Abstract:
Over recent decades, the emergence of satellite missions dedicated to the observation of
land surface states and processes has created an excellent opportunity to improve land
surface models (LSMs) by confronting them with observations. Options are to either
directly improve LSM simulations through state data assimilation (DA) or to improve
overall model skill through parameter estimation or analysis of DA results to inform model
structural developments. While land DA is undoubtedly a valuable tool, questions
regarding how to maximize the observation impact and fully exploit DA’s potential for
Earth System Model development remain. In this presentation, I will highlight some of
these questions through a number of case studies. First, I will investigate how the
complementary information provided by multiple satellite sensors can be used most
efficiently in a DA framework, by comparing the simultaneous assimilation of individual
soil moisture retrieval products from an active and passive sensor to the assimilation of a
joint retrieval. Next, I will examine the potential of using land DA to partially correct model
systematic errors in addition to random errors through the use of a machine learning based
bias correction approach. Third, I will discuss the advantages and caveats of assimilating
land observations to optimize model parameters using the example of a vegetation
parameter calibration study. Finally, I will explore the potential of land DA to improve
model forecasts in an operational framework and will examine the scenarios in which land
DA could most contribute to the forecast skill. Over recent decades, the emergence of
satellite missions dedicated to the observation of land surface states and processes has
created an excellent opportunity to improve land surface models (LSMs) by confronting
them with observations. Options are to either directly improve LSM simulations through
state data assimilation (DA) or to improve overall model skill through parameter
estimation or analysis of DA results to inform model structural developments. While land
DA is undoubtedly a valuable tool, questions regarding how to maximize the observation
impact and fully exploit DA’s potential for Earth System Model development remain. In this
presentation, I will highlight some of these questions through a number of case studies.
First, I will investigate how the complementary information provided by multiple satellite
sensors can be used most efficiently in a DA framework, by comparing the simultaneous
assimilation of individual soil moisture retrieval products from an active and passive
sensor to the assimilation of a joint retrieval. Next, I will examine the potential of using land
DA to partially correct model systematic errors in addition to random errors through the
use of a machine learning based bias correction approach. Third, I will discuss the
advantages and caveats of assimilating land observations to optimize model parameters
using the example of a vegetation parameter calibration study. Finally, I will explore the
potential of land DA to improve model forecasts in an operational framework and will
examine the scenarios in which land DA could most contribute to the forecast skill.