Improving the Observability of Soil Moisture Estimates from Noah-MP Land Surface Model

Systematic errors in the modeled background values can significantly limit the utility of data assimilation applications. These differences arise due to multiple reasons, including the differences in the spatiotemporal representations of modeled and observed values, the actual measured quantity being a proxy or a retrieval of the modeled variable, and the model equations being an inexact reproduction of the real-world physics. When there are biases between the observations and the modeled background, data assimilation methods that depend on specific conditions for optimality are usually not satisfied. The current practice in soil moisture data assimilation is to employ rescaling approaches that ignore the systematic differences and focus on the assimilation of anomaly information. Recent studies have shown that such approaches have fundamental limitations, particularly when unmodeled processes are present in the observations being assimilated.

To eliminate the model versus observation bias as preparation for a soil moisture assimilation application, we explore the utility of calibrating the parameters of a land surface model (LSM) to in-situ observations of soil moisture. We use observations from 1345 sites of the International Soil Moisture Network (ISMN) over the Continental United States (CONUS) to calibrate the Noah-Multiparameterization (Noah-MP) LSM version 3.6. We use a Genetic Algorithm to get deterministic calibrated parameter values of Noah-MP for different soil types as a new parameter table. Results from a systematic evaluation of the calibrated model documenting the added utility in soil moisture estimates and the impact on other model states will be presented. The development of a model configuration that generates soil moisture estimates consistent with ground measurements is a necessary step to improving the observability of model simulations and efficiency of data assimilation environments.