This paper aims at characterizing the groundwater flow in a highly dynamic karst aquifer using a global modeling approach based on rainfall and spring discharge time series. The Dardennes aquifer (SE France) was studied as it is used for drinking water supply and it also produces karst flash floods that increase the flood hazard downstream in urban areas. Three years of data were available, including a normal rainy year, a wet year and a dry year. Modeling was performed with the new platform KarstMod, a rainfalldischarge model with calibration tools. The Dardennes aquifer model was structured with three interconnected reservoirs: Epikarst, Matrix, and Conduit. Using this modeling approach, we were able to determine the groundwater hydrograph separation of the karst spring discharge, at the annual scale and at the event scale (flood). This gives insight into the low flow (Matrix) available for the drinking water demand and the fast flow (Conduit) that generates flash floods. In such a dynamic aquifer, part of the water budget cannot be accounted for by water resources as fast flow is not stored within the aquifer and is not available for the drinking water demand. The results were compared with the current groundwater management to determine whether the withdrawal is sustainable. Depending on whether it is a wet or a dry year, the proportion of slow flow ranges from 27 to 61% of the total discharge, respectively. During floods in high water periods, the proportion of quickflow increases drastically up to more than 90% of the spring discharge. In the case of a 300 mm/d simulated Mediterranean rainfall event, the mean daily peak value may reach 74m3/s. This discharge can be reduced if the aquifer is previously depleted, which increases the storage within the aquifer. Coupling the geological context and the model results opens up future perspectives for the active management of the karst aquifer.

Karst systems are hierarchically spatially organized three-dimensional (3D) networks of conduits behaving as drains for groundwater flow. Recently, geostatistical approaches proposed to generate karst networks from data and parameters stemming from analogous observed karst features. Other studies have qualitatively highlighted relationships between speleogenetic processes and cave patterns. However, few studies have been performed to quantitatively define these relationships. This paper reports a quantitative study of cave geometries and topologies that takes the underlying speleogenetic processes into account. In order to study the spatial organization of caves, a 3D numerical database was built from 26 caves, corresponding to 621 km of cumulative cave passages representative of the variety of karst network patterns. The database includes 3D speleological surveys for which the speleogenetic context is known, allowing the polygenic karst networks to be divided into 48 monogenic cave samples and classified into four cave patterns: vadose branchwork (VB), water-table cave (WTC), looping cave (LC), and angular maze (AM). Eight morphometric cave descriptors were calculated, four geometrical parameters (width-height ratio, tortuosity, curvature, and vertical index) and four topological ones (degree of node connectivity, α and γ graph indices, and ramification index) respectively. The results were validated by statistical analyses (Kruskal-Wallis test and PCA). The VB patterns are clearly distinct from AM ones and from a third group including WTC and LC. A quantitative database of cave morphology characteristics is provided, depending on their speleogenetic processes. These characteristics can be used to constrain and/or validate 3D geostatistical simulations. This study shows how important it is to relate the geometry and connectivity of cave networks to recharge and flow processes. Conversely, the approach developed here provides proxies to estimate the evolution of the vadose zone to epiphreatic and phreatic zones in limestones from the quantitative analysis of existing cave patterns.