Improving an Atmosphere General Circulation model through Parameter Optimization

Abstract

This thesis presents an implementation and evaluation of two different multivariate data assimilation techniques for the optimization of parameters of a global primitive equation Atmospheric General Circulation Model (AGCM), the Planet Simulator (PlaSim). The hypothesis used is that the source of uncertainty in the model is related to parameters from the cloud parameterizations, vertical and horizontal dif- fusion time scales in the model. The results are evaluated by comparing basic physi- cal state variables of the atmosphere such as surface temperature, precipitation, net heat flux, winds and sea level pressure predicted by the model with observations. Initially, sensitivity analysis of PlaSim with respect to various parameters used in its different parameterizations is carried out. The variation of the cost function with respect to changes in each control parameter is studied and the most sensitive parameters are identified. The results of the sensitivity analysis serve as a guideline for identifying sensitive model parameters optimization procedures. Green’s function (GF) method of parameter optimization is applied on two differ- ent model configurations, with and without moisture related processes (wet and dry configurations, respectively) in an identical twin model framework. The results are inter-compared with existing results from 4D-variational (4D-var) assimilation scheme in PlaSim. GF procedure successfully estimates model parameters for both shorter (30 days) and longer time (365 days) scales using 3, 5 and 6 control param- eters. However, when using real world observations, the GF method is unable to minimize cost function even using a single control parameter. Another optimization procedure based on stochastic approximation, the simulta- neous perturbation stochastic approximation (SPSA) method is implemented in PlaSim and the results are discussed. The advantage of using SPSA method is its ease of implementation and its robustness to noise in cost function. In identical twin experimental framework SPSA method reliably recovers the control parame- ters. When real observations are used, the errors in optimized state, for example in surface temperature and net heat flux are reduced by 16% and 30% respectively. In addition, the optimized state of PlaSim shows improvement in sea level pressure, zonal winds at 500 hPa and surface precipitation. This study demonstrates the use- fulness of a simple data assimilation scheme in a highly non-linear chaotic system and its potential application in tuning of the climate models.