We test how different groups of observations, which reach the computer that run weather models at different times, impact the model forecast. We find that by not using the observations that arrive late, the forecast is much worse, showing the importance of quick transmission of observations. Direct observations (those measured within the atmosphere) have a greater impact than remote observations (those viewed from afar, such as by satellites).
Millions of observations are used in global weather forecast models every day. The use of these observations relies on their quick transmission to where the models are run. This study looks at the impact of two datasets, one with all available observations and one smaller one representing what was received by NOAA by a particular time, on the accuracy of NOAA’s Global Forecast System (GFS). The GFS is run four times each day (every 6 h); the model starts from what we call initial conditions, which is the combination of all the most recent weather observations and a short forecast from the previous run (or cycle). Typically, the smaller dataset is used to create initial conditions for 14-day forecasts, and the full dataset, containing about 20% more data, is used to create the short forecast for the next cycle.
- Using the full dataset for both forecasts yields a small, statistically significant improvement in forecast skill.
- Using the smaller dataset for both 14-day and 9-hour forecasts leads to a large degradation in forecast skill.
- The greatest impact comes from removing observations taken from direct contact with the atmosphere, such as from weather balloons and airplanes.
- If a single observation source’s time to reach the model increases, the impact on forecast skill may not be significant; however, if the time increases across all observation types (or all observations of a single type), the impact can be large.
For more information, contact email@example.com.
This study was partially funded by the NESDIS OPPA Technology Maturation Program, and supported by NOAA/OAR/AOML and U.Miami/CIMAS grant NA15OAR4320064.