As a preparation of the discretisation, the physical space is vertically divided into layers. This is done by introducing internal surfaces , which do not intersect, each depending on the horizontal position and time . Let
define the local layer depths with
for . For simplicity, the argument is omitted in most of the cases.
The most simple layer distribution is given by the socalled transformation (see Phillips (1957) for a first application in meteorology and Freeman et al. (1972) for a first application in hydrodynamics) with
and
for .
The coordinates can also be refined towards the surface and the bed:
such that levels are obtained as follows:
(17) 
The grid is refined towards the surface for and refined towards the bottom for . When both, and are larger than zero, then refinement towards surface and bed is obtained. For the transformation (14) with is retained. Figure 1 shows four examples for vertical layer distributions obtained with the transformation.
Due to the fact that all layer thicknesses are proportional to the water depth, the equidistant and also the nonequidistant transformations, (14) and (16), have however one striking disadvantage. In order to sufficiently resolve the mixed layer also in deep water, many layers have to be located near the surface. The same holds for the bottom boundary layer. This problem of coordinates has been discussed by several authors (see e.g. Deleersnijder and Ruddick (1992), de Kok (1992), Gerdes (1993), Song and Haidvogel (1994), Burchard and Petersen (1997)) who suggested methods for generalised vertical coordinates not resulting in layer thicknesses not proportional to the water depth.
The generalised vertical coordinate introduced here is a generalisation of the socalled mixedlayer transformation suggested by Burchard and Petersen (1997). It is a hybrid coordinate which interpolates between the equidistant and the nonequidistant transformations given by (14) and (16). The weight for the interpolation depends on the ratio of a critical water depth (below which equidistant coordinates are used) and the actual water depth:
with
For inserting in (19) and and in (16), the mixed layer transformation of Burchard and Petersen (1997) is retained, see the upper two panels in figure 2. Depending on the values for and , some nearsurface layer thicknesses will be constant in time and space, allowing for a good vertical resolution in the surface mixed layer.
The same is obtained for the bottom with the following settings: , and , see the lower two panels in figure 2. This is recommended for reproducing sedimentation dynamics and other benthic processes. For and or a number of layers near the surface and near the bottom can be fixed to constant thickness. Intermediate states are obtained by intermediate settings, see figure 3. Some pathological settings are also possible, such as , and , see figure 4.




The strong potential of the general vertical coordinates concept is the extendibility towards vertically adaptive grids. Since the layers may be redistributed after every baroclinic time step, one could adapt the coordinate distribution to the internal dynamics of the flow. One could for example concentrate more layers at vertical locations of high stratification and shear, or force certain layer interfaces towards certain isopycnals, or approximate Lagrangian vertical coordinates by minimising the vertical advection through layer interfaces. The advantages of this concept have recently been demonstrated for onedimensional water columns by Burchard and Beckers (2004). The threedimensional generalisation of this concept of adaptive grids for GETM is currently under development.