Uncertainty assessment for hydrologic modeling

Researcher: Lien Loosvelt
Promotors: Niko Verhoest

 

The objectives of this research are:  

  • To assess the uncertainty on the available soil and vegetation data as used for the model input;
  • To develop an uncertainty analysis tool for the TOPLATS model;
  • To quantify the impact of soil and vegetation data on the model predictions (e.g. soil moisture, energy fluxes, evaporation, runoff) and the output uncertainty;


To realize the proposed objectives, the following research questions can be formulated:

  • Which methodology can be applied to assess the uncertainty on the selected input variables and model parameters? How important is the integration of edges into uncertainty analysis? How important is the temporal evolution of uncertainty?
  • Which kind of information do we need know in order to account for uncertainty? Is this information available as a standard source?
  • Which methodological approaches can be used to propagate the several sources of uncertainty through the model?
  • Which measure is the most appropriate to evaluate the output uncertainty? On which scale the evaluation needs to be done?
  • In which manner is the uncertainty on the image data and model parameters best related? What is the relative importance of both, regarding the generated model error? 
  • Which methodoloy can be applied to account for the joint contribution of related sources of uncertainty?
  • How large is the impact of using 'standard' available maps and 'standard' parameter values from literature as input in distributed hydrological models? What is the impact of lacking field or laboratory measurements on the modelled uncertainty? 
  • How can the modelled uncertainty be validated?


The study can be divided into four mayor components:

  1. Data collection: distributed hydrological models require a lot of input information, including meteorological data, image data, model parameters etc. These data are not always 'standard' available from literature or governement organisations. Concerning vegetation parameters (e.g. leaf area index) and soil properties, field experiments may need to be conducted to properly represent their value. In assessing the uncertainty, additional data may be required.
  2. Uncertainty assessment: following the amount of information available, the uncertainty on both the map inputs and the model parameters needs to be quantified. Because every type of information has a different source of uncertainty, different methods of uncertainty assessment need to be applied. These methods can be found among the 'state of the art' concerning uncertainty modelling.
  3. Building an uncertainty analysis tool: this tool allows the modeller to integrate the uncertainty on the selected inputs into the TOPLATS model. Therefore an appropriate method needs to be chosen to propagate the input error through the model. By implementing this method into a source code and by building it around the model, predicted values will be accompanied by uncertainty bounds.
  4. Assessing model prediction uncertainty: in this part a method needs to be chosen to properly quantify the output error. Also the use of different uncertainty measures will be explored. The selected method needs to be able to compare the generated model output uncertainty for different model run scenarios.