Automatic detection of buried channel deposits using high resolution laser altimetry data (FLI-MAP)

Abstract

The formation of the current Rhine-Meuse delta has mainly taken place during the Holocene (the last 12,000 years). This period is characterised by avulsions that lead to sudden shifts in the location of river channels. The channel deposits left behind by abandoned rivers are often distinct sandy layers in the shallow sub-surface, called buried channel deposits. Knowledge of the location of buried channel deposits is interesting from a historical point of view and is essential for planning, constructing and maintaining structures that intersect these channels. Traditionally, buried channel deposits are traced using labour intensive soil drillings and visual interpretation. However, some buried channel deposits can also be detected by using local elevation differences (1 - 2 m) caused by differential compaction. Automating the detection process and using Digital Elevation Models could result in faster processing and a higher detailed, more objective map. This leads to the following research question: is it possible to automatically derive a detailed map of buried channel deposits from high resolution laser altimetry data? The dataset used for this purpose is a rasterized elevation model measured by the FLI-MAP system with the same specifications as AHN-2. First, two filtering methods are applied to remove infrastructure and other objects using existing topographical data (GBKN) and terrain characteristics (variability and density). If these objects are not removed they cause errors in the final result. In the detection procedure four structural attributes are calculated using the elevation dataset: slope, curvature, relative elevation (TPI) and smoothed TPI. This is done to use as much of the available information in the detection procedure. With these attributes a multi-band image is formed. Classification is then performed using a Maximum Likelihood classifier were each point is assigned to a predefined class. A majority filter is applied on the result and empty areas are interpolated using a conditional dilation to get a full coverage of the area. Processing speed is increased by downsampling datapatches that are used for computational intensive algorithms. Additional automation of the detection process is performed by developing a method to resolve edge effects on tiled datasets. Validation is performed by comparing the detection result to two independent datasets. A palaeogeographic map is used to make a numerical and visual comparison. An analysis using shallow drilling measurements is performed to determine the depth of the detected sand layers. The results indicate that buried channel deposits (between 4,620 and 1,700 years old) can be mapped roughly in a fully automatic way. The probability that these channels are within a certain depth can be calculated. Validation shows that this new method is strongly hampered due to human intervention in the landscape. Further research is recommended to improve the filtering and detection method.