Optimisation of the simulation particle number in a Lagrangian ice microphysical model
This paper presents various techniques to speed up the Lagrangian ice microphysics code EULAG-LCM. The amount of CPU time (and also memory and storage data) depends heavily on the number of simulation ice particles (SIPs) used to represent the bulk of real ice crystals. It was found that the various microphysical processes require different numbers of SIPs to reach statistical convergence (in a sense that a further increase of the SIP number does not systematically change the physical outcome of a cirrus simulation). Whereas deposition/sublimation and sedimentation require only a moderate number of SIPs, the (nonlinear) ice nucleation process is only well represented, when a large number of SIPs is generated. We introduced a new stochastic nucleation implementation which mimics the stochastic nature of nucleation and greatly reduces numerical sensitivities. Furthermore several strategies (SIP merging and splitting) are presented which flexibly adjust and reduce the number of SIPs. These efficiency measures reduce the computational costs of present cirrus studies and allow extending the temporal and spatial scales of upcoming studies.