Distinctive zones of inundation water during floods were shown to originate from different sources in some major floodplains around the world. Recent research showed that the zonation of water in rivers and floodplains is related to vegetation patterns. In spite of this, water source zones were not used for vegetation modeling due to difficulties in their delineation. In this study, we used simulation results of a fully-coupled groundwater-surface water integrated hydrological model (IHM) HydroGeoSphere and the Hydraulic Mixing-Cell method to provide standard hydrological predictors (e.g. water depth, inundation length, groundwater depth, exchange flux) and the extent of inundation zones having a certain water source (discharged groundwater, river, rainfall, and snowmelt). These variables were used to train a vegetation model for the lower Biebrza floodplain (about 290 km2) using vegetation maps from 1960, 1980, and 2000. We used a one-at-a-time (OAT) approach, where each map was validated based on a model trained on the remaining two maps to obtain realistic error estimates. We also used a fractional approach in which a fraction of each map was used for training and validation. The single model from the fractional approach was used to assess the importance of predictors and to predict vegetation for the 20th century and for the 21st century using IHM simulation forced by the Twentieth Century Reanalysis data and EURO-CORDEX RCP 2.6, 4.5, and 8.5 model ensembles. The model which used both water sources extent and standard predictors performed the best overall and was sensitive to the future trends. The extent of river water within the inundation area was by far the most important vegetation predictor. The models that neither used the water sources extent predictors nor the exchange flux were not able to predict the trends of areas covered by certain vegetation types under future climate. The advantage of the water sources extent predictors was their ability to represent the spatial effect of local hydrological phenomena. This was not possible with the standard predictors, because they show only the source of the phenomena (e.g. groundwater discharge zone), but do not indicate the actual area affected by its physical and chemical properties, which is more relevant for vegetation development. Our results highlight the relevance of using water extent predictors due to their ability to explain spatiotemporal ecological processes, such as vegetation development. We suggest to use water extent predictors in modelling for developing more accurate decision support for wetland floodplains.