Superstorm Sandy and the importance of polar orbiting satellites in forecasting

On the 23rd of October, the 18th named storm of the 2012 Atlantic hurricane season, Tropical Storm Sandy, was born in the Central Caribbean. As is common for late-season storms in the Caribbean, Sandy moved northwards across Cuba. The official forecasts from the National Hurricane Center issued on October 23 called for Sandy to turn to the northeast after crossing Cuba, and head into the Central Atlantic. This forecast was based on the output from five of our top six computer models, which all predicted that an upper-level low pressure system in the Central Atlantic would be strong enough to pull Sandy northeastwards. However, the global weather forecast model run by the European Center for Medium-Range Weather Forecasts (ECMWF) showed a disturbing possibility: the upper-level low pressure system in the Central Atlantic would not be strong enough to turn Sandy to the northeast. The hurricane would instead get caught up in the flow associated with a trough of low pressure approaching the U.S. East Coast, and Sandy would get slung into New York or New Jersey on October 29. While the ECMWF model was the best performing model for tracking Atlantic hurricanes in both 2010 and 2011, and had done very well again so far in 2012, the American GFS model had outperformed the ECMWF model several times during the 2012 season. NHC elected to discount the ECMWF forecast for Sandy as an outlier, and went with the forecast from the GFS and other models. By October 25, it was clear that the ECMWF model had the right idea all along. More models were now showing the turn towards New York, and the official NHC forecast now called for Sandy to make landfall in New York or New Jersey on October 29. The ECMWF model's early forecast of a track for Sandy into the Northeast was critical for allowing additional time for residents to prepare for arrival of the devastating storm. So what enabled the ECMWF model to make such an excellent forecast for Sandy, six days in advance?


Figure 1. This image uses the model output from the ECMWF experiment, showing where Sandy was predicted to be located five-days out with the normal satellite data inputs into the model (left) and without any polar-orbiting satellite data (right). Both position and intensity forecasts were affected--Sandy stays out to sea without the polar-orbiting satellite data, and the closer isobar lines encircling the storm also imply a more organized and stronger system. Image credit: NOAA.

Polar satellite data: a key to ECMWF model success
The ECWMF has a very sophisticated technique called "4-D Var" for gathering all the current weather data over the Earth and putting the data on a 3-dimensional grid that is then used as the initial "reality" of the current weather for the model to use for its forecast. The old expression, "garbage in, garbage out" is a truism for weather forecast models. If you don't properly characterize the initial state of the atmosphere, the errors you start off with will grow and give a lower-quality forecast. Data from geostationary satellites, which sit continuously at one spot above the globe, are easy to assimilate, and all the models use this data. However, the ECMWF model's superior technique used to assimilate the initial data allows inclusion of data from a large number of polar-orbiting satellites, which the other models cannot do as well. Polar-orbiting satellites orbit Earth at an altitude of 540 miles twice per day, circling from pole to pole. Their data is difficult to use, since the it is only available twice per day at each spot on the Earth, and the time of availability is different for each location. According to an email I received from Jean-Noël Thépaut, the chief of the Data Division of the Research Department at the European Center for Medium-Range Weather Forecasts, the ECMWF model uses data from at least fourteen polar orbiting satellites: N-15, N-19, N-19, N-17 (ozone SBUV instrument only), Metop-A, AQUA, NPP (ATMS instrument only), AURA (ozone OMI data only), F-17, TRMM (TMI data), COSMIC, GRACE-A, TERRASAR, and the GPSRO data on top of METOP-GRAS. The data of most importance is the data collected in the infrared and microwave wavelengths, as well as atmospheric density data obtained via GPS radio occultation (as a polar orbiting satellite goes over the horizon, the GPS signals from the satellite get bent by Earth's atmosphere, with the amount of bending proportional to the density of the atmosphere. This GPS Radio Occultation data is gathered from eight polar orbiting satellites, and fed into both the ECMWF and GFS models.) You can find a nice summary of the impacts of polar orbiting satellite data on weather prediction models at this link.)


Figure 2. Forecast track error for four of our top models used to predict Hurricane Sandy, for their runs that began at 00Z October 25, 2012. By this time, the GFDL model had joined the ECMWF in predicting that Sandy would make landfall in Southern New Jersey in five days. The GFS and HWRF models made good 1 - 3 day forecasts, but failed to anticipate Sandy's north-northwestward turn towards the U.S. coast. Image credit: Morris Bender, NOAA/GFDL.

As originally reported by the Washington Post's Capital Weather Gang, then confirmed in a NOAA press release, a study done by ECMWF research scientist Tony McNally found that if the ECMWF model did not have all of the data from the fourteen polar orbiting satellites, the five-day forecast of the model for Hurricane Sandy would have shown Sandy missing the Northeast U.S. This brings up a concern, since the U.S. polar orbiting satellite program is behind schedule. As explained by Andrew Freedman of Climate Central, the program is plagued by mismanagement, billions in cost overruns, and technical development challenges. The next polar orbiting satellite is not scheduled to be launched until 2017, and one or more of the existing polar orbiting satellites are expected to fail before then. This will result in a degradation of our ability to observe and predict the weather, and may result in poorer forecasts for storms like Hurricane Sandy. Given that the ECMWF model used data from fourteen polar orbiting satellites, the failure of just one satellite may not have made a significant difference in its forecast for Sandy. But if we lose several of these key satellites by 2017, our hurricane forecasts in 2017 may be worse than they were in 2012. To figure out how to cope with the loss of satellite-derived data, NOAA is conducting a Gap Risk Study that seeks ideas from researchers and the public on how NOAA can preserve the quality of its weather model forecasts in the event of the failure of one or more polar orbiting satellites in the coming years.


Figure 3. A tanker rests on the southern shore after being swept onto land by a storm surge due to Superstorm Sandy, Friday, Nov. 2, 2012, in the Staten Island borough of New York. (AP Photo/ John Minchillo)

Links
Eric Berger of the Houston Chronicle has an interview with Jean-Noël Thépaut, chief of the Data Division of the Research Department at the European Center for Medium-Range Weather Forecasts, on why the European model did so well with Hurricane Sandy.

Jeff Masters