Glaciers play an important role in sea level predictions and are an important supplier of fresh water. Understanding the physics and dynamics of glaciers is important in local and global climate predictions and predictions of fresh water supplies. The Himalayas are the biggest storage of fresh water outside the polar regions with many different local climates and glaciers. One important glacier parameter used in dynamic studies is the flow velocity, as the flow velocity is used as a boundary condition in mass balance and run-off models. The flow velocity of a glacier is relatively easy to measure on a large scale using satellite missions and the increased coverage of satellite missions provides the means to large scale velocity monitoring. Robust methods for measuring large scale seasonal and long term velocity dynamics of glaciers, however, remain an elusive goal.

The optical Landsat mission has a long history of monitoring and is thus useful for long term flow velocity analysis. Feature tracking algorithms applied on these Landsat images provide means of automatically calculating large scale velocities. Automated approaches for large scale analysis are difficult because of the lack of validation and good filtering techniques to deal with shadows, surface changes and clouds. Large scale temporal analysis is even harder because the errors arising in the flow velocity calculations are often large compared to the velocities due to the short periods between the images from which the velocities are calculated.

This research proposes, implements and tests a new method for automatically creating large scale velocity time series using the optical Landsat database and feature tracking in the Himalayas. Where normally velocity time series consists of consecutive single velocity fields, the novel method uses combinations of velocities to estimate these single velocities. This method is tested against results from single velocity fields for the Everest region and the Karakoram region. A sensitivity and parameter analysis provides the best parameter settings for the new method.

The result is a novel method that provides validation, robustness and acts as a filter for erroneous velocities. When possible, the new method increases or retains the number of results while increasing the precision. Furthermore, when the number of results is lower, it is due to filtering of erroneous velocities. The new method also provides an error indication which could be of use in future research. The main sources of errors: geo-location and precision of the feature tracking algorithm, are shown to have a large effect on the results and should be as small as possible. The magnitude of these errors make it difficult to measure seasonal changes in flow velocity for slow moving glaciers. Furthermore, Landsat 4-5 TM results are shown not to be useful for dense time series as the geo-location accuracy and artifacts from the TM scanner have a large effect on the precision of these results. The Landsat 8 velocity results, however, show good similarity to results from other researches and show promise for future research.

The sensitivity analysis of this research was incomplete as the complete algorithm was not optimized and future developments might focus on improving this. Also, it was found that the proposed method was very time intensive. A correct choice for the method of calculating flow velocities might drastically decrease the computational effort and improve applicability to large area research. In total, the new method is not only useful for use on the Landsat database, but could be used to improve any optical or radar flow velocity time series.