Classification of Fast Ice under Uncertainty

Abstract

This thesis covers a stochastic approach on fast ice detection on simulated satellite observations. The main focus lies on the usage of the time dependence of the data points, and how to correct defective and noisy inputs. Monte Carlo simulation is the method of choice, utilizing its versatility. Another approach briefly looks at linear stochastic programming and its similarities to a weighted average in the context of the problem. The numerical comparison shows the predominant performance of the Monte Carlo approach with an average of over 86.5% accuracy in detecting the fast ice, at least on the simulated data.