Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Royaume-Uni → 2010 → Hydrological modelling in data-sparse snow-affected semiarid areas

Imperial College London (2010)

Hydrological modelling in data-sparse snow-affected semiarid areas

Mirshahi Babak

Titre : Hydrological modelling in data-sparse snow-affected semiarid areas

Auteur : Mirshahi Babak

Université de soutenance : Imperial College London

Grade : Doctor of Philosophy (PhD) 2010

Résumé
In this study, using daily data from a region in northeast Iran, the problem of hydrological modelling in snow-affected mountainous semiarid areas is addressed, and the potential of an approach is evaluated. Since in such regions the inputs to rainfallrunoff (RR) models, i.e. rainfall, temperature and snowmelt are spatially variable, a semidistributed model is required. However, since observed input data are usually limited at the required spatial scale, such variables first need to be modelled. For this, a stochastic approach, namely Generalised Linear Models (GLMs) is used for precipitation modelling over a 2700 km2 region. The snow water equivalent time series are estimated using 6 snowmelt algorithms, with different levels of complexity. The significance of these inputs on runoff generation are investigated using a conceptual rainfall-runoff (RR) model, IHACRES, which itself is applied at three levels of complexity.
The results of the GLM-based precipitation model are generally found promising. The results indicate that rainfall properties including extremes are generally well captured by the models, particularly if spatial dependence effects are included. Although the annual maximum daily rainfall tends to be overestimated at a few sites, performance is generally satisfactory at more than 75% of sites. Unlike previous results for humid climates, effects of temporal memory are limited to the previous day’s rainfall. For the period 1968-2007, the observed rainfall data follow a cyclical pattern with a period of around 5.5 years, mainly due to spring rainfall periodicity. Teleconnection indices including the El Niño-Southern Oscillation are not found significant in the modelling, and did not account for the inter-annual periodicity.
The RR model with the least number of snowmelt parameters is found the best model structure. Setting the interacting parameters to a fixed value improves the model performance. Although, the performance for the simple model of Jakeman & Hornberger (1993) is found superior to the model by the Ye et al (1997), a longer calibration period is required before being able to confirm this. As for snowmelt estimation, a simple temperature index method has a better performance than more complex methods which use satellite-derived snow-cover data. It is also found that the variable degree-day factor performs better than the fixed one. The simulated stochastic precipitations are well estimated for the low flows. Hydrograph properties for high flows are generally well captured, though the accuracy of estimated peak flow varies between precipitation realisations.
Overall, precipitation properties are generally well modelled (in over %75 of sites) by the GLM-based precipitation model. The thesis demonstrates that for data-sparse mountainous snow-affected semiarid regions, although more complex RR models may structurally be more accurate ; in combination with uncertain simulated inputs, they have a worse performance. This research therefore indicates that generally the simpler RR and snowmelt models perform better in such regions than more complex methods. The simplified semi-distributed IHACRES, in combination with a simple snow meltaccumulation model, which in total requires 5 parameters, potentially is found appropriate for runoff estimation in the region ; modelling needs to be repeated by a longer data, before being able to confirm.

Version intégrale

Page publiée le 13 mars 2015, mise à jour le 1er février 2017