The potential of UAV photogrammetry for hydro-glaciological forecasts

Abstract

In recent decades, glaciers worldwide have lost a significant part of their volume due to increasing atmospheric temperatures. In the future, these large storages of fresh water resources will continue to shrink. Glacio-hydrological models predict that glacier retreat will first induce an increased release of melt water until a certain peak is reached, followed by a decrease in runoff due to a reduction in glacier size and volume. For some glaciers, this peak has already occurred. By 2100, many regions of the world will likely lose more than 75% of their current glacier volume. Declining glacier volumes, in turn, will lead to a smaller contributions of glacier meltwater to streamflow, resulting in direct consequences for agriculture, drinking water and hydropower for instance. The anticipated trend in future glacier runoff is very similar across glacio-hydrological models. However, different sources of errors in the models themselves and in the model input, lead to large uncertainties in the predictions. This thesis, composed of three main studies, examines the uncertainties linked to the models’ input, namely the meteorological forecasts and the glacier-related datasets (i.e. Digital Surface Models (DSMs), ice thickness or snow height distribution). The ultimate goal is to help improving the next generation of runoff forecasts. The first study is related to the model meteorological inputs. The aim was to assess how the skill (i.e. the quality) of temperature and precipitation forecasts are influencing the skill of the resulting runoff predictions. A synthetic experiment was set up in which meteorological forecasts ranging over a spectrum of different skills were created. The meteorological forecasts were fed into a glacio-hydrological model, and the skill of the resulting runoff forecasts was assessed. The same experiment was performed for catchments with different degrees of glacierization. The results show that temperature and precipitation have an influence on runoff that depends on the catchment’s degree of glacierization, with the importance of accurate precipitation forecasts increasing at the expenses of accurate temperature projections with a decreasing degree of glacierization. The second and third studies are related to the models' glacier-related inputs. The aim was to investigate the potential of Unmanned Aerial Vehicle (UAV) photogrammetry to derive datasets including high-resolution ortho-images, DSMs of the glacier surface, and ice-flow velocity fields. First, the focus was set on assessing the accuracy of the DSMs, and how the number of Ground Control Points (GCPs), used to geo-reference the DSMs, affects them. To answer this question, several UAV surveys were performed on three alpine glaciers during different seasons. The results show that not only the number but also the distribution of GCPs affects the accuracy of UAV-derived DSMs. Once the recipe to derive accurate datasets was found, the ortho-images and DSMs were used to derive surface velocity fields, which is the subject of the third study. The goal was to develop a new procedure that derives (a) robust surface displacements and (b) an estimate of the result’s accuracy. The procedure consists of generating many different surface displacements based on various filters and matching functions, and to stack the so-obtained ensemble of results into a final surface velocity field. This procedure was embedded in a new tool that works in a semi-automated way. The tool was tested on three different types of glaciers, namely a calving glacier, an alpine glacier and a rock glacier. The experiments showed that it successfully generated robust velocity fields for all test sites and that the vector fields cover a large spatial extent. This thesis demonstrates that in high mountain areas, UAV photogrammetry allows to obtain high-resolution datasets with centimeter to decimeter accuracy. The potential of UAV photogrammetry to derive insightful snow and glacier-related datasets is high, but the method is currently limited to small catchments of several square kilometers, due to short flying time and geo-referencing requirements. Technical advancements and innovation in the field are fast, and UAV photogrammetry might become a valuable tool for surveying larger areas in the near future.