2D non-uniform polygonal meshes allow representation of the impact of landscape elements and small infrastructures on water flows. The initial vectorial mesh, derived from the intersection of several geographical information systems' layers, can have highly non-convex or sliver polygons. These bad-shaped elements compromise accurate numerical flow computation. We propose a flexible divide-and-conquer strategy to decompose polygons into physiographical meaningful parts using shape descriptors to better represent the surface terrain and hydrologic connectivity. We use the convexity index (CI) and the form factor (FF) to consider convex and square like optimum shapes. The strategy was applied to two peri-urban areas whose hydrologic response was simulated using distributed modeling. Good-quality meshes were generated with threshold values of CI≈0.8 and FF≈0.2, and CI≈0.95 and FF≈0.4 for undeveloped and highly urbanized zones, respectively. We concluded that the mesh segmentation facilitates the representation of the spatially distributed processes controlling not only the lumped response of the catchment, but also the spatial variability of water quantity and fluxes within it at medium and small scales.