Daniel Bittner, Steven Mattis, Barbara Wohlmuth, Gabriele Chiogna and I prepared a document on the calibration of parameters for a lumped karst aquifer model by discharge data. We used the active subspace method to study the degree at which the parameters (more precisely, the directions in the parameter space) are informed by data.
I have to thank all the contributors for putting much effort in this preprint. I am looking forward in continuing and refining the research with these guys.
Abstract. Hydrological models of karst aquifers are generally semi distributed and physical processes such as infiltration and spring discharge generation are described in a lumped way. Several works previously addressed the problems associated with the calibration of such models, highlighting in particular the issue of model equifinalty and the importance of multi-objective optimization. In this work, we investigate the problem of model calibration from a different point of view. We apply the active subspace method, to investigate how the parameters of a newly proposed hydrological model for karst aquifers (LuKARS, Land use change modeling in KARSt systems) are informed by discharge data. We showed that each one of the three hydrotopes (i.e. distinct landscape units characterized by homogeneous hydrological properties as a result of similar land use and soil types) used in the model are similarly informed by the data. We found that all the 21 model parameters are informed by the measured data and therefore it is not immediately possible to apply any model reduction scheme. The full-dimensional inverse problem is formulated and solved to identify local optima for the misfit function.