Sensitivity to space and time resolution of a hydrological model using digital elevation data

Abstract

The space and time resolutions used for the input variables of a distributed hydrological model have a sufficient impact on the model results. This resolution depends on the required accuracy, experimental site and the processes and variables taken into account in the hydrological model. The influence of space and time resolution is studied here for the case of TOPMODEL, a model based on the variable contributing area concept, applied to an experimental 12 km2 catchment (Coet-Dan, Brittany, France) during a two month winter period. A sensitivity analysis to space and time resolution is performed first for input variables derived from the digital elevation data, secondly for the optimized values of the TOPMODEL parameters and finally for modelling efficiency. This analysis clearly shows that a relevant domain of space and time resolutions where efficiency is fairly constant can be defined for the input topographic variables, as opposed to another domain of larger resolutions that induces a strong decrease of modelling effciency. It also shows that the use of a single set of parameters, defined as mean values of parameters on this relevant domain of resolution, does not modify the accuracy of modelling. The sensitivity of the parameters to space and time resolution allows the physical significance of the parameter values to be discussed.

Scalability of grid- and subbasin-based land surface modeling approaches for hydrologic simulations

“…That is, instead of establishing a critical scale for optimal prediction [Wood et al, 1988;Famiglietti and Wood, 1994;Bruneau et al, 1995;Woods et al, 1995;Liang et al, 2004;Shrestha et al, 2006], we evaluate the ability of the two modeling approaches to produce robust predictions across multiple spatial scales. For this purpose, we systematically compare the two modeling approaches: (1) to determine if they differ in their scalability in hydrologic flux simulations across multiple spatial resolutions; (2) to explore the sources of their scalability differences; and (3) to determine the significance of their scalability differences.…”

Resolution Sensitivity of a Compound Terrain Derivative as Computed from LiDAR-Based Elevation Data

“…"This non-linear effect of space resolution may be due to differing effects on the two variables [gradient, SCA] used in determining the topographic index" [6]. This change in the shape of the TWI distribution was also reported by Saulnier et al [8].…”

“…Bruneau et al [6] observed a change in the shape of TWI statistical distributions that may affect model runs within TOPMODEL. "This non-linear effect of space resolution may be due to differing effects on the two variables [gradient, SCA] used in determining the topographic index" [6].…”

“…Since TOPMODEL is semi-distributed there is no difference in the computational effort, after calculation of the TWI, between a model run at a resolu- tion of 1000 m or 1 m. As a result, various studies have examined the use of TOP-MODEL at a range of resolutions [4][5][6][7][8][9][10], though to our knowledge only one paper has applied LiDAR-based elevation data with resolutions of ~2 m [11]. Much existing research into the resolution sensitivity of the TWI has been performed in order to understand the resolution sensitivity of TOPMODEL and has therefore focused only on the sensitivity of the (aspatial) distribution of the TWI, since this is the only TWI property used in TOPMODEL.…”

Génèse des débits dans les petits bassins versants ruraux en milieu tempéré : 2 - Modélisation systémique et dynamique

“…-le résultat du calage dépend largement du choix de la méthode d'ajustement, de la fonction-objectif et des périodes et variables de référence, voire même de la discrétisation choisie : ainsi, le calage de TOPMODEL donne des valeurs de conductivite hydraulique croissant avec la taille des mailles du MNT utilisé -ce qui explique les valeurs irréalistes fréquemment obtenues pour ce paramètre de calage lorsque cet effet n'est pas corrigé (BRUNEAU et al, 1995 ;SAULNIER étal., 1997); -contrairement à ce qui est souvent supposé, interdépendances et nonlinéarités rendent généralement la surface de réponse peu lisse voire discontinue, même pour des modèles à peu de paramètres : les multiples extrema locaux rendent plus difficile la détection de l'extremum absolu ; -la « surparamétrisation » de la plupart des modèles spatialisés -par rapport à la quantité des données disponibles pour le calage -peut conduire à une forte indétermination, les rendant non identifiables : des simulations très semblables peuvent être obtenues avec des jeux de paramètres calés très différents, du fait de compensations liées à l'interdépendance des paramètres : comment alors identifier le « bon » jeu de paramètres ?…”