A bird’s-eye view on river floodplains: Mapping and monitoring land cover with remote sensing

Abstract

Natural lowland rivers are dynamic environments with a high ecological value. However, 90% of the European and North-American river floodplains are in a degraded state. The functions of floodplains are strongly determined by land cover and they often compete for space in narrowed floodplains. Integrated river management (IRM) tries to take care of floodplains in such way that land cover is optimized for multiple functions. For IRM, monitoring is essential to capture the dynamics, to evaluate changes, and to document the state of floodplains over time. The main objective of this thesis was to establish remote-sensing methods for the monitoring of floodplain land cover over multiple spatial and temporal scales. Several remote-sensing based solutions have been developed for the monitoring of land-cover dynamics in river floodplains and tested in floodplains of the lower Rhine. The phenological change of floodplain vegetation over the course of one year was studied using temporal profiles of its height and greenness. Using multitemporal UAV images, vegetation height was determined with an accuracy similar to much more expensive airborne LiDAR data. Multitemporal elevation models yielded meaningful profiles of greenness and vegetation height over time, which enabled discriminating the different land-cover types. The same dataset combined with a powerful machine learning model (Random Forest) yielded unprecedented high classification accuracies for floodplain vegetation (> 90%), even for similar vegetation types such as grassland and herbaceous vegetation. This method is a practical and highly accurate solution for monitoring areas of a few square kilometres. For large-scale monitoring of floodplains, the same method is recommended, but with data from airborne platforms covering larger extents. Land-cover change over the course of five years was studied for a 100-km river section using satellite images. Using an object-based approach, a sequential deviation of a land-cover object from its class mean was used to detect land-cover change. For most classes the method was unsuccessful (accuracy < 60%) but changes of natural forest, grassland and agricultural fields were detected with an accuracy > 75%. The developed method has important advantages, such as high observation frequency, independence of repeated land-cover classification, and fast processing. At sub-daily frequency, it was assessed how accurate water temperature in a floodplain side channel can be documented from thermal UAV maps. The associated habitat suitability for native and alien fish assemblages was estimated based on the produced temperature maps. Water surface temperatures were mapped four times during a hot summer day with an overall RMSE of 0.53 oC. During the day, temperatures in the side channel increased rapidly to values detrimental for many fish species. The study showed that thermal imagery from UAVs is an efficient and accurate information source to monitor spatiotemporal patterns of thermal habitat suitability. The presently available range of spaceborne and airborne platforms and sensors offers great opportunities to collect information on land-cover change across a range of spatial and temporal scales. This may advance our management of floodplains and help us recovering and protecting these rich ecosystems and the benefits they provide us.