Semi-automated detection and delineation of earthflows in New Zealand using remote sensing - challenges and opportunities

Earthflows are complex landslide phenomena that can occur on gentle to moderate slopes in plastic, mixed, and disturbed earth with significant internal deformation. They exhibit a wide range of sizes (from tens of metres to kilometres in length) and can form complexes with slowly deforming bodies or fails along multiple shear surfaces, resulting in a lobate flow-like morphology. While they can show different movement rates, typical earthflows move slowly and intermittently with active and inactive states, whereby velocities are usually measured in meters per year. They mainly occur under saturated conditions, and trigger factors include prolonged or intense rainfall or snowmelt, stream erosion at the bottom of a slope, or the lowering of adjacent water surfaces and the related drawdown of the groundwater table. Earthflows can cause damage to infrastructure, affect the productivity of farmland, potentially dam rivers with subsequent flooding upstream, pose a risk to downstream areas, and impact water quality due to sediment input to streams.

Earthflows are usually mapped manually using orthophotos, but the quality of existing inventories differs significantly. Owing to their complexity, the semi-automated detection and delineation of earthflows is highly challenging. Boundaries are generally transitional rather than discrete, and a range of factors influence the internal homogeneity of the landslide body, such as topographic relief, landform properties, and scale. Terrain and topographic characteristics of earthflows, such as small scarps, hummocks, and flow lobe shadows, are difficult to discern based only on optical imagery; thus, the integration of high-resolution topographic data in the recognition process is important. While a human interpreter can use such specific topographic characteristics, implementing the required expert knowledge into automated mapping approaches based on remote sensing data is challenging.

In this study, we addressed these challenges and aimed to semi-automatically detect and map earthflows in the Tiraumea catchment, which is an upper catchment of the Manawatū catchment located in the Manawatū-Whanganui region of the North Island of New Zealand, using aerial photography and a photogrammetrically derived high-resolution digital surface model (DSM) within an object-based image analysis (OBIA) framework. A flexible segmentation approach was followed, creating different sizes of connected image objects at different hierarchical segmentation levels, whereby the earthflow boundaries were stepwise adapted and refined. Statistics derived from a range of terrain derivatives informed the selection of the most suitable derivatives for knowledge-based classification, which relied on specific earthflow characteristics, such as the connection to streams and the existence of bare ground, rushes, and surface water. The results show that the automated delineation of earthflow bodies is particularly difficult and requires further improvement. However, the mapping outcomes indicate potentially unknown earthflow locations that should be confirmed or refuted by local experts or in the field. An approach that combines semi-automated with manual feature detection could improve the entire mapping process and lead to acceptably accurate mapping results with the potential to greatly reduce the time and effort needed to generate earthflow maps.