Hans Roelofsen MSc
- Position:PhD student
My background is in Geographical Information Systems (GIS), remote sensing, geostatistics and environmental sciences. I am interested in using remote sensing techniques to investigate and quantify the spatial variation in plant trait values. Plant traits are increasingly recognized as a tool to classify plant species on functional grounds, but are inherently measured at the individual plant level. This may be problematic for spatial ecological models that require full-coverage plant trait values as input. Remote sensing techniques may provide a solution for this problem, given their full spatial coverage of data acquisition. My research aim then, is to investigate which and how plant traits can be quantified from remote sensing and how they can be used in ecological models.
My position was created and financed by KWR Watercycle Research Institute. Because of the interdisciplinary character, I have three supervisors who contribute their specific knowledge. Flip Witte from KWR is my promotor, Peter van Bodegom (VU) advises me on plant traits and Lammert Kooistra from Wageningen University acts as the remote sensing specialist. As such, I distribute my time over all three institutes.
Research project
Maps showing the spatial distribution of vegetation types provide information for nature management, but are expensive to produce and inaccurate. My first research project aims to develope a remote sensing based approach to map the spatial extent of Dutch national vegetation types. These types are defined based on individual species abundance and co-occurrence, which is typically not discernable with remote sensing. We overcome this by not looking at vegetation types directly, but using functionality of the vegetation as intermediate between reflectance and vegetation.
We intend to produce a map of the Dutch national vegetation typology on Ameland in two steps. Firstly, we will define indicator values as an ordinal measure of vegetation preference for soil moisture, nutrient availability and salinity. Using Partial least squares regression, we link observed indicator values at a set of local vegetation plots to remotely sensed reflectance values, and subsequently, to predict indicator values for the complete study area. Secondly, we note that vegetation plots of a similar vegetation type tend to cluster in a three dimensional indicator value space. The indicator value space will be distributed among vegetation types using Gaussian Mixture Density modelling, yielding occurrence probability of a vegetation type as a function of three indicator values. The remote sensing predicted indicator values will then be used to calculate vegetation type occurrence probabilities for a set of vegetation types for the complete study area. As each pixel will be assigned to the vegetation type sporting the highest occurrence probability, the end result will be a map of the spatial variation in vegetation types. In addition, the highest occurrence probability of each pixel will be mapped to indicate spatial variation in model certainty.
Supervisors
Flip Witte
Lammert Kooistra
Peter van Bodegom
Links
KWR Watercycle research institute
Centre for Geo-information, Wageningen University
