Surrogate-based uncertainty quantification and parameter optimization in simulations of the West African monsoon.

Format:

Book

Author/Creator:

Fischer, Matthias, author.

Series:

Schriftenreihe des Instituts für Technische Mechanik Band 41.

Language:

English

Subjects (All):

Uncertainty (Information theory)--Mathematical models.

Uncertainty (Information theory).

Weather forecasting--Mathematical models.

Weather forecasting.

Climatic changes--Mathematical models.

Climatic changes.

Gaussian processes--Mathematical models.

Gaussian processes.

Genre:

Academic theses.

Physical Description:

1 online resource

Place of Publication:

Karlsruhe : KIT Scientific Publishing, 2025.

Summary:

The West African monsoon (WAM) is a key climatic system with far-reaching impacts, and its complex interactions pose a challenge for weather and climate models. Uncertainties in model parameterizations, for example, in deep convection or cloud micro-physics, significantly affect forecast accuracy. This dissertation presents a surrogate-based approach to quantify these uncertainties and systematically improve model parameters. To achieve this, a framework is developed that first transforms the parameter space into a uniformly distributed input space, where sampling techniques and surrogate methods are applied. Results from simulations with the IcosahedralNon-Hydrostatic (ICON) model of the German Meteorological Service are analyzed using Gaussian process regression and principal component regression to reduce computational costs. The influence of model parameters is examined through global sensitivity analyses and parameter studies, followed by optimizations incorporating meteorological reference data. The results show that among the investigated parameters, the entrainment rate, which governs the mixing between rising air and the surrounding atmosphere, the terminal fall velocity of ice particles and the evaporative soil surface fraction, which controls soil moisture and evaporation, have the greatest impact on the WAM system. Reducing the entrainment rate leads to amore accurate simulation of precipitation, near-surface humidity and mean sea-levelpressure. However, improvements across all variables remain limited, suggesting thatthe default configuration of the ICON model is already well-calibrated. The findingshighlight structural challenges, such as the limited ability to achieve desired spatialpattern changes through parameter adjustments and the trade-offs between differentmeteorological variables. This underscores the need to refine not only parameter valuesbut also model physics and spatial resolution.V

Contents:

Danksagung Kurzfassung Abstract Introduction Motivation State of Research Objectives and Structure Fundamentals Fundamental concepts of probability theory Sources of uncertainty Probability distributions$ $t Probability integral transformation Sampling techniques Gaussian process regression Simple kriging Universal kriging Principal component regression Principal component analysis Regression model Model Validation Global sensitivity analysis.

Notes:

Includes bibliographical references (pages) and index.

Part of the metadata in this record was created with the help of AI Metadata Assistant