Development of an imaging spectroscopy based method for mapping and monitoring plant functional types in river floodplains: A case study using Spectral Mixture Analysis in the Millingerwaard

Laboratory of Geo-Information Science and Remote Sensing
  Education
  Research
  Publications
  Models
  News & Calendar
  News
  Calendar
  Archive
  News
  Calendar
  2011
  2010
  2009
  2008
  2007
  2006
  2005
  2012
  Staff
  Equipment
  Contact details
  Workshops

28 Jun 2007 09:00 - 28 Jun 2007 09:30
Unit: Laboratory of Geo-Information Science and Remote Sensing
Location: Gaia 1, Droevendaalsesteeg 3.
Organisation: Wageningen University

 By:  Lucia Sanchez Prieto    

Abstract:   Due to flood risk increase in the Netherlands during the last decades, new river management strategies are being developed in areas such as river floodplains. As floodplains of the Rhine River are part of the National Ecological Network of the Netherlands, these strategies are meant to safeguard both flood protection as well as nature rehabilitation objectives. Nature rehabilitation implies that former agricultural land is transformed into natural areas, but vegetation is also an important component which influences the hydraulic roughness of the floodplains which increases peak discharge. In order to prevent catastrophic flood events river managers need to forecast vegetation dynamics. Dynamic Vegetation Models (DVM) are used to predict vegetation succession, but to initialize and validate these models, information about the spatial distribution of vegetation is required.
This research is focused on establishing a methodology for mapping and monitoring floodplain vegetation by the application of imaging spectroscopy techniques. Vegetation classes are defined according to the concept of Plant Functional Types (PFTs), because of its appropriateness when being used by DVMs. PFTs are defined in a wide variety of terms, but in this study they were defined as vegetation clusters that have a similar response to water flow impact. This response was measured by what is known as hydraulic resistance which was characterized by quantifying specific plant traits: height, density and flexibility. Since the heterogeneity of PFTs leads to intimate mixture of classes in a sub-pixel scale Spectral Mixture Analysis (SMA) was considered an appropriate technique. SMA was used to classify into PFTs imaging spectroscopy data acquired by HyMap and CASI in 2004 and 2001, respectively.
The methodology was developed using HyMap as base image. In this first approach, nine PFTs based on plant size were defined. Linear spectral unmixing was applied to the first 23 MNF bands from HyMap using ten endmembers (nine PFT classes and soil). These were extracted from the image based on ground truth knowledge derived from field data. Three methods were applied: unconstrained, semi-constrained and fully constrained linear unmixing. Finally, a temporal analysis was performed in which both HyMap (2004) and CASI (2001) were subject to previous methodology. In this second approach, five PFTs were redefined based on plant species and six endmembers were used.
Overall classification accuracy of HyMap improved in second approach with respect to first from 66 % to 40 %. Classification of woody PFTs showed better performance (68 %) than herbaceous PFTs (47 %). Dwarf shrubs (Crataegus monogyna) were classified with 57 % accuracy, shrubs (Sambucus nigra) with 81 % and pioneer trees (Salix sp.) with 64 % in the second approach. Performance of CASI image showed worse results with an overall accuracy of 21 %. The results from this research reveal that it is possible to map and monitor PFTs in river floodplains by applying Spectral Mixture Analysis to hyperspectral images.
Print this activity