In wind-driven coastal upwelling, the alongshore variation of the shelf topography largely constrains the orientation of the alongshore flow and geostrophically modulates the intensity of the cross-shelf transport [Gan and Allen, 2002]. The cross-shelf gradient of isobaths over the continental shelf, on the other hand, controls the intensity of the upwelling jet and the ensuing cross-shelf bottom frictional transport [Allen et al., 1995; Weisberg et al., 2005]. The upward motion over the nearshore waters in the upwelling is regulated not only by the characteristics of the topography over the inner shelf, but also by those over the middle and outer shelves [Weisberg et al., 2005]. The variable shelf topography, in both along-shelf and across-shelf directions, markedly controls the shoreward advection of denser deep water by virtue of the bottom Ekman transport and geostrophy. The combined topographic influence from nearshore and farther offshore regions is expected to interact each other and control the path and intensity of the coastal current, the shoreward transport and the associated correlation scale of the cross-shelf flow over the continental shelf. The forcing process induced by the shelf topography and exerted on the wind-driven upwelling circulation, particularly on the cross-shelf circulation that transports large gradients of physical and tracer properties upslope over a sufficiently wide shelf with variable topography, remains poorly understood. This study reports on the variations in coastal upwelling as a result of control by the shelf topography over the broad continental shelf in the northeastern South China Sea (NSCS) (Figure 1).
 The shelf topography in the NSCS is characterized by the complex coastline variation in the nearshore region and by the existence of a prominent eastward widened shelf (referred to as the widened shelf hereafter) formed by an abrupt offshore extension of isobaths east of the Pearl River Estuary and bounded by the 50 m isobath at its southern edge (Figure 1). A shallow bank, the Taiwan Shoals, is located between the 50 m isobath in the south and the 30 m isobath in the north at the eastern end of the widened shelf. Prominent features in the widened shelf includes the embedded capes along the coastline and shoreward convex isobaths that converge near the head of the widened shelf about half a degree southwest of Shanwei.
 Under the influence of the East Asian monsoon, southwesterly upwelling favorable winds prevail over the NSCS during the summer in June, July and August [Li, 1993]. As a result, coastal upwelling is generated along the coast of the NSCS in which cold and nutrient-rich deep waters upwell to the surface layer over the broad shelf. The upwelling in the NSCS has a strong spatial variation. The intensified upwelling (IU) is found over the widened shelf (Figures 2 and 4) with a maximum intensity at the lee of a coastal cape near Shantou (116.7°E (Figures 1 and 2)). The spatial structure of the upwelling appears to be largely correlated with the topographic characteristics of the widened shelf. Following early findings on upwelling in the NSCS [Wyrtki, 1961], strong upwelling centers have been identified either in Shantou by Li  or at relatively less frequent occurrences, in Shanwei (115.5°E) by Zeng  and Han and Ma  from many hydrographic measurements. However, there have been few investigations that describe the basic process and the forcing mechanism of the IU in the NSCS. In particular, the effect from the unique NSCS shelf topography on the upwelling circulation has not been explored.
 The intensity of the coastal upwelling is governed by the strength of the wind-driven surface Ekman transport and the shoreward return currents in the interior and bottom boundary layer in a cross-shelf, two-dimensional view. For a given wind forcing, the two-dimensional upwelling is characterized to a great extent by the cross-shelf topography. In a cross-shelf, two-dimensional modeling study, Allen et al.  showed that the coastal jet in an upwelling at a given time is stronger over a steep shelf, while the frictional bottom boundary layer develops more strongly and carries a larger fraction of the onshore Ekman transport over a wider shelf. A three-dimensional upwelling flow field can deviate substantially from the two-dimensional flow as a result of variable forcing in the alongshore direction. Besides an alongshore variation in wind forcing, the alongshore variation in the pressure gradient induced by the shelf topography [e.g., Lentz et al., 1999; Weisberg et al., 2005] plays a key role in forming the alongshore variation in the upwelling in many coastal regions around the world.
 Over a sufficiently wide shelf like in NSCS, wind-forced coastal circulation varies across the shelf with distinctive regions defined as inner, middle and outer shelf as a result of variable dynamical forcing regimes. The inner shelf is the nearshore waters of overlapping surface and bottom Ekman layers [Mitchum and Clarke, 1986; Lentz, 1995], or inshore of the upwelling front where weakly stratified or unstratified waters are located [Austin and Lentz, 2002]. A more comprehensive definition for the inner, middle and outer shelf was given by Li and Weisberg [1999a, 1999b] and Weisberg et al. , in which the inner shelf is the transition region between the Ekman and Ekman geostrophic balances where the surface slope results from mass adjustments through overlapping surface and the bottom Ekman layer, the outer shelf is the region that extends a baroclinic Rossby radius of deformation landward from the shelf break and the middle shelf is located between them. Weisberg et al.  also pointed out that the inner shelf can include a geostrophic interior region so long as the surface and bottom Ekman layers are connected via divergence.
 The upwelling over the nearshore water is controlled by the interaction of the wind-driven cross-shelf exchange over the inner, middle and/or outer shelves. The intensity of the interaction, besides other controlling factors like stratification and wind forcing, is governed by the nature of the shelf topography. In the alongshore direction, Gan and Allen  found that a stronger upwelling occurred in the lee of a coastal promontory along the coast of California, where the cross-shelf circulation is geostrophically strengthened by a locally amplified alongshore pressure gradient. A similar response extends to the middle shelf as the scale of the topography characteristic is expanded beyond the inner shelf [Song et al., 2001]. In the cross-shelf direction, Janowitz and Pietrafesa  showed that the diverging shelf isobaths produce a variation in the vorticity and leads to the formation of a vertical vorticity as well as a cross-isobath flow. On the other hand, the accelerated along-shelf flow over the continental shelf with converging isobaths leads to an increase in the frictional bottom transport favorable to the upwelling [Oke and Middleton, 2000; Weisberg et al., 2000]. Pringle  used a barotropic, linear and steady potential vorticity equation, showed that the onshore transport of deep waters, primarily in the bottom layer enhances in the place where the shelf narrows down wave during upwelling. In the recent Coastal Ocean Advances in Shelf Transport (COAST) program, Barth et al. , Kosro  and Gan and Allen [2005a] showed that the intensity and the pattern of the upwelling circulation are greatly regulated by the existence of a coastal bank embedded over a relatively uniform shelf off Oregon. They found that stronger cooling was introduced by the intensified cross-shelf circulation and heat advection over the coastal bank. In spite of the obvious correlation between the coastal circulation and topographic forcing, a dynamical rationalization remains unclear. In particular, the processes and dynamics involved in the interactive response of wind-driven, cross-shelf transport over the inner and middle shelves to their respective topographic forcing are poorly understood.
 In this study, the hydrographic measurements in the NSCS during the upwelling season are analyzed to elucidate the characteristics of the IU process in the NSCS (section 2). A three-dimensional model is then employed to simulate the observed processes (section 3) and to dynamically interpolate the governing forcing mechanism (section 4).