A method for distinguishing different water sources during flood events in river floodplains, northern Poland

Abstract

The Biebrza wetlands are located in an ice-marginal valley in northeastern Poland and cover some 195 000 hectares. The Biebrza river is not regulated, the valley is not reclaimed and still has marsh, fen and grassland vegetation whereas large scale human interventions in hydrology are absent. The most characteristic feature of the Biebrza river is flooding, which occurs almost every year during the spring thaw. During these floods the inundated wetlands covers many km2. This water may in principle derive from three different sources: 1) rain and snow-melt water, 2) river water and 3) groundwater. We developed a method to distinguish these water sources enabling us to estimate the area inundated by the different water types. Our analysis was based on long term monitoring of water levels via gauge stations and an in depth analysis of floods for the period 2001-2008. We used Landsat images, field surveys during floods and chemical analysis of water samples. The satellite images of the Landsat 5 TM and the Landsat 7 ETM+ captured during maximum flood were processed. This included the calculation of ratio of bands 7/4, the Normalized Differential Vegetation Index (NDVI) and a Principal Component Analysis (PCA). The results of these transformations were stored in new bands and were classified by using the maximum likelihood classifier for identification of inundated areas. The image of 2002 was verified with high overall accuracy on the basis of 796 sample points. Samples obtained from hydro-chemical transects were statistically analyzed using PCA. The results of the statistical analysis were imported to ArcGIS. Geographical analysis resulted in maps depicting the spatial distribution of the first three principal components. K-means cluster analysis was performed by establishing clusters, which finally were interpreted as representing the three different water sources: river, groundwater and snowmelt. Validation was performed by comparing the results with vegetation maps. There was a good match between water sources and generalized vegetation maps. We conclude that the method was demonstrated to be useful for analysing vast floodplain wetlands in which it is relevant to know the spatial distribution of different water sources during inundations. In the case of ecological studies aiming at conservation/restoration of habitat condition within floodplain wetlands, there is a need to know the balance between water from different sources. Distinguishing different water sources during inundation of river floodplains is also important for constructing and verification of floodplain hydrodynamic models.