Are Dynamically Evolving Models the Future of Hydrologic Modelling? A Data Assimilation Approach

Tuesday, 16 December 2014
Sahani Darshika Pathiraja1, Lucy Amanda Marshall1, Ashish Sharma1 and Hamid Moradkhani2, (1)University of New South Wales, School of Civil and Environmental Engineering, Sydney, NSW, Australia, (2)Portland State University, Civil and Environmental Engineering, Portland, OR, United States
Can a single model adequately represent a catchment’s runoff response? Is it possible to design a modelling framework that can adequately represent catchment processes now and in the future? These are just a few of the questions related to catchment modelling that continue to intrigue hydrologists. An appreciation of catchment non-stationarity is critical to answering these questions. However this has been given little consideration in the past. Typically a single model parameterisation is adopted and assumed to be adequate outside the period of calibration/validation. We present a novel method for dealing with catchment non stationarity that allows model parameterisations to evolve in time through a Data Assimilation framework. Data Assimilation for hydrologic parameter estimation has mainly focused on deriving a stationary parameter distribution. We investigate the potential for Data Assimilation to estimate time varying model parameters, an application which has not been examined before in the hydrologic sciences. This is undertaken using the popular Ensemble Kalman Filter with the Probability Distributed Model (PDM), a lumped conceptual hydrologic model. It is shown that careful consideration of the artificial parameter evolution step is critical in this context. A range of traditional artificial parameter evolution techniques are considered and found to be problematic when parameters exhibit temporal variability. A new hierarchical approach is presented that relies on a more meaningful description of parameter evolution to improve the prior parameter distribution. The hierarchical approach is applied to multiple known case studies and shown to produce superior results when compared to traditional parameter evolution techniques. This method provides a general framework for establishing dynamically evolving models that will be particularly useful for rapidly changing catchments.