ERRORS NEAR THE POLES GENERATED BY A SEMI-LAGRANGIAN INTEGRATION SCHEME IN A GLOBAL SPECTRAL MODEL

When a semi‐Lagrangian integration scheme was introduced into a global spectral model by Ritchie (1988), large errors developed in the neighbourhood of the poles. It took approximately 6 months of effort to diagnose the problem and find an appropriate correction. The method that was used to diagnose the source of error was quite tedious, but it was successful. Furthermore, it seems that this method could be used to diagnose other sources of error that occasionally show up in numerical integrations. For this reason, it was felt that this method should be described in a separate article. This is the main objective of the presentation that follows.An integration is carried out with the original version of the model and some results are presented in order to illustrate these errors. In order to identify their exact cause, the model is stripped down in two steps. At each step, some checks are made to ensure that the errors are stillpresent in the degraded version of the model. In the end, the remaining equations are sufficiently simple to ensure that the cause of the errors becomes obvious. The diagnosis immediately suggests some alternative computational designs. A modification that completely eliminates these errors is then proposed and tested. An integration with the modified spectral model is carried out and results are presented to show that the errors have disappeared. © 1990 Taylor & Francis Group, LLC.