Predictive modelling and monitoring of Ellenberg moisture value validates restoration success in floodplain forests
How does floodplain forest vegetation react to a changed hydrological regime induced by a restoration project? Is it possible to model the Ellenberg moisture value (mF) using variables representing site moisture conditions? Does mF help predict changes in vegetation composition and thus help in validating restoration success?
Danube floodplain, Neuburg a. d. Donau, Bavaria, Germany.
A Generalized additive model (GAM) and a generalized linear model (GLM) were used for spatial prediction of mF using groundwater, relief and soil variables as predictors. Vegetation response to a changed hydrological regime was tested through replacing the old groundwater model with a new model representing the new water regime. With the help of a GIS, a spatial distribution and a scenario map of mF was produced. Model validation was done by cross-validation and comparison between predicted and observed data from repeated vegetation sampling on permanent plots.
Groundwater level (minima of 2009) and soil thickness entered the final model as significant predictors of mF. The GAM of mF showed good predictive power, with more than 57% of deviance explained. The corresponding GLM had slightly lower accuracy, with 51% of deviance explained. Using the new groundwater model (minima of 2011) it was possible to forecast mF after restoration. Changes in mF are to be expected in areas directly affected by restoration measures. The model validation has proven that the quality of model prediction is good and represents the reality fairly well.
The study has shown that mF is a useful indicator for a summary prediction of ecological change. It can be expected that restoration will turn overall conditions wetter, and vegetation will respond to these changes with an increase of wetland plant species. With the help of the model, it is possible to measure the success of the restoration project. It is demonstrated that GAM can be used for intensive data exploration, after which a less complex and easily computable GLM can be parameterized.