Exploring strongly coupled land/atmosphere data assimilation for numerical weather prediction

Abstract The National Oceanic and Atmospheric Administration (NOAA) is currently the only major numerical weather prediction (NWP) center not to have soil‐temperature and soil‐moisture analysis in its global NWP system. Rather than developing a separate soil‐analysis scheme, as has been done elsewhere, this study explores the possibility of introducing soil analysis by (i) expanding the control vector of a pre‐existing ensemble‐based atmospheric data assimilation (DA) and (ii) adding the assimilation of screen‐level temperature () and humidity () observations. Here, we use an ensemble Kalman filter (EnKF) to test the viability of this approach, before we invest the effort to expand our full atmospheric DA to perform the soil analysis also. Even without assimilating the screen‐level observations, adding the soil states to the EnKF control vector induces small, but significant, improvements in the fit between the model and the screen‐level observations (1% reduction in the root‐mean‐square difference (RMSD) for and ). By using the same method for the land and atmospheric DA, we can share information more easily between the two components, including the possibility of applying more strongly coupled DA. In our experiments, the best results were obtained from a strongly coupled land/atmosphere DA (screen‐level RMSD reduction 4% for , 7% for ), and we are now exploring this option at NOAA. However, our weakly coupled experiment, in which the standard suite of atmospheric observations was assimilated into the atmosphere and the screen‐level observations were assimilated into the land (soil), had similar results to the strongly coupled DA experiment (screen‐level RMSD reduction 3% for , 5% for ). These results highlight that weakly coupled land/atmosphere DA is also an attractive option for NWP, while additionally having the flexibility to configure the DA separately to best fit the specific behavior of each model component.