Al nivel
superior
Sitio en Ingles
Tabla de contenidos:
Al nivel
superior
Sitio en Ingles
Introduction A la tabla de contenidos
The PhD research is related to distributed catchment modelling, a relative new topic with growing interest. Such interest in the subject is increasing rapidly under the society's pressure to accurately assess the impact of human activity on the natural resources, among which, water is one of the most vulnerable. Clear understanding of how human interference affects the water resources of a river basin will assist policy makers and water administrations in the definition of the most appropriate regulations and measures that conserves and protects the basin's water resources.
The specific objective of the doctoral research was the performance assessment of a physically based distributed hydrologic model using the MIKE SHE code and data of the river basin Gete (586 km2). The MIKE SHE code was selected as the modelling tool for its capacity to model the entire hydrologic cycle, the underground, soil and surface component in an integrated way (capitulo 1). The Gete catchment was chosen as test basin for its size, rolling landscape and shallow to medium depth of the phreatic groundwater table. The catchment is very representative for the loamy region of Belgium, being an important zone for the natural recharge of the groundwater system (capitulo 1).
In constructing the model of the study site, care was exerted to integrate (i) distributed data with different accuracy and scale characteristics and (ii) time varying data to characterise the hydrological stresses and catchment outputs (capitulo 2). Calibration and evaluation approaches were selected and applied (capitulo 3 and capitulo 4). For avoiding over-parameterisation and to address uncertainties associated to the model predictions, several sensitivity analyses were carried out including the assessment of the effect of the grid size, the potential evapotranspiration data and the digital terrain model generation-methods on the model performance, as well as, the values of the effective parameters (capitulo 4 and capitulo 5).
An uncertainty analysis was then performed to determine confidence intervals for the modelled variables (capitulo 6). The latter analysis involved the Monte Carlo simulation technique within the framework of the GLUE (Generalised Likelihood Uncertainty Estimation) methodology, which is based on the equifinality philosophy (many models that are consistent with the available observations). The sensitivity and uncertainty analyses (capitulo 4, capitulo 5 and capitulo 6) revealed (i) the equifinality of parameter sets apparently due to non-linearities and parameter interactions; (ii) the significant uncertainties associated to distributed model predictions, apparently caused by the significant uncertainty of the data available for the modelling; and (iii) the sensitivity of some of the components of the model structure to the referred data uncertainty.
capitulo I: INTRODUCTION To Table of Contents
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
1.1).
Fig. 1.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
capitulo II: DATA SOURCES AND INITIAL PARAMETERISATION OF THE MODEL OF THE STUDY SITE A la tabla de contenidos
This capitulo outlines the general data requirements, data sources, data accuracy and data flow for building the distributed model of the study site with the aid of the MIKE SHE code. In doing so, this capitulo also provides more details of the conceptual model of the study site (Fig. 2.1).
Fig. 2.1: Sketch of Belgium subdivided into the northern, central and
southern regions and schematization of the main hydrological processes.
Publications related to this capitulo are Feyen et al.(2000), Vázquez et al.(2002) and Vázquez et al.(2004a).
capitulo III: MULTI-CRITERIA FRAMEWORK FOR ASSESSING THE PERFORMANCE OF DISTRIBUTED MODELS A la tabla de contenidos
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
The study site is the Gete catchment,
located in Belgium (Fig.
3.1).
Fig. 3.1: Flowchart for the creation of
DTMs (Vázquez et al.,
2002).
Publications related to this capitulo are Feyen et al.(2000), Vázquez et al.(2002) and Vázquez et al.(2004a).
capitulo IV: CONVENTIONAL CALIBRATION AND EVALUATION OF THE DISTRIBUTED MODEL OF THE STUDY SITE A la tabla de contenidos
This capitulo describes the calibration of the distributed model of the study site (600x600 m²) within a multi-objective framework. Firstly, the model was calibrated against both daily discharge measurements and observed water levels and then evaluated using a simple Split-sample (SS) test. In addition, a Multi-site (MS) evaluation test was performed for 2 internal discharge stations and 6 observation wells that were not considered during model calibration.
The main calibration and evaluation
periods are sketched in (Fig.
4.1).
Fig. 4.1: Main calibration and evaluation
periods (Vázquez et al.,
2002).
Publications related to this capitulo are Feyen et al.(2000) and Vázquez et al.(2002).
capitulo V: SENSITIVITY ANALYSES WITH RESPECT TO KEY INPUT AND MODELLING VARIABLES A la tabla de contenidos
The potential of distributed codes (DC) for integrated water management is usually in contrast with the significant amount of input variables and parameters necessary to represent and schematise a particular catchment. As such, it is important to investigate which of those input variables are relevant to the simulation of the dynamics of the modelled hydrological system. This investigation should be based on reliable sensitivity analyses (SA) and implemented according to the distributed nature of those codes.
This capitulo reports the results of some SA that considered few input variables and their effect on the model performance and on the magnitude of the model effective parameters(Fig. 5.1). In the same framework, this capitulo also depicts some results dealing with the incidence of additional sampling of the parameter space on the performance and effective parameter values.
Fig. 5.1: Flowchart of the Multi-resolution (MR) test for investigating
the sensitivity to grid size
(Vázquez et al.,
2002).
Publications related to this capitulo are Feyen et al.(2000), Vázquez et al.(2002) and Vázquez et al.(2004a).
capitulo VI: UNCERTAINTY ASSESSMENT OF THE MODEL PREDICTIONS To Table of Contents
This study aimed to estimate the uncertainty associated with the prediction of the basin-wide streamflow in the study-catchment. The Generalised Likelihood Uncertainty Estimation (GLUE) process was applied to assess the performance of the hydrological model with randomly defined parameter sets and to estimate confidence simulation limits. The GLUE analysis showed that the horizontal and vertical components of the hydraulic conductivity of one of the geological layers had the most influence on the streamflow model performance. The analysis also revealed the significant impact of data uncertainty in the hydrological modelling of the study site.
The study site is the Gete catchment,
located in Belgium (Fig.
6.1).
Fig. 6.1: Bayesian approach to upgrade probability estimates conditioned
on newly observed data (Vázquez,
2003).
Publications related to this capitulo are Vázquez et al.(2002) and Vázquez et al.(2004a).
capitulo VII: CONCLUSIONS AND RECOMMENDATIONS To Table of Contents
"So far as the laws of mathematics refer to reality, they are not certain. So far as they are certain, they do not refer to reality" ................Albert Einstein (1879-1955).
The main objectives of the doctoral research were to outline and test an approach for assessing the performance of spatially distributed catchment models and to quantify the uncertainty associated to the model outputs, given current limitations in terms of data availability and differences among the scales at which the physical structure of the model was derived, the scales at which the data are collected and the scales at which medium size catchments are modelled. To accomplish previous, research was mainly based on the assessment of the relevance of some key input and modelling variables on the simulation outputs and the magnitude of the effective parameters, the assessment of the sensitivity of certain components of the MIKE SHE model structure to the accuracy of these input and modelling variables and the realisation of the spatially physically based hydrological code MIKE SHE in a simple stochastic context. In this respect, the following research activities were developed: (i) the definition of appropriate evaluation statistics and evaluation tests to assess the performance of distributed models; (ii) the application of sensitivity analyses inspecting some input and modelling variables; and (iii) the definition of simulation bands characterising the total uncertainty associated to model outputs. Although those activities are part of a holistic modelling protocol, an attempt was made to depict separately their relevance through the different capitulos of the dissertation.
Apendice A: ASSESSMENT OF DTM QUALITY FOR HYDROLOGICAL USE A la tabla de contenidos
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
A.1).
Fig. A.1: Flowchart for the creation of
DTMs (Vázquez et al.,
2002).
A publication related to this apendice is Feyen et al.(2004b).
Apendice B: PROCESSING OF THE SOIL DATA To Table of Contents
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
B.1).
Fig. B.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
A publication related to this apendice is Feyen et al.(2000).
apendice C: LITHOSTRATIGRAPHY OF THE STUDY SITE To Table of Contents
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
C.1).
Fig. C.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
A publication related to this apendice is Feyen et al.(2000).
apendice D: POTENTIAL EVAPOTRANSPIRATION (ETp) FOR THE DISTRIBUTED MODELLING OF BASINS A la tabla de contenidos
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
D.1).
Fig. D.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
Publications related to this apendice are Feyen et al.(2003a), Vázquez et al.(2003b) and Vázquez et al.(2004a).
apendice E: BRIEF QUALITY ASSESSMENT OF THE CLIMATIC DATA FOR ESTIMATING ETp A la tabla de contenidos
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
E.1).
Fig. E.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
Publications related to this apendice are Feyen et al.(2003a), Vázquez et al.(2003b) and Vázquez et al.(2004a).
apendice F: FILTERING OF SUBFLOWS FROM TOTAL STREAMFLOW TIME SERIES To Table of Contents
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
F.1).
Fig. F.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
Publications related to this capendice are Feyen et al.(2000), Vázquez et al.(2002) and Vázquez et al.(2004a).
apendice G: GENERALISED ESTREME VALUE ANALISYS (GEV) CONSIDERING THE POT APPROACH A la tabla de contenidos
In principle, spatially distributed codes can accept experimental data at each grid element or calculation node, predicting the response of catchments based on the knowledge of its physical properties without the need of model calibration. Those models would allow the prediction of the hydrological responses of ungauged catchments and of the changing responses resulting from changes in land characteristics. In practice, because of sub-grid heterogeneity of parameter values, differences between measurement and model grid scales, and experimental and financial constraints, specification of parameter values is difficult. Indeed, in most cases the experimental techniques to measure or estimate the model parameter values at the scale of the grid element do not exist. Thus, effective model parameters have to be defined normally through model calibration. Traditional evaluation by means of a Split-Sample (SS) test still remains in many practical cases the only attainable option; however, this method is inconsistent with spatially distributed modelling.
A brief description of both, the study site as well as the hydrological code used in this study are given in this capitulo.
The study site is the Gete catchment,
located in Belgium (Fig.
G.1).
Fig. G.1: Location of the study site within Belgium and a detailed view
of the modelled area (Vázquez
et al., 2002).
A publication related to this apendice is Vázquez et al.(2003a).