CONCLUSION
Here we present an approach for automated mapping of fluvial
hydromorphology from remote sensing imagery. The main advantages of this
approach are that it is quick, flexible, and based on readily available
imagery, and classifications follow a strict rule-based system, so are
unaffected by researcher-subjectivity and associated biases. Firstly,
image processing is fast. The approach does not rely on field surveys or
hydraulic modelling, both of which are time consuming, and application
of an automated decision rule system is potentially faster than
qualitatively interpreting images. The method is flexible and can be
easily adapted to various criteria. For example, our HMU classification
system used a gradient of 4% as a threshold to distinguish betweenmild and steep gradients, but this could be adjusted if it
is producing results that appear unrealistic, which may be the case for
low-lying, downstream reaches. With the measurement of additional river
properties, such as depth or velocity, the classification can be refined
to provide a more detailed hydromorphological delineation; for example,
depth could be used to distinguish between cascades (shallow) and rapids
(deep), and velocity could be used to distinguish between walks (slow)
and glides (fast). Imagery of the required types are usually readily
abundant. Within a Norwegian context, orthophotos are available for all
rivers, and there is a complete high-resolution topographic
LiDAR-derived DTM/DSM covering the country. In locations where existing
aerial photographs are not available, imagery of a suitable resolution
(e.g. 10 cm) can easily be obtained using UAVs. Additionally, a range of
national and global data sources exist that can be used for estimating
channel surface gradient (e.g. ASTER DEM; see Azizian & Brocca 2020).
Finally, using such a system removes researcher-subjectivity, and
therefore means that classifications are consistent among different
rivers and that the reasons for the classifications are documented.