Adaptive Estimation of ATMS Observation Uncertainty to Improve Atmospheric Prediction

Abstract The present era is marked by an unprecedented number of novel observations for data assimilation (DA) systems. The abundance of new observations necessitates finding optimal ways to use them. One key challenge is the estimation of their uncertainty. In this study, we allow the uncertainty of Advanced Technology Microwave Sounder (ATMS) radiance observations to vary spatially and/or temporally. ATMS is a sensor that provides temperature and humidity profiles in all-weather conditions, as well as precipitation information. We estimate observation uncertainty using the well-known Desroziers method, which uses observation-minus-analysis ( O − A ) and observation-minus-background ( O − B ) values from a cycling DA system to produce an updated observation error covariance matrix . We adaptively estimate the uncertainty (in terms of standard deviation) of each channel, as well as the interchannel error correlations, for ATMS observations across various geographic regions and time. Next, the impact of adaptively estimating the ATMS observation uncertainties is evaluated in the U.S. Navy Global Environmental Model (NAVGEM) forecasts initialized from analyses using the updated . The analysis shows that using derived from global O − A and O − B values from a 3-month period can lead to statistically significant improvements in temperature, moisture, and geopotential height forecasts up to 3% relative to using the current operational static . Finally, the use of derived from O − A and O − B values from the previous 10 days of cycling can further improve NAVGEM forecasts of temperature and moisture by over 1%. Significance Statement Numerical weather predictions benefit greatly from information collected by satellites. Observations of radiation emitted by the atmosphere are assimilated into our models to generate better initial conditions for forecasts. To do so, we need to know how uncertain the satellite observations are relative to prior predictions and other observations. As the number of active satellites rapidly grows, we can ingest a greater wealth of information. At the same time, it becomes increasingly challenging to keep accurate estimates of the observation uncertainty. To this end, we employ a well-known diagnostic that estimates the observation uncertainty of satellite observations online using prior values of observation-minus-forecast departures. We then show the improvements brought to global forecasts made possible by this diagnostic.