Re-thinking Sensitivity of Model Parameter Values in Soil Moisture Assimilation Using the Evolutionary Data Assimilation

The sensitivity of land surface model parameters is usually examined for one parameter at a time in response to changes in observation data and/or the model estimated output. This parameter independence approach assumes that there are limited interactions between model parameters a precondition which is highly unlikely for land surface models. Additionally, the model parameter values are widely assumed as time-invariant, particularly, in most data assimilation (DA) studies where model states are updated in response to changes in observation data. However, a demonstration of time-variant nature of physically representative model parameters values has not been thoroughly investigated in the DA literature. The sensitivity of model parameter values is important because their variability constitute an integral component of the overall accuracy of land surface model outputs for quantities such as soil moisture. This study assimilates the Soil Moisture and Ocean Salinity (SMOS) Level 2 soil moisture data into the Joint UK Land Environment Simulator (JULES) to investigate the sensitivity of its model parameter values. The study demonstrates the sensitivity of JULES model parameter values in the estimation of soil moisture through the Evolutionary Data Assimilation (EDA) procedure. The EDA employs stochastic and adaptive properties of multi-objective evolutionary strategies to evaluate the continuous interaction between several model parameters. Analysis of these parameter sets across all assimilation time steps allows the sensitivity of model parameter values in response to changes in observation data to be determined. This approach is evaluated through comparison to conventional soil moisture assimilation for dual state-parameter estimation using the Ensemble Kalman Filter (EnKF).