UZH-Logo

Maintenance Infos

Modelling topographic uncertainty: Impacts on large scale environmental modelling.


Hebeler, F. Modelling topographic uncertainty: Impacts on large scale environmental modelling. 2008, University of Zurich, Faculty of Science.

Abstract

Uncertainty can be apprehended as lack of knowledge about a certain phenomenon. Decisions about whether and how to react to this uncertainty depend on a number of factors. These factors include the ability to estimate the amount of uncertainty and thus estimate the involved risk, available options to decrease either the uncertainty or its relevance, and the costs for responding or ignoring uncertainty.
In GIScience, the modelling of processes is subject to uncertainties from a number of sources. Above all, the abstraction inherent in any model results in uncertainty,
created from the assumptions made to simplify complex processes and interrelations in order to formalise and model them. Additionally, uncertainty in any input data
propagates through a model into the results. For topography-based models, i.e. models characterising and detecting topographic form, or models simulating processes that act upon this topography, digital elevation models (DEMs) are a potential source of uncertainty. DEMs consist of measured or digitised elevation values, and as such are
subject to any error in the data capturing process. Widespread DEMs such as GLOBE or SRTM are distributed with accuracy figures that only give global measures such as
root mean square error (RMSE) lacking any information on the spatial distribution of error. Where uncertainty from DEM accuracy has to be modelled to assess its impact on the results of associated topographic models, assumptions have to be made about the spatial distribution of uncertainty. Within this dissertation it has been shown that these assumptions influence the impact of uncertainty on modelled ice sheets. Besides DEM accuracy, a number of factors in handling DEM data introduce additional uncertainty. These factors include the choice of data model, processing such as projecting and resampling of a DEM data, as well as algorithms used to extract and process elevation based information.
Within this dissertation, the influence of resampling on uncertainty in topography has been explored. This was done by assessing the variation in resampled DEMs introduced
by changing the source and target resolution, choice of resampling algorithms and resampling origin. When these uncertainties were modelled and added to input topographies for the GLIMMER ice sheet model, they had noticeable influence on modelled ice sheet configurations. Where higher accuracy reference data for a DEM is available, error can be derived and analysed to provide information about spatial autocorrelation and possible dependencies of error with topographic attributes such as elevation, slope or roughness. Within the course of this dissertation, an uncertainty model was developed which allows modelling of GLOBE DEM uncertainty for areas without higher accuracy reference data such as Scandinavia. The model is based on derived dependencies of GLOBE error with topographic attributes, derived from areas where SRTM data was available to be used as a reference. The model includes both deterministic and stochastic components and reproduces GLOBE DEM uncertainty well for different test areas.
The developed uncertainty model was applied to investigate the impact of DEM uncertainty on different types of models in three case studies. The first case study applied a geomorphologic and hydrologic model (TARDEM), the second case study used two snow melt models, and in the third case study the GLIMMER ice sheet model was employed. Results showed the impact of uncertainty to be depending on a number of facts. Generally, modelled DEM uncertainty had less impact on derived global topographic variables such as mean slope length or the number of derived watersheds when
applied to a hydrological model. Higher impacts were recorded where the model focus was on local processes, such as the delineation of a certain watershed and calculation
of associated parameters such as hypsometry. For process models like the ice sheet model, factors such as terrain configuration (smooth vs. rough topography, abundant
ridges or valleys) influenced the impact of DEM uncertainty on ice sheet model (ISM)results.
Additionally, the amount of uncertainty and its spatial correlation, as well as the relative influence of topography within a model were found to play key roles. This implies that for process models, the impact of uncertainty can vary over time. In the case of the ice sheet model, uncertainty had the greatest impact on ice sheet configuration during phases of inception and retreat, and its impact was shown to be dependent on the overall size of the ice masses.
In another set of experiments, a range of sensitivity tests using different ISM parameters and input data were conducted, and the results of these tests were used to
conduct a full parametric uncertainty analysis (PUA) for a steady-state climate scenario on Fennoscandia. Results from this analysis allowed the comparison of the influence
of uncertainty in other parameters to that of DEM uncertainty, which was found to be equivalent to a 1degC change in climate. The impact of DEM uncertainty was found to be comparable to that of various ‘internal’ ISM parameters. However modelled DEM uncertainty resulted in significantly different ice sheet configurations. This underlines the importance of DEM uncertainty to be considered in ice sheet modelling.
Using different temperature index models (TIM) to model potential snow melt across different resolutions revealed significant impact of scale and resampling on modelled melt rates. This effect was substantially decreased by the use of subgrid model approaches. While it was shown that these subgrid approaches are subject to an increased susceptibility to DEM uncertainty, this effect was more than compensated for by an increased performance in terms of modelled melt rates.
In summary, the results of this dissertation underline the necessity of detailed information on the statistical and spatial distribution of DEM uncertainty to be included
with the data. Additionally, in topographic modelling, uncertainty from other sources such as resampling have shown to be of importance, and modellers and end-users
should account for these uncertainties introduced into model results.

Uncertainty can be apprehended as lack of knowledge about a certain phenomenon. Decisions about whether and how to react to this uncertainty depend on a number of factors. These factors include the ability to estimate the amount of uncertainty and thus estimate the involved risk, available options to decrease either the uncertainty or its relevance, and the costs for responding or ignoring uncertainty.
In GIScience, the modelling of processes is subject to uncertainties from a number of sources. Above all, the abstraction inherent in any model results in uncertainty,
created from the assumptions made to simplify complex processes and interrelations in order to formalise and model them. Additionally, uncertainty in any input data
propagates through a model into the results. For topography-based models, i.e. models characterising and detecting topographic form, or models simulating processes that act upon this topography, digital elevation models (DEMs) are a potential source of uncertainty. DEMs consist of measured or digitised elevation values, and as such are
subject to any error in the data capturing process. Widespread DEMs such as GLOBE or SRTM are distributed with accuracy figures that only give global measures such as
root mean square error (RMSE) lacking any information on the spatial distribution of error. Where uncertainty from DEM accuracy has to be modelled to assess its impact on the results of associated topographic models, assumptions have to be made about the spatial distribution of uncertainty. Within this dissertation it has been shown that these assumptions influence the impact of uncertainty on modelled ice sheets. Besides DEM accuracy, a number of factors in handling DEM data introduce additional uncertainty. These factors include the choice of data model, processing such as projecting and resampling of a DEM data, as well as algorithms used to extract and process elevation based information.
Within this dissertation, the influence of resampling on uncertainty in topography has been explored. This was done by assessing the variation in resampled DEMs introduced
by changing the source and target resolution, choice of resampling algorithms and resampling origin. When these uncertainties were modelled and added to input topographies for the GLIMMER ice sheet model, they had noticeable influence on modelled ice sheet configurations. Where higher accuracy reference data for a DEM is available, error can be derived and analysed to provide information about spatial autocorrelation and possible dependencies of error with topographic attributes such as elevation, slope or roughness. Within the course of this dissertation, an uncertainty model was developed which allows modelling of GLOBE DEM uncertainty for areas without higher accuracy reference data such as Scandinavia. The model is based on derived dependencies of GLOBE error with topographic attributes, derived from areas where SRTM data was available to be used as a reference. The model includes both deterministic and stochastic components and reproduces GLOBE DEM uncertainty well for different test areas.
The developed uncertainty model was applied to investigate the impact of DEM uncertainty on different types of models in three case studies. The first case study applied a geomorphologic and hydrologic model (TARDEM), the second case study used two snow melt models, and in the third case study the GLIMMER ice sheet model was employed. Results showed the impact of uncertainty to be depending on a number of facts. Generally, modelled DEM uncertainty had less impact on derived global topographic variables such as mean slope length or the number of derived watersheds when
applied to a hydrological model. Higher impacts were recorded where the model focus was on local processes, such as the delineation of a certain watershed and calculation
of associated parameters such as hypsometry. For process models like the ice sheet model, factors such as terrain configuration (smooth vs. rough topography, abundant
ridges or valleys) influenced the impact of DEM uncertainty on ice sheet model (ISM)results.
Additionally, the amount of uncertainty and its spatial correlation, as well as the relative influence of topography within a model were found to play key roles. This implies that for process models, the impact of uncertainty can vary over time. In the case of the ice sheet model, uncertainty had the greatest impact on ice sheet configuration during phases of inception and retreat, and its impact was shown to be dependent on the overall size of the ice masses.
In another set of experiments, a range of sensitivity tests using different ISM parameters and input data were conducted, and the results of these tests were used to
conduct a full parametric uncertainty analysis (PUA) for a steady-state climate scenario on Fennoscandia. Results from this analysis allowed the comparison of the influence
of uncertainty in other parameters to that of DEM uncertainty, which was found to be equivalent to a 1degC change in climate. The impact of DEM uncertainty was found to be comparable to that of various ‘internal’ ISM parameters. However modelled DEM uncertainty resulted in significantly different ice sheet configurations. This underlines the importance of DEM uncertainty to be considered in ice sheet modelling.
Using different temperature index models (TIM) to model potential snow melt across different resolutions revealed significant impact of scale and resampling on modelled melt rates. This effect was substantially decreased by the use of subgrid model approaches. While it was shown that these subgrid approaches are subject to an increased susceptibility to DEM uncertainty, this effect was more than compensated for by an increased performance in terms of modelled melt rates.
In summary, the results of this dissertation underline the necessity of detailed information on the statistical and spatial distribution of DEM uncertainty to be included
with the data. Additionally, in topographic modelling, uncertainty from other sources such as resampling have shown to be of importance, and modellers and end-users
should account for these uncertainties introduced into model results.

Downloads

409 downloads since deposited on 21 Oct 2008
35 downloads since 12 months
Detailed statistics

Additional indexing

Item Type:Dissertation
Referees:Weibel R, Purves R S, Hulton N, Fisher P
Communities & Collections:07 Faculty of Science > Institute of Geography
Dewey Decimal Classification:910 Geography & travel
Language:English
Date:April 2008
Deposited On:21 Oct 2008 15:30
Last Modified:05 Apr 2016 12:28
Number of Pages:214
Funders:SNF
Official URL:http://www.geo.unizh.ch/~fhebeler/fhebeler2008dissertation.pdf
Related URLs:http://www.geo.unizh.ch/~fhebeler (Author)
Permanent URL: http://doi.org/10.5167/uzh-3712

Download

[img]
Preview
Filetype: PDF
Size: 13MB

TrendTerms

TrendTerms displays relevant terms of the abstract of this publication and related documents on a map. The terms and their relations were extracted from ZORA using word statistics. Their timelines are taken from ZORA as well. The bubble size of a term is proportional to the number of documents where the term occurs. Red, orange, yellow and green colors are used for terms that occur in the current document; red indicates high interlinkedness of a term with other terms, orange, yellow and green decreasing interlinkedness. Blue is used for terms that have a relation with the terms in this document, but occur in other documents.
You can navigate and zoom the map. Mouse-hovering a term displays its timeline, clicking it yields the associated documents.

Author Collaborations