this MODIS image of the Adriatic Sea.EuroSTRATAFORM researchers deployed tripods (triangles), recovered box cores (squares), and made water-column observations (circles) in the WACC.(b) Researchers working on ACE, EACE, WISE, and ADRICOSM projects established radars for measuring surface currents (gray shading), deployed acoustic Doppler profi lers (squares), tracked drifters (red lines), completed hydrographic surveys (blue circles) and made other water-column measurements during the 2002-2003 experiments (circles).
transfer information from global scales to local scales (consisting of a few square kilometers).The coupling of atmospheric, hydrologic, wave, and oceanographic models allows us to evaluate the importance of regional and local forcing on the detailed physics of the bottom boundary layer, which we can examine in detail with the gravity-fl ow models.Coupled circulation and sediment-transport models allow us to verify processes against measurements during specifi c events and over the course of a winter transport season.Stratigraphic models let us extend the effect of annual processes over decades and centuries and factor in longterm changes in sea level and sediment supply.In the end, we hope to achieve a cascade of models that, although they are calibrated at small scales and over short periods, still allow us to build insight about cumulative effects because they are based on a fundamentally correct representation of the processes.OBSERVATIONS EuroSTRATAFORM researchers deployed instrumented tripods and moorings for nine months (November 2002 to May 2003); made water-column measurements in the western Adriatic coastal current (WACC) during voyages in November, February, and May; made shortterm (~2 day) studies of particle dynamics; and analyzed cores to determine characteristics of near-surface sediments

Figure 2 .
Figure 2. (a) Mud initially deposited in shallow water (20-40 m) beneath the Eel River plume is remobilized by waves and moved downslope as a gravity current, producing a fi nal deposit in deeper water (50-60 m), according to these model simulations.Th e fi rst fi eld observations of this process were made during the STRATAFORM project, and fl uid muds were again observed on the fl anks of the Po River delta during EuroSTRATAFORM.Th e model (b) indicates that these processes occur in shallower depths in the Adriatic Sea, where waves are smaller.Note the deposits are plotted with ~10x vertical exaggeration.Nittrouer et al. (this issue) show maps of the fi nal location and thickness of fl uid mud deposits after the Po fl ood of 2000 predicted from a two-dimensional model.
modeled waves in the Adriatic Sea on a rectangular grid with ~2-km spacing.Evolution of the non-stationary wave fi eld was calculated at 15-minute time steps, driven by wind fi elds provided at three-hour intervals and ~7-km spacing from the LAMI model output.Circulation and Suspended-Sediment Transport Oceanic circulation and sediment transport were simulated using ROMS (Regional Ocean Modeling System) (more information available http://marine.rutgers.edu/po/models/roms/).ROMS solves fi nite-difference approximations of the three-dimensional Reynolds-averaged equations for conservation of mass, momentum, and heat using a two-equation submodel for turbulent mixing.Effects of waves in the bottom boundary layer, which include increased drag and bottom stresses, were parameterized with a modifi ed version of the Grant and Madsen (1979) model.Sediment supplied by rivers or resuspended from the seafl oor was advected, vertically mixed, and allowed to settle in the model, which also records changes in upper stratigraphy (to ~0.1 m) of the seafl oor caused by erosion or deposition.Wind-driven circulation, mixing, and heating or cooling of surface waters were calculated using bulk fl ux algorithms with input from the atmospheric models.Initial water temperatures and salinities were set by interpolating from hydrographic data col-lected during September 2002 surveys.A range of model runs were performed on a ~4-km grid with 20 vertical levels using both atmospheric models and alternative parameterizations for various processes to explore the fate of sediment supplied by rivers or resuspended from existing deposits.Gravity Flows of Fluid Mud Gravity-fl ow processes on the Po River delta front were simulated with a one-dimensional (cross-shelf) time-dependent model that resolves the suspended sediment concentration in the thin wavecurrent boundary layer and calculates transport and deposition associated with gravity-driven downslope fl ows.The model was forced with observed waves and currents, and assessed for skill against near-bottom measurements of fl ow and suspended-sediment.A two-dimensional model (across shelf and along shelf) based on several simplifying assumptions was also used to predict the spatial distribution of the fl ood deposit on the Po delta (see Nittrouer et al., this issue).These models allow us to evaluate the displacement of fl ood deposits from their initial depocenters beneath the river plume to their eventual locus of long-term accumulation.Stratigraphy Stratigraphy was considered on two scales: detailed (upper 2 to 10 cm) and shallow (upper ~2 to 30 m).Two models, HSM-3D and CST (Coastal Sediment Tract), have been developed to allow diagnosis and predictions at these scales.HSM-3D is a three-dimensional, time-dependent numerical model for hydrodynamics, sediment dynamics, and morphodynamics.It includes suspended transport, bedload, and the effect of gravity fl ows in the bottom boundary layer and incorporates evolving bottom morphology caused by erosion or deposition.The model has variable grid spacing in the water column and fi xed bins for stratigraphy, which can extend several meters below the seafl oor.We embedded the HSM-3D grid in the ROMS domain to resolve length scales of ~0.1 km, while incorporating information about the larger scale mean circulation and wave fi elds.The CST model simulates sedimentation over decades and millennia by parameterizing sediment dynamics in terms of time-averaged and spatially averaged relationships that are consistent with long-term system forcing (Cowell et al., in press).The CST model represents a number of sedimentary systems including the river watershed, estuary, inlet, beach/surf zone, and shoreface/shelf systems (see Pratson et al., this issue).
. Wave fi elds and ocean heat content naturally integrate meteorological forcing; good agreement among measured and modeled ocean temperatures, waves, and currents in the Adriatic Sea model suggests the meteorological models were generally accurate.COAMPS/NCOM simulations demonstrate the advantages of two-way coupling, where more accurate water temperatures improve the parameterizations for atmospheric boundary layer structure and heat fl uxes (Figure 3).But the one-way LAMI/ROMS simulations seem to capture key aspects of the oceanic response to atmospheric forcing, and produced heat fl uxes and sea-surface temperatures that fell within the uncertainty of estimates derived from fi ve cloud-free satellite images (Figure 3b).Atmospheric forcing is the most important input to our sediment-transport models in the Adriatic Sea and an ongoing research topic.The wave climate in winter 2002 to 2003 was dominated by a few Sciroccos and a series of Bora wind events.Maximum modeled waves reached 5 m (signifi cant wave height) in open waters of the northeast Adriatic during a Scirocco that inundated Venice lagoon with the sixth highest aqua alta (storm surge) in the last 60 years (more information available at Centro Previsioni e segnalazioni Maree, Comune di Venezia; http://www.comune.venezia.it/maree/).Bora events were remarkable for their spatial structure (Figure4).Comparisons

Figure 3 .
Figure 3. (a) As cold Bora winds cross the Adriatic, they fl ow over surface waters that are ~10 degrees warmer.Th is induces vigorous air-sea exchange as the atmosphere extracts heat from the ocean.Th e vertical contour plot shows COAMPS/NCOM two-way coupled potential temperature at the beginning of a Bora.(b) Bulk heat transfer rates calculated by ROMS fall within error bounds around fl uxes estimated from satellite data, but LAMI algorithms overpredict rate of heat transfer.

Figure 4 .
Figure 4. (a) Wave data were recorded by pressure and velocity sensors on the EuroSTRATAFORM tripods, and by a network of wave-monitoring buoys maintained by the Italian Agency for Environmental Protection.(b) Th e SWAN model was forced with LAMI wind fi elds (black arrows) and produced fi elds of wave heights.Th e colors shown here are results for a January Bora event with red = 5 m signifi cant height.(c) Th e model results (black) usually compare favorably with measured wave heights (blue) and near-bottom orbital velocities (red).

Figure 5 .
Figure 5. (a) Bora winds form a distinct pattern across the northern Adriatic as they funnel through topography in Croatia.(b) Th e resulting circulation is complex, as shown here by simulated mean fl ow over four days at water depth of 5 m.Model results from ROMS generally correspond well with measurements from the ACE array of acoustic Doppler current meters (magenta arrows).(c)Drifter tracks simulated with ROMS using wave-enhanced near-surface mixing (solid green) match tracks measured during ACE (black) better than tracks simulated without wave-enhanced mixing (dashed green).
of SWAN model hindcasts with Italian buoy records indicate the model captures most of the features in the records of wave height.Modeled near-bottom wave orbital velocities also agree fairly well with PASTA tripod measurements, but there are some intriguing differences during some events (Figure 4) that may be related to incorrect wind forcing or problems with swell dissipation in SWAN (see Rodgers et al., 2003).Circulation, Sediment-Transport, and Stratigraphy Models Modeled circulation in the Adriatic agreed well with the extensive data col-lected in 2002 to 2003.The mean fl ow,

Figure 6 .
Figure6.Th e western Adriatic current causes southward water transport along the Italian coast.Model results (blue) match the timing and often the magnitude of transport estimated from an array of four acoustic Doppler current profi les on a transect near Senigallia (see Figure5b).
terns.Transport in the WACC calculated in the model simulation was compared with estimates made from four acoustic Doppler current profi les (ADCPs) along a transect off Senigallia (Figure 6).Early in the simulation, modeled transport tended to overestimate measured transport, but the model tended to underestimate the transport associated with pulses in November and early December.The early December pulse signaled the arrival of freshwater discharged during the Po River freshet, which peaked on November 30.Later in the season, transport calculated with the model and inferred from the measurements agreed more closely, and the modeled and measured mean transports over the nine-month period matched closely.Sediment transport is strongly nonlinear, however, and the impact of errors in circulation during specifi c events is under investigation.Model simulations of river-supplied sediment naturally indicated that rapidsettling material tends to accumulate near the river mouths (particularly the Po), and that unfl occulated material is transported farther along the coast and accumulates in elongated deposits near the shelf edge (Figure 7).The depositional pattern produced by a model simulation for winter 2002 to 2003 mimics aspects of the long-term depositional pattern observed in Holocene stratigraphy.Specifi cally, deposition on the Po delta, near the Apennine river mouths, and near the shelf edge north of the Gargano promontory occurs where late Holocene deposits are thickest.Typical results from the HSM-3D model showed depositional patterns that indicate alongshore transport produces coalescing clinoforms off the Apennine rivers.Larger-scale features that develop over millennia are predicted with the CST model (Pratson et al., this issue).The encouraging agreement suggests that the linked system of models incorporate the key mechanisms for transport and delivery of sediment from shelf environments to depocenters located farther south in deeper water.Work is ongoing to investigate the sensitivity of these patterns to winds, settling velocity, variations in erosion formulae, and other parameters.
winter conditions in the Adriatic showed that near the mouth of the Po River, rapid accumulation after fl oods produces ephemeral deposits of sediment that are subsequently remobilized by waves to form density fl ows.This was observed on three occasions in 2002 to 2003, and these fl uid-mud fl ows moved sediment from ephemeral shallow-water deposits beneath the river plume to deeper water.These results lead us to believe that fl uid mud processes may control the geometry of deltaic deposits; EuroSTRATAFORM researchers are developing models to calculate the resulting depth profi les.The mechanisms that move Po River sediment hundreds of kilometers to depocenters off the Gargano promontory must be associated with a larger-scale phenomena like the western Adriatic coastal current.The WACC is partially buoyancy driven, a forcing that increases during fl oods when suspended-sediment supply is highest.Observations and models demonstrate that Po water fl ows around the Gargano Promontory, but because the sediment aggregates and settles rapidly, it is unlikely that much remains in suspension for the entire jour-ney.It appears that Po sediments must travel southward in a series of episodic transport events, and that wave-induced resuspension is important in these hops.Bora winds generate large waves in the western Adriatic and enhance fl ow in the WACC, so it is reasonable to expect that correlated wave resuspension and stronger-than-average southward fl ow in the WACC combine to generate southward sediment fl ux.However, our data and model results show only a weak correlation between wave height and southward fl ow velocity in the WACC on the Chien-ti transect.The best correlation between WACC fl ow at Chienti and modeled wind is with wind near Trieste, nearly 200 km from the Chienti transect.This is evidence that wind-driven circulation in the northern Adriatic has a very large scale, and that waves are not necessarily collocated with Bora-enhanced southward fl ow.The model indicates that Scirocco winds can generate large waves that resuspend sediment without reversing fl ow in the WACC, suggesting that southward transport is likely to prevail regardless of the timing or mechanisms for resuspension.The system of models we have applied to the Adriatic Sea incorporates processes at scales ranging from hours and kilometers up to millennia and basin dimensions.Large collaborative fi eld efforts such as those in the Adriatic Sea during 2002 to 2003 are the only viable way to obtain enough synoptic data on waves, currents, and sediment to evaluate such complex models.Although research is ongoing to assess the skill of these models, results from the simulations are already helping us gain a better

Figure 7 .
Figure 7. (a) Results of sediment-transport simulations for nine months (November 2002 -May 2003) reveal patterns of deposition for (a) fast-settling aggregates (1 mm/s) and (b) slow-settling particles (0.1 mm/s) supplied by rivers.Th e depositional patterns are dominated by local deposition of material from the Po River in the northern Adriatic and by southward dispersal of Apennine sediment along much of the Italian coast.Deposition is enhanced north of Ancona and the Gargano promontory, similar to the pattern seen in the Holocene thicknesses.