Detecting hydromorphological structures using an AI-based analysis of high-resolution drone imagery to access physical river habitat development

The detection of hydromorphological structures gained more attention during the last decades. Many approaches of different scopes, scales and purposes have been developed. They can either be classified as stand-alone methods, like the German River Habitat Survey, which evaluates the hydromorphological integrity on a catchment scale or as methods being part of an ecological assessment, which includes the estimation of hydromorphological characteristics on the scale of respective study sites. The main purposes of detecting hydromorphological structures are to investigate the spatial characteristics and temporal scale of change to collect reliable and comparable data in a sampling setup of an ecological multi habitat sampling. Especially river restoration projects often lack the detection of positive effects on aquatic biota induced by missing or wrong development of physical river habitat structures (PRHS).

Most methods available for determining PRHS are insufficient for this task as they lack sufficient temporal and spatial resolution. Examples thereof include overview methods based on topographic maps and remote sensing. On the other hand, visual assessment methods do not reach the required accuracy and objectiveness or are too general if too few hydromorphological structures are assessed. Therefore, this research proposes the combination of Unmanned Areal Vehicle (UAV) and high-resolution sensors. This combination creates high-resolution imagery or point clouds by using multispectral sensors or Lidar scanner.

In a case study of the river Lippe, the methods for detecting PRHS on Structure from Motion (SfM) high-resolution imagery with deep learning, based classification methods are applied. Results indicate the potential from different deep learning classification approaches to identify physical river habitat structures being able to assess the development over time.