The land data assimilation (DA) used in NOAA’s global numerical weather prediction (NWP) system is much less advanced than that used at other major international NWP centers, and as a part of the GFSv17 upgrade we are developing a new state-of-the-art land data assimilation system. This seminar will review the planned design of the new land data assimilation system, and progress towards its development and implementation. The first priority for the new land data assimilation system is to replace the current snow depth analysis. The current analysis is quite outdated, and consists of a simple rule-based merging of an externally generated snow depth product. This is being replaced with an Optimal Interpolation (OI) snow depth analysis, based directly on the methods used at other NWP centers. Tests of the snow depth OI with GFSv16 (with the current land surface model, Noah) showed that it significantly improves the model simulated snow depth, while generating small but consistent improvements to the simulated atmospheric temperatures over snow-affected land. Based on these tests, we are preparing the snow depth OI for use in GFSv17. This has included adapting the OI to the Noah-MP land surface model (which will replace Noah in GFSv17), and also implementing the OI within the JCSDA’s JEDI data assimilation platform. The second priority of the new land data assimilation system is to introduce a soil moisture and soil temperature analysis. Currently, NOAA does not apply a snow analysis in our global NWP systems, while other centers have done so for decades. For the soil analyses, we are developing a Local Ensemble Transform Kalman Filter (LETKF) assimilation, initially based on assimilation of screen level temperature (T2m) and specific humidity (q2m). Early tests with GFSv16 using the LETKF to update the model soil temperature from T2m observations show very small improvements in the subsequent simulations of T2m, with negligible effect above the surface. Additionally, the impact of the assimilation is limited by the difficulty of obtaining sufficient ensemble spread without introducing biases into the ensemble mean. Work is ongoing to address this issue.
