Abstract:
Accurate estimates of model and observation error parameters are key ingredients of a data assimilation system. To date, two main approaches for obtaining model error parameters have been used in soil moisture data assimilation. The first approach derives model error parameters by comparing the open loop trajectory to validating measurements (from field or synthetic data) outside of the cycling data assimilation system. The second approach is based on repeating the entire data assimilation experiment many times over with differents sets of model error parameters, that is, the model error parameters that produce the best validation of assimilation estimates with the cycling assimilation system are selected by enumeration. While theoretically correct, the second approach of enumeration is computationally not feasible for large systems. We demonstrate in a fraternal twin experiment for the Red-Arkansas river basin that a computationally affordable, adaptive assimilation system provides improved assimilation estimates. In the adaptive assimilation system, the model error parameters are continually adjusted in cycling assimilation mode in response to the innovation information provided by the observation minus forecast misfits.