The Impact of Reducing the Assimilation Interval in the NOAA Global Data Assimilation System

Most operational centers use a 6-hour assimilation window for global weather prediction, creating analyses at the ‘synoptic’ times of 00,06,12 and 18 UTC. Recently, a study by Slivinski et al showed that an overlapping-window approach could successfully be applied to deal with observational data latency (the delay between when an observation is taken and when it arrives at the operational center) in a rapidly-updating version of the NOAA operational global data assimilation system. This approach was used to reduce the assimilation window from 6 hours to 1 hour without any degradation in performance, despite the fact that most of the observations were not available at the operational center within an hour of when they are taken. In fact, some evidence was presented showing that an hourly cadence actually improved the fit of short-term forecasts to observations, suggesting that assimilating observations closer to the time they were taken is beneficial. However, the issue of observational data latency and data receipt cutoffs makes it difficult to design ‘apples-apples’ experiments to identify the impact of changing the assimilation window. In the study we have eliminated the role of data latency by assuming all observations can be used as soon as they are taken in the GDAS. We will present results of experiments with varying window lengths (from 6 down to 1 hour), comparing the skill of forecasts informed by the same observations to measure the impact of reducing the assimilation interval. The difference between these results and results that take into account observational data latency can be used to assess the impact of investments to improve network transmission capabilities that will allow observations (particularly from satellites) to arrive at the operational centers more quickly.