The terms of reference for a monitoring tool were defined in close
cooperation with the end-users which in first instance consisted of the
team responsible for floodplain vegetation management and the state
forestry agency, one of the bigger landowners in the floodplains. We had
ample discussion on the requirements of the end-users, such as ‘the tool
must be usable in the field e.g. using a tablet with GPS functioning’,
‘the tool must allow for quick analysis per cadastral plot’ and
‘information must be downloadable for further processing in GIS if
needed’. The basic terms of reference were:
- Produce a floodplain vegetation map and show the difference to the
legal permitted “state of the vegetation” of the whole river system;
- The map should be up to date, providing on the fly user selectable
single-date maps;
- As a reference map for all users, a year-map should also be available;
- Use Sentinel-2 data and classify as robust and accurate as achievable;
- Use the legal vegetation map classes;
- Data and interface should be open and accessible for all stakeholders;
- Added in second instance: differentiate between field use (including
GPS location) and desktop use; export of maps to GIS;
- Added in second instance: timeseries to visualize longer term trends.
The added value of year-maps was only later recognized as there was a
need for a map with the highest possible accuracy that could be used in
communication with all stakeholders, and would be less affected by
inaccuracies in single image classifications as a result of e.g.
atmospheric disturbances. The on-the-fly single date-maps gave a good
up-to-date insight, but different stakeholders could confront each other
with different day-maps. Also, for reference in communication, a more
stable year-map was thus preferred. Also, time series were added to
allow for a better understanding of the development of vegetation
through the years and have a better approach for areas where changes in
vegetation type fluctuated through the years, due to the management of
the area.
The overall technical classification workflow is visualized in Figure 3.
For data-retrieval, pre-processing and classification is carried out on
the Google Earth Engine (GEE; Basic remotely sensed data consisted of
Sentinel-2 MSI (Level-1C) data and the Dutch LiDAR based AHN (0.5m raw
samples), both are available through GEE. Including Lidar-based data in
the classification improves results for floodplain vegetation