Distributed simulation models are valuable tools for addressing challenging questions such as the impact of land use or climate change on water resources, or for assessing the efficiency of mitigation solutions for reducing water pollution. Modelling platforms fasten the development of such models by providing tools to perform ancillary tasks such as input pre-processing or output visualisation, templates for process modules construction and coupling mechanisms to build integrated models from the available modules. Such a platform, called LIQUID (Viallet et al., 2006) is under development by Hydrowide with the collaboration of Cemagref. The main stream of models built in LIQUID follows object-oriented principles: the landscape is discretized into hydro-landscapes (Dehotin and Braud, 2008) representing hydrological functional units, which construction depends on the model objectives, data availability and process representation. They usually form very irregular geometries with surfacic, linear or point elements. A model takes as input data vector descriptions of these hydro-landscapes, querying a PostgreSQL/PostGIS database. This paper deals with the pre-processing operations and difficulties encountered for obtaining these input data using open-source GIS. Input data are derived from Digital Elevation Models (raster), and vector maps such as geology, soil properties, land use, but also the hydrographic network, possible other networks of interest (roads, sewer network) or punctual informations, as for example measurement points. Vector data are usually in the ESRI shape, MapInfo MIF/MID or PostgreSQL/PostGIS (WKT/WKB) formats. Data pre-processing operations include DEM processing (calculations and classification of slope, and flow directions), delineation of subcatchments according to the hydrographic network, successive intersections/fusions of polygon layers (land use/soil properties for instance), verification of the connectivity of the hydrographic network from upstream to downstream, boundary recognition and topology extraction from the hydro-landscape layer. As we could not find a unique open-source GIS providing all the required functionalities, we developed strategies using combinations of several softwares: SAGA GIS or MapWindow GIS for DEM processing, OpenJump, Quantum GIS or PostGIS for vector editing and processing, OpenJump and OrbisGIS for topology extraction. Yet one of the most severe limitations in these vector-processing GIS is the way to verify and handle properly topology constraints. We encountered several problems (failure of PostGIS functions for example) due to slightly incorrect geometries that were sometimes generated by the processing operations themselves. Another limitation is the lack of facilities for developing procedures (PL/pgSQL for example) that are needed to automate the pre-processing operations. Therefore, up to now, some of these operations have been processed either manually, or within the ESRI ArcInfo software. We are considering using the open OrbisGIS platform or GRASS GIS to develop such functionalities. The presentation will highlight the problems through representative examples encountered in real case studies. These case studies deal with the impact of land use features, for examples hedges, ditches, or even urbanization on the hydrological regime of small to medium size catchments, or with the influence of rainfall and soil description on the simulation of flash floods. They cover several spatial and temporal scales. As perspective, the presentation will also give insight into the needs for visualization of model results like GIS temporal extensions or links with statistical computing open-source softwares to visualize spatial time series.