Should we use a simple or complex model for moisture recycling and atmospheric moisture tracking?

Abstract. This paper compares state-of-the-art atmospheric moisture tracking models. Such models are typically used to study the water component of coupled land and atmosphere models, in particular quantifying moisture recycling and the source-sink relations between evaporation and precipitation. There are several atmospheric moisture tracking methods in use. However, depending on the level of aggregation, the assumptions made and the level of detail, the performance of these methods may differ substantially. In this paper, we compare three methods. The RCM-tag method uses highly accurate 3-D water tracking (including phase transitions) directly within a regional climate model (online), while the other two methods (WAM and 3D-T) use a posteriori (offline) water vapour tracking. The original version of WAM is a single-layer model, while 3D-T is a multi-layer model, but both make use the "well-mixed" assumption for evaporation and precipitation. The a posteriori models are faster and more flexible, but less accurate than online moisture tracking with RCM-tag. In order to evaluate the accuracy of the a posteriori models, we tagged evaporated water from Lake Volta in West Africa and traced it to where it precipitates. It is found that the strong wind shear in West Africa is the main cause of errors in the a posteriori models. The number of vertical layers and the initial release height of tagged water in the model are found to have the most significant influences on the results. With this knowledge small improvements have been made to the a posteriori models. It appeared that expanding WAM to a 2-layer model, or a lower release height in 3D-T, led to significantly better results. Finally, we introduced a simple metric to assess wind shear globally and give recommendations about when to use which model. The "best" method, however, very much depends on the research question, the spatial extent under investigation, as well as the available computational power.