Land–Atmosphere Covariances for Coupled DA from Ground-Based Observations and NOAA's Next Generation Global Prediction System

Coupled land/atmosphere data assimilation could benefit atmospheric forecasts by extracting skill from lower-frequency land states, and allowing a more seamless prediction across a range of time scales. It could also benefit land surface forecasts by updating the model land states with observed information from the comparably well-observed atmosphere. However, coupled data assimilation faces a number of difficulties, including differences in the temporal scales of different Earth system components, large uncertainties in the simulated cross-component covariances (and error covariances), and the potential for transmission of biases between components. In this study, we will examine the observed and modeled covariances between time series of land and low-level atmosphere states, and use the results to guide the design of an assimilation strategy for coupled land/atmosphere data assimilation that ameliorates the above-listed difficulties. Specifically, we will examine the observed and modeled anomaly covariances between well-observed variables that could potentially be assimilated (near-surface soil moisture, land surface temperature, screen-level temperature and relative humidity) and land surface model variables that could potentially be analyzed (soil moisture and temperature states). The observed covariances will be estimated from ground-based USCRN observations at close to 120 locations across the contiguous US, while the modeled covariances will be estimated from NOAA’s Next Generation Global Prediction System (NGGPS) at the same times and locations. Additionally, we will examine whether the inclusion of temporal operators, such as averaging or differencing, in the observation operator could help to address the differences between the time scales of the land and atmosphere, and/or the large biases ubiquitous to many modeled land surface states. The understanding obtained in this study of the observed land/atmosphere covariances and of the realism of the modeled covariances, will directly contribute to the design of an ensemble-based land/atmosphere data assimilation system, to be developed and tested for use in NOAA’s NGGPS. Specifically, the outcomes of this study will inform the selection of observations to be assimilated and any temporal operators applied to those observations, as well as the design of the model forecast error covariances within the data assimilation (the latter through ensemble simulations, possibly combined with static error covariances).