Changes of the ice sheets in Greenland and Antarctica have a significant effect on the global sea level. In order to predict future climate scenarios, it is essential to understand ice sheet processes and therefore to observe current changes. As ice sheet changes are strongly influenced by the dynamics of their outlet glaciers, an accurate and continuous monitoring of these outlet glaciers is of paramount importance. One outlet glacier process that has been identified as particularly important in this context is the change in calving front position. Consequently, calving front change data is of major importance. However, due to manual delineation, most current data sets are limited in either spatial coverage or temporal resolution. This cumulative dissertation addresses this problem by applying deep learning to multispectral Landsat imagery in order to achieve accurate, robust and automated delineation of outlet glacier fronts. Throughout three publications, the developed processing system is presented, assessed and applied. The workflow is based on a convolutional neural network that is trained and tested using manually delineated calving front positions. The mean delineation error ranges from about 45 m to 75 m depending on the test data set. Thus, most automatically extracted glacier fronts are indistinguishable from manually digitised ones. Furthermore, the benefit of multispectral, topographic, and textural input features is assessed using a dropped-variable approach. The resulting feature importance assessment emphasises the utilization of multispectral inputs as they lead to more accurate predictions, especially under challenging situations such as difficult ice mélange conditions. Eventually, this processing system is applied to outlet glaciers in Greenland and on the Antarctic Peninsula. The resulting Greenland data product contains 9243 calving front positions across 23 glaciers for the period 2013 to 2021 and the Antarctic Peninsula data product contains 4817 calving front positions across 42 glaciers for the period 2013 to 2023. Unlike most existing manually delineated products, these data sets reliably combine high temporal resolution with large spatial coverage. Derived time series resolve not only long-term and seasonal signals, but also sub-seasonal patterns. These data products will help to better understand the dynamics of outlet glaciers at intra-annual scales, study ice-ocean interactions in more detail, and constrain glacier models. In future, comparable deep-learning based workflows are likely to become integral to the monitoring of glacier calving fronts. This thesis contributed to establishing a new link between deep learning, remote sensing and glaciology – a link that will persist in the long term.