Geo-MHYDAS: A landscape discretization tool for distributed hydrological modeling of cultivated areas $ P. Lagacherie a,n , M. Rabotin a , F. Colin b , R. Moussa a , M. Voltz a a INRA, Laboratoire d’e ´tude des Interactions Sol-Agrosyst  eme-Hydrosyst  eme (LISAH) UMR INRA-IRD-SupAgro n1 1221, 34060 Montpellier, France b SupAgro, Laboratoire d’e´tude des Interactions Sol-Agrosyst  eme-Hydrosyst  eme (LISAH) UMR INRA-IRD-SupAgro n1 1221, 34060 Montpellier, France article info Article history: Received 24 March 2009 Received in revised form 3 October 2009 Accepted 15 December 2009 Keywords: Landscape Hydrology Discretization Topology GIS Cultivated catchment abstract The representation of landscape variabilities by means of an adequate landscape discretization is of major importance in distributed hydrological modeling. In this paper, we present Geo-MHYDAS, a landscape discretization tool that allows explicit representation of landscape features, particularly man-made ones, that are known to have a great impact on water and mass flows across the landscape. The landscape discretizations that are produced include user-controlled delineated irregular, linear or areal units connected to each other along a tree-like structure. Geo-MHYDAS includes three steps: (i) processing (i.e., the importation or the creation), followed by the modifications of the geographical objects, the limits of which are considered in the modeling as hydrological discontinuities, (ii) creation of the areal and linear units for hydrological modeling by a ‘‘selective cleaning’’ procedure after overlay that preserves, as much as possible, the object limits defined in the previous step, while having sizes and shapes that remain compatible with the application of the water flow functions of the hydrological model and (iii) building an oriented topology between irregularly shaped areal and linear units that allows the routing of the simulated water flows across the landscape. Geo-MHYDAS was developed using the open source free Geographic Information Systems (GIS) software GRASS. The use of Geo-MHYDAS was illustrated by running the hydrological model MHYDAS for different scenarios of man-made features, their presence and spatial organization within a small vineyard catchment located in the south of France (the Roujan catchment). Comparisons with hydrological modeling performed with usual landscape discretizations showed significant differences in the simulated hydrograms. This comparison illustrates well the strong impact of landscape discretizations on hydrological modeling, specifically on the man-made landscape features represented in Geo-MHYDAS. & 2010 Published by Elsevier Ltd. 1. Introduction Distributed hydrological models provide effective simulation tools for exploring basin hydrological processes and predicting the effects of changes on catchment response. Each model has its own method to represent landscape variabilities, which strongly influences its performance and often limits its use. Therefore, it is important to provide adequate spatial discretizations of land- scapes that could be coupled with physically based representa- tions of hydrological processes. This is all the more true in cultivated landscapes, where the number of man-made features (e.g., ditches, field boundaries, benches or hedges), recognized as strongly influencing water flows, must be taken into account (Louchart et al., 2001; Carluer and De Marsily, 2005). Grid-based segmentation of the landscape is commonly used by hydrological distributed model because grid-based Digital Elevation Models (DEM) are the most commonly available, the methods of analysis are computationally efficient and simple and the structure is compatible with remote sensing techniques and GIS. Furthermore, the coupling with the equations that represents hydrological processes is straightforward. Some examples of models based on this structure include the Areal Nonpoint Source Watershed Environment Response Simulation model (ANSWERS) (Beasley et al., 1980), the Syst  eme Hydrologique Europe ´ en model (SHE) (Abbott et al., 1986a, b), the TopModel (Beven et al., 1987), the Agricultural Non-Point Source pollution model (AGNPS) (Young et al., 1989) or the Limburg Soil Erosion Model (LISEM) (De Roo and Offermans, 1995). However, the elementary pixels considered by grid-based landscape representations do not correspond to real landscape features as recognized by landscape surveyors, which results in unsatisfactory solutions for representing irregularly sized landscape features and gradi- ents. This is all the more true in cultivated landscapes where many human-made features are to be considered as hydrological Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/cageo Computers & Geosciences 0098-3004/$ - see front matter & 2010 Published by Elsevier Ltd. doi:10.1016/j.cageo.2009.12.005 $ Code available from authors’ web site www.umr.lisah.fr/openfluid and on server at http://www.iamg.org/CGEditor/index.htm. n Corresponding author. Tel.: + 33 4 99 61 25 78; fax: + 33 4 67 63 26 14. E-mail address: lagache@supagro.inra.fr (P. Lagacherie). Computers & Geosciences 36 (2010) 1021–1032