Time stepping
If we combine all the terms that must be evaluated at time step $n + \frac{1}{2}$ into a variable $G$, then we have
where $\bm{G}_{\bm{u}} = (G_u, G_v, G_w)$. Together with \eqref{eq:projection-step} allows us to write the discretized momentum equation as
where we have brought back the hydrostatic pressure anomaly $\phi_{HY}^\prime$ and non-hydrostatic pressure $\phi_{NH}$.
Doing the same for tracer quantities yields
and
We evaluate the $G^{n + \frac{1}{2}}$ terms explicitly using a weighted two-step Adams–Bashforth (AB2) method
AB2 has the advantage of being quasi-second-order accurate in time and yet does not have a computational mode (???). Furthermore, it can be implemented by evaluating the source terms $G$ only once and storing them for use on the next time step, thus using less memory than higher-order time stepping schemes.
It turns out that for a second-order accurate approximation of $G^{n+\frac{1}{2}}$ we require $\chi = \frac{1}{8}$ \citep{Ascher95}. Note that $\chi = 0$ reproduces the unweighted Adams-Bashforth method which calculates a second-order accurate approximation of $G^{n+1}$. Also note that $\chi = -\frac{1}{2}$ reproduces the first-order accurate forward Euler method, useful for initializing the model when $G^{n-1}$ is not available, such as at the first time step.