Data assimilation of observed cloud fields in LES model

Abstract

This thesis investigates the implementation of three-dimensional nudging into large-eddy simulation (LES) to assimilate observed atmospheric data into an LES model. 3D-nudging 'pushes' the thermodynamic fields in a simulation towards the desired observed fields. The aim is to test if such a method is useful in improving solar forecasts of stratocumulus-topped boundary layers. For this purpose 3D-nudging LES solar forecasts are compared to persistence forecasts and conventional LES-based forecasts. As a proxy for observations, exact thermodynamic fields from LES were used in this research. Using LES fields is advantageous as it provides full 3D thermodynamic fields but also dynamic fields for checking the turbulence in the different methods. Results show that 3D-nudging is quite capable of replicating the desired thermodynamic fields. Unfortunately, nudging comes with a penalty as it causes the turbulence built up in a simulation to be flawed. This effect is mitigated by the design of variations on the nudging technique, the most promising of which is multiple time fields nudging, which nudges the thermodynamic fields in a simulation to subsequent desired fields every 10 minutes during the nudging period. Solar forecasts found by this method are found to be more accurate than the persistence and regular LES methods on forecast horizons of 30 minutes and larger. Approaches proposed in this study to approximate thermodynamic fields from observational data estimate thermodynamic fields to a reasonable accuracy but are far from perfect, and thus it should be noted that solar forecast accuracy of the discussed methods will be less accurate when applied to real observations. Further research is recommended to focus on the use of the 3D-nudging methods in more LES case studies, and on devising better methods for the estimation of thermodynamic fields from observations.