How to Quantify the Quality of Karst Network Models?

Abstract:

In the last years, new models have been developed for the geometry of karstic conduit networks. Some are based on purely statistical techniques, others are based on process based methods. Are the resulting models properly representing the main features of real karstic systems?

In this talk, we introduce a set of statistical indicators to characterize the geometry and topology of karstic networks. These indicators have been applied on a relatively large set of real karst networks. We then use these measures to compare synthetic karstic networks with their field counter parts. We use a supervised learning algorithm to automatically classify any new simulated network based on its statistical properties. Issues such as incomplete mapping of the karst system or systematic bias due to exploration conditions are discussed. We show that the method is able to estimate objectively how distant the simulated network is from the real karst networks. This distance can then be used as a quality measure of the simulation and may indicate the direction to take in order to improve future simulations.