AbstractIncreased urbanisation is a global trend, with more people living in urban areas
causing more soils to be sealed by impervious surfaces, increasing the effective impervious
area of an urbanized catchment. As a consequence, the bigger urban population
will be facing hydrological events with greater magnitude and frequency. This thesis
aims to combine high resolution multispectral data with LiDAR measurements to optimally
map impervious surfaces in the urbanized catchment of the Biala river. This
detailed land cover map of Bialystok, Poland will be used for hydrological modelling
in WetSpa, a model that estimates the amount of discharge at the outlet of the catchment.
The effects of spatial distribution and resolution of the input land cover map
on the modelled discharge is evaluated.
The detailed land cover map was obtained in different steps. First, a support vector
machine learning approach was applied to classify the WorldView 2 data. LiDAR
measurements (elevation and curvature) were used to improve classification accuracies
in post-classification. A separate support vector machine was set up to classify
shadow pixels in Bialystok. Finally, the detailed land cover map (containing 16 land
cover classes) was aggregated to a map containing only 5 useful land cover classes
for hydrological modelling in WetSpa. The land cover classifications had an overall
kappa index of agreement of 0.87 and 0.96 respectively.
Different WetSpa scenarios are created by using different inputs for the land cover
map and the runoff coefficient parameter: a distributed scenario at 30 meter resolution,
a non-distributed scenario at 30 meter resolution and a distributed scenario at 10
meter resolution. By parametrizing eachWetSpa scenario individually and comparing
them, the influence of spatial distribution and resolution on the modelled discharge
is assessed. It is proven that, when modelling urbanized catchments, high quality
spatial data providing detailed information on the spatial distribution of land cover
is crucial for accurate hydrological modelling. Using smaller grid cells in Wetspa did
not result in more accurate discharge estimates.
|Date of Award||2016|
|Supervisor||Frank Canters (Promotor), Jarek Chormanski (Promotor), Frederik Priem (Advisor) & T. Berezowski (Advisor)|