Aspects of Error Quantification and Evaluation in Digital Elevation Models for Glacier Surfaces

Sammanfattning: Aspects of Error Quantification and Evaluation in Digital Elevation Models for Glacier Surfaces This study explores methods to quantify and evaluate error in digital elevation models (DEMs) built from remotely sensed elevation data of the Hintereisferner glacier. A special focus lies on glacier surfaces because glaciers are often inaccessible for field observations but at the same time prone to measurement errors. They would therefore particularly benefit from a comprehensive error assessment. One of the primary aspects of the study is to find suitable methods that also include the spatial distribution of error because this is generally a somewhat neglected aspect of quality assessment of DEMs, although it has a potentially significant impact on the way DEMs are used in research. Especially if geomorphological or topographical aspects are part of this. In addition to identifying and discussing methods to quantify and evaluate existing errors in DEMs, this study also looks at some of the major sources from which the errors stem to find out if the influence of these sources on the resulting errors can be estimated. To this end, two out of three major categories of error sources were selected: interpolation effects and spatial resolution effects. The explored methods include quantitative error measures, such as RMSE, but also more evaluative approaches, such as correlation analysis. Including spatial distribution as part of the quantification and evaluation of error in DEMs is done by exploring methods like the creation of error surfaces or approaches that quantify the spatial patterning of error values in DEMs, such as Moran’s I. The results show that when it comes to quantification and evaluation of error in DEMs, error surfaces, in combination with a mapped overview of the Local Moran’s I values, are presumably the most powerful methods to gain insight into both the absolute error values and the spatial distribution of them. Particularly spatial outlier detection is a useful part of this approach.