Rainfall-runoff modelling : comparison of modelling strategies with a focus on ungauged predictions and model integration

Abstract

This work is an integrative study of methods for rainfall-runoff simulation. The thesis addresses process description considerations, problems associated with ungauged predictions, and tools for sharing data and computational procedures over the Internet.

Simple runoff model structures can be preferred as they facilitate systematic uncertainty assessments and can fit streamflow data as well as more complex structures. A more complex model structure can be invoked when assessing hydrological problems involving water quality considerations, which often necessitates explicit representation of runoff generation mechanisms and different flow pathways. Models that have more ambitious objectives than merely to reproduce streamflow should also be validated against other measured data than just flow records. It is demonstrated that two different model parameterisations can yield both good quality fits to observed streamflow data, but generate drastically different evapotranspiration time series. This thesis also presents a case study where a physics-based hydrological model is calibrated and validated using groundwater levels and isotope tracer results.

Runoff predictions for a catchment lacking streamflow records can be based on establishing relationships between physical catchment attributes and runoff model parameters using flow data from other catchments belonging to the same region. Results of this thesis suggest that consideration of correlation among runoff model parameters can improve performance of such a regionalisation exercise. Ideally, one would wish to exploit available information on catchment properties in a more physics-based way where observed values of physical catchment properties are incorporated into the model structure and parameters directly. Although physics-based models still face problems which are not readily solvable, results of this study show some promise in predicting streamflow in a physically more consistent way with a minimum amount of parameter calibration.

The final part of this thesis explores tools for distribution of environmental data sets and simulation models over the Internet. The idea is to promote openness in environmental simulation studies by providing means for data and model integration from resources published by different parties.