Hyperspectral analysis of surface materials for water balance estimation in urbanized areas

: A case study on the Brussels Capital Region

Scriptie/masterproef: Master's Thesis


This thesis aims to use high resolution remote sensing data, both spatially and spectrally, to improve the input used in hydrological modelling in an urban context. The improved input seeks to represent the surface materials present in urban area more accurate and precise. This is important given the continuing urbanization of the world, which leads to a need for more accurate information on what happens with the water balance in an area after it has been built-up. In the first part of the thesis an overview will be given of the problems that arise when using remote sensing in an urban area. Since urban areas show high heterogeneity in surface material cover on a local scale and a wide variety of materials are present, it is difficult to obtain good maps of the surface materials in an area. The usage of high spatial and high spectral resolution remote sensing data is required to create a detailed surface map of an urban area. In this study data is used from the hyperspectral APEX sensor with a resolution of 2m available from the 2013 Belair campaign over Brussels. The high spatial resolution of the data ensures that the urban objects present in the area, like buildings, gardens, roads and roofs can be represented; while the high spectral resolution is necessary to differentiate between all the different surface covers. The newly identified surface material classes, in the sense that they were previously generalized for hydrological modelling, are parameterized in terms of their hydrological properties, using porosity and manning roughness coefficient values taken from the literature, and then integrated into the model.
The hydrological modelling was done on the small Watermael catchment in the Brussels Capital Region, making use of the WetSpa model developed by the Department of Hydrology and Hydraulic Engineering of VUB. The impact of the new inputs on the modelling results were tested by comparing a previous run of the model with new runs that incorporate these new inputs. In total 5 different runs were performed. Firstly, in the reference run, previous inputs based on older, less precise remote sensing data were used. Secondly, in the so-called impervious run, the per-cell impervious fraction used in the model was changed using the updated surface map of the study area so that sealed surfaces are more accurately represented. Thirdly, in the manning run, the manning roughness coefficient map was updated taking into account the diversity of identified surface types, so that the flow speed of the runoff can be more accurately represented. Fourthly, in the porosity run, urban interception parameters were estimated for the newly mapped land-cover classes so that the water retained by surface materials can be more accurately modelled. Fifthly, in the combined run, all the previous changes to the inputs are added together in one run to give an overall picture of the impact of these changes.
The results of the different model runs are compared by looking at each run’s overall water balance and the discharge at the catchment outlet for large precipitation events. The results of the runs show that with the updated impervious fraction map, there is a higher overall runoff, mainly surface runoff. The manning run shows that the discharge peaks arrive later and the runoff is spread over a longer period. The interception run shows little change overall but there are small seasonal discharge variations between summer and winter compared to the reference run. Finally in the results of the combined run, the changes of the previous runs are reflected as expected. Our study shows that the use of high spatial and high spectral data changes the
outcome of the hydrological model simulation significantly. After looking at the results more closely the changes seem rather due to the more accurate representation of the surface materials, both urban and non-urban, than to the updated class-specific parameters. It can be concluded that high-resolution hyperspectral remote sensing data is useful, if not required, in the context of detailed urban hydrological modelling. Future improvements in the modelling should focus on integrating the artificial drainage systems, and as such taking into account the more complex flow that is imposed on runoff by urban structures like houses and roads.
Datum Prijs2015
Toekennende instantie
  • Vrije Universiteit Brussel
BegeleiderFrank Canters (Promotor), Boud Verbeiren (Co-promotor) & Frederik Priem (Advisor)

Citeer dit