Monitoring Vegetation Structure in Floodplains for Flood Risk Estimation

Floodplains are among the most valuable ecosystems of the world. In the Netherlands they have a double function, serving both safety and nature, recreational purposes. Safety standards for flood protection of this densely populated country are the highest world-wide, protecting against events expected to occur with an annual probability of 1/10,000. Rivers in the Netherlands are all embanked by dikes to protect the low-lying polders from flooding. Along the main river channels, embanked floodplains carry excess water during high-water periods. Objects within these embanked floodplains, whether natural or man-made, increase the hydraulic roughness and hamper the water flow, causing flood waters to increase. In 2006, the Dutch government adopted the Spatial Planning Key Decision Room for the River, aiming at reducing flood-water levels, together with restoring riverine ecosystems. To account for changes in floodplain vegetation over time, the government requires five-yearly updates of vegetation maps, to monitor hydraulic roughness patterns and ecological quality of the floodplain. We developed a monitoring method which combines object-based analysis of CIR photos with knowledge on vegetationsuccession paths. We selected the nature reserve De Blauwe Kamer, in the floodplain of the river Neder-Rijn as a pilot study. In 1992 an open connection was created here between the main river channel and the floodplain, accompanied by creating relief differences in this formerly agricultural area, to increase landscape dynamics. Image interpretation of natural vegetation is generally hampered by spectral overlap between vegetation types. Within our monitoring method we combine two improvements to reduce this effect and increase classification accuracies. Object-based interpretation is the first step, because it deals with shape and internal variation of vegetation types, thus reducing spectral confusion. Next, we introduce succession rules based on succession paths observed in the reserve and similar areas. Knowing the vegetation stage at t=0, possible stages at t=1 can be deduced, reducing the number of possible classes. Together with the number of classes, spectral overlap will scale down. * Corresponding author.