|
Hi Billy, We're run into a similar problem making GNOME a grid independent trajectory model. http://response.restoration.noaa.gov/software/gnome/gnome.html Our idea has been to make GNOME work with any circulation model we come in contact with so that we can leverage the best forecast model available during a response regardless of whether it is rectangular, curvilinear or finite element. We've been working with NOAA/CSDL for create a description of netCDF files that will work with different grid models. Our solution has been to interpolate all the velocities to one point on the grid (corner or center) in easting and northing coordinates (rather than grid relative). Although not perfect, we've found them functional. As you mentioned, the narrow channel is a problem. From our experience we would recommend either interpolating all four velocities onto the center of the grid cell (works best for narrow channels), or taking the two nearest sides to a corner and moving them to that corner (gotta trust your grid). The latter may seem really odd, but the problem of no velocity through the boundary really isn't such a bad problem, as you end up with currents that parallel the shoreline as they should. You can follow some of the discussion from Tom Gross' web page on the HYDRONetCDF web page. http://ccmp.chesapeake.org/HYDRONetCDF/HYDRONetCDF.html Best Regards, CJ William S Kessler wrote: This is only partly a Ferret question, but I suspect many Ferret users have thought about how to deal with this problem. I have OGCM output on a C-grid. The temperature and velocity components are on a staggered grid structure. A single gridbox looks like: v u T u v where the u-values are the currents in/out the east-west sides of the box, and the v-values out the north-south, and T is the temperature value defined in the center of the box. All these are on the same vertical level. I want to make a vector plot of the velocity components, but since the u's and v's are not at the same location this is not straightforward. One possibility is to define a grid consisting of all the u and v points together, and interpolate all the velocities to all the points that exist on either grid. Another is to average the u's in x and v's in y onto the temperature gridpoint. A third is to do no interpolation, but treat the upper v and right-side u as the flow at the northeast corner of the box, and the lower v and left-side u as the flow at the southwest corner. The problem becomes particularly acute when I want to make vectors in a region of complicated land masses, with narrow straits, but I want the plot to accurately reflect the model's actual flow, and not to omit any gridpoints. All of the above ideas result in losing some values, because, for example, there is no left-side u-value on gridpoints against a western boundary, while the right-side u and upper and lower v's do exist in this case. Anybody have a good idea how to do this as correctly as possible? Billy K --
CJ Beegle-Krause, Ph.D.
NOAA/NOS/ORR/Hazmat
7600 Sand Point Way NE
Seattle, WA 98115
voice: (206) 526-6961
fax: (206) 526-6329
<:}}}}}><
<:}}}}}><
\!/ \!/ \!/ >^<**>^< \!/ \!/ \!/
|