Dynamically Evolving Models for Dynamic Catchments: Application of the Locally Linear Dual EnKF to a Catchment with Land Use Change

Catchments are dynamic, constantly undergoing change be it naturally or due to anthropogenic influences. Changes in land surface conditions such as disturbance due to bushfire or erosion, urbanisation, deforestation or afforestation will affect a catchment's hydrologic regime. Models calibrated to pre-change conditions will lead to biased streamflow predictions, unless the change is explicitly accounted for in the model. A modelling methodology that is capable of adjusting its form (for instance, through time varying parameters) as catchments undergo change is therefore needed. We developed a framework for automatically and objectively detecting time variations in model parameters using Data Assimilation. The so called Locally Linear Dual EnKF was previously tested against a range of synthetic case studies and shown to reproduce known temporal variations from assimilating streamflow observations only. In this study, we apply the Locally Linear Dual EnKF to the Wights and Salmon paired catchments in Western Australia. Both were initially forested and monitored for a 3 year period, after which Wights was fully cleared whilst Salmon remained unchanged. The lumped conceptual hydrologic model (PDM) was calibrated over the stationary period and the optimal parameterisation used to initialise the Locally Linear Dual EnKF. Resultant parameter trajectories for the Salmon catchment were relatively stationary, whilst parameters for the Wights catchment were automatically adjusted to produce greater flood peaks, sooner after rainfall, consistent with observations. A significant improvement in both streamflow prediction and catchment soil moisture was obtained with the Locally Linear Dual EnKF, compared to the time invariant parameter case. This application has demonstrated the usefulness of this framework for improving predictions in rapidly changing catchments.