Observing system simulation experiments (OSSEs) are an important tool for evaluating the potential impact of proposed new observing systems, as well as for evaluating trade-offs in observing system design, and in developing and assessing improved methods for assimilating new observations. Over the last three decades, we have conducted extensive OSSEs, early on at NASA's Goddard Space Flight Center, and more recently at the National Oceanic and Atmospheric Administration's (NOAA's) Atlantic Oceanographic and Meteorological Laboratory.1, 2 These experiments correctly determined the quantitative potential for several proposed satellite observing systems to improve weather analysis and prediction prior to their launch, evaluated trade-offs in orbits, coverage, and accuracy for space-based wind lidars (light detection and ranging systems), and were used to develop an approach that led to the first beneficial impacts of satellite surface-wind data on numerical weather prediction.
OSSEs for hurricanes are much more limited and first became possible as numerical models acquired sufficient resolution to simulate hurricanes quasi-realistically. The objectives of these OSSEs are to evaluate the potential impact of new (proposed) observing systems on hurricane track and intensity prediction, evaluate trade-offs in the design and configuration of these observing systems, optimize sampling strategies for current and future airborne and spaceborne observing systems, and evaluate and improve data assimilation and vortex initialization techniques for hurricane prediction.
The first OSSE to evaluate the effect of observing systems on hurricane prediction was one we conducted as part of a series of experiments to determine the potential impact of space-based lidar wind profiles (and other advanced remote sensing systems). OSSEs use a ‘nature run,’ generated by a very realistic numerical model, to represent the real atmosphere. As an example, Figure 1shows the evolution of the first hurricane in the nature run, as it moved toward the southeast coast of the United States and then weakened after making landfall.
Figure 1. Sea-level pressure analyses for the first hurricane in the Finite Volume General Circulation Model (FVGCM) nature runs at 24-hour intervals. Axes show latitude and longitude. Values for isobars (lines of constant pressure) within each panel are in millibars. 00Z: Midnight, Greenwich mean time. L: Low pressure center.
Following a detailed assessment of the realism of the nature run and the differences between that model and the assimilation/forecasting model, we validated the entire OSSE system through a comparison of parallel real data and simulated data experiments. We then used this system to perform parallel assimilation experiments and forecasts to evaluate the impact of space-based lidar wind profiles. As in earlier OSSEs, one of the major metrics for assessing the potential effect of lidar wind data was the anomaly correlation for pressure and height forecasts. We also evaluated a number of additional metrics, such as impact on the central pressure and displacement of cyclones or the landfall of hurricanes.
This evaluation showed a substantial improvement in forecast accuracy resulting from the assimilation of space-based lidar winds. In the Southern Hemisphere, average forecast skill was extended by 12–18 hours, whereas in the Northern Hemisphere, average forecast skill was extended by 3–6 hours. This development was associated with a meaningful (10%) reduction in position error for all cyclones averaged over the globe and all time periods. For very intense cyclones, the reduction of position error exceeded 200km. Figure 2 shows a significant improvement in hurricane landfall prediction as a result of assimilating lidar data. This result was obtained for the first hurricane in the nature run, shown in Figure 1. The predicted landfall position error for this and another tropical cyclone to hit the US mainland in the nature run was improved by approximately 240km.
Figure 2. Illustration of the potential impact of lidar (light detection and ranging) wind data. Green (easternmost track): Actual track from nature run. Red (westernmost track): Forecast beginning 63 hours before landfall with all currently used data. Blue (middle track): Improved forecast for same time period with simulated wind lidar added.
These results demonstrate considerable potential for space-based lidar wind profile measurements. We carried out additional experiments to assess the relative impact of upper- and lower-level winds, as well as to isolate the specific lidar data responsible for the improvements, and the effects of horizontal coverage. These experiments showed that mid-upper-level winds contributed more of the beneficial impact on track forecasts, and that the improvements were lost when only a single line of data was assimilated.
We are now developing new, more realistic OSSEs related to hurricane track and intensity prediction that use very high resolution models to determine the potential impact of unmanned aerial systems, as well as the relative impact of alternative concepts for space-based lidar wind sensors and for polar and geostationary hyperspectral sounders. These OSSEs are also beginning to evaluate the implications of proposed observing systems for hurricane track and intensity prediction, and the design and configuration of these systems. Finally, these experiments are being used to optimize sampling strategies for current and future airborne and spaceborne observing systems and to evaluate and improve data assimilation and vortex initialization methods for hurricane prediction.
NOAA Atlantic Oceanographic and Meteorological Laboratory
Robert Atlas is the director of NOAA's Atlantic Oceanographic and Meteorological Laboratory. He holds a PhD in meteorology and oceanography and was the first person to demonstrate the beneficial impact of quantitative satellite data on weather prediction.
1. R. Atlas, Atmospheric observations and experiments to assess their usefulness in data assimilation, J. Meteorol. Soc. Jpn. 75(1B), p. 111-130, 1997.
2. R. Atlas, Observing system simulation experiments to assess the impact of remotely sensed data on hurricane prediction, Proc. SPIE
8515, p. 85150P, 2012. doi:10.1117/12.927595