Three sources of bathymetry were considered for the 2008 tests: an improved version of the Smith & Sandwell (version 9, 2007) topography, the 1 minute Smith & Sandwell v10.1 (2008) topography and the 1 minute NCOR bathymetry (courtesy of Dr. San Jan). The improved v9 Smith & Sandwell (2007) has somewhat overly smooth shelf and a questionable trench following the west coast of Taiwan. The newer version v10.1 Smith & Sandwell has numerous unverifiable small scale structures in the shallower regions. In our 2008 simulations, the NCOR bathymetry was selected as the starting point (in our older 2007 simulations, the improved version of the Smith & Sandwell (version 9, 2007) was used).

Improved S&S v9 (2007)	Smith & Sandwell v10.1 (2008)	NCOR

Part of our operating procedures is to condition the bathymetry to control computational errors over regions of high slope or high reduced slope. Below are plots of the shallow regions of the topography in our modeling domain at three stages of the conditioning process

1. Original: The NCOR topography has simply been interpolated to our modeling domain
2. Clipped: All depth shallower than 10m have been set to 10m to avoid singularities in our terrain-following coordinate system. Also a 2D median filter has been applied to primarily remove two grid-point features.
3. Smoothed: The final stage, after all conditioning procedures have been applied.

A research question that we are investigating now is which features do we need to maintain to produce realistic simulations. This includes physical (slope values, position of shelfbreaks, sill depths, etc) as well as numerical properties to be preserved.

Original	Clipped	Smoothed