Challenging Bill Proenza's QuikSCAT numbers
A political storm engulfed the National Hurricane Center this week, with a majority of the senior hurricane forecasters calling for Bill Proenza's removal as director. The most visible issue revolved around the extraordinary focus on the aging QuikSCAT satellite. The public argument put forth by Mr. Proenza was that QuikSCAT data was so vital to hurricane track forecasting that without it, track forecast errors would increase significantly, leading to larger warning areas and increased costs for evacuation and emergency planning.
Focus on QuikSCAT--out of proportion?
On March 16th Proenza went public with the QuikSCAT concerns and associated statistics for the first time, stating that "two- and three-day forecasts of a storm's path would be affected. The two-day forecast could be 10 percent worse while the three-day one could be affected up to 16 percent," with the conclusion," that would mean longer stretches of coastline would have to be placed under warnings, and more people than necessary would have to evacuate." As a result of these comments, a perception arose in the public and among lawmakers that without QuikSCAT, NHC would not be able to provide accurate hurricane forecasts. Legislation was hastily introduced into both the House and Senate to provide an immediate replacement for the $375 million satellite.
Proenza's statements raised several questions: 1) Why the focus on track forecast errors in landfalling situations, when QuikSCAT was widely known to be used in intensity forecasting and for tropical cyclones too far at sea to be accessed by the Hurricane Hunters? 2) Could such specific and significant gains in track forecast error truly be attributed to QuikSCAT? Where did these numbers come from, and why was no uncertainty being attached to them?
Since QuikSCAT data became available, starting in 1999, average track errors for 48-hour and 72-hour forecasts have been reduced by 43 miles and 62 miles respectively. Fully one quarter of this improvement was being attributed by Proenza to QuikSCAT. This was an extraordinary performance increase to attribute to one satellite, and seemed doubtful.
We find out where the QuikSCAT numbers came from
In mid-June, Margie Kieper and I asked Proenza to comment on how he got his QuikSCAT numbers. He cited an unpublished study, "A Two Season Impact Study of Four Satellite Data Types and Rawinsonde Data in the NCEP Global Data Assimilation System", by Tom H. Zapotocny, James A. Jung, John F. LeMarshall and Russ E. Treadon. I contacted one of the authors, who informed me that the study was submitted for publication on January 26, 2007, and accepted for publication in the journal Weather and Forecasting on May 23, 2007. It will probably appear in the October-November time frame, according to the publisher. This raises an immediate problem, since only a privileged few are able to read unpublished research. This limits the possibilities for an informed debate on the issue, and basing important policy decisions on unpublished research is thus normally to be avoided. However, making accurate hurricane forecasts is important enough that such considerations can be excused. Proenza didn't give me any details on the study, other than the fact that QuikSCAT data improved 72-hour and 48-hour hurricane track forecasts by 16% and 10% respectively, for a select group of storms from the 2003 hurricane season. One of the authors graciously sent me a copy of the study, though, and after reading it, I had these observations:
1). The study looked at a very limited number of cases over a six-week period during 2003--only 19 cases were available for 72 hour forecasts. The 19 cases were not 19 storms, just 19 separate forecasts from the 4 hurricanes and 2 tropical storms that occurred during the 6-week study period. This sample is too small to draw definitive conclusions about the impact of the QuikSCAT on tropical cyclone forecasts. The two longest-lived storms during the test period were Fabian and Isabel, storms that spent the majority of their lifetimes far away from land. Since the quality of the observing network increases close to land, particulary when reconnaissance data from the Hurricane Hunters is available, it is reasonable to conclude that the impact of the QuikSCAT data for storms within 72 hours of landfall would be less than for the sample as a whole. The study was not primarily designed to study tropical cyclone track accuracy, so there was no separation out of the cases we really care about--storms 72 hours or less from landfall.
2). The study was done with only one model, the GFS. NHC official forecasts make use of several models, including the GFDL, UKMET, NOGAPS and ECMWF. Consequently, a change in the accuracy of a single model will have only a partial effect on NHC official forecast accuracy. As far as I know, there have not been studies done of the impact of QuikSCAT on tropical cyclone forecasts in the GFDL, UKMET or the ECMWF models. Past studies on the impact of dropsonde data from the Hurricane Hunters, however, show that the GFDL is less sensitive to these data than the GFS is.
3). When I attended the AMS hurricane conference in May 2006 in Monterey, I came across a poster presentation by Dr. Jim Goerss that evaluated the impact of QuikSCAT on the NOGAPS model. His study was far more comprehensive, and included 12 hurricanes, 5 typhoons, and 7 tropical storms from a 6-week period in 2004. The number of cases was 212 at 72 hours, eleven times as many as the study Proenza cites. Dr. Goerss found that QuikSCAT probably improved 24-hour track forecasts by 2.5% (90% confidence of this), but at all other forecast times (48, 72, 96, and 120 hours), QuikSCAT had no statistically significant effect (i.e., zero effect).
It is hard to compare the results from these two studies, since they used two different data assimilation systems. We do not know if they used all the data, or how they treated the vertical impact of the data. The uncertainties are high, and Proenza's simple statement that QuikSCAT data improves hurricane tracks forecasts by 10% and 16% is unreasonable, without at least making mention that these numbers are highly uncertain.
I believe that NHC official forecasts for landfalling storms in the Atlantic would not be significantly affected by the loss of the QuikSCAT satellite. I can't think of a hurricane scientist out there who would defend using a study with only 19 cases that didn't focus on landfalling storms, to make the case Proenza is making--particularly in light of the data from the unpublished Goerss study showing no effect of QuikSCAT data on NOGAPS model tropical cyclone track errors. Proenza should have at least attached some measure of uncertainty to his numbers, which he did not.
One could argue that the study cited by Proenza has undergone peer review, and is thus the only scientific study one can use to make arguments on QuikSCAT's effectiveness. The Goerss study has not been published in a journal, and has not undergone peer review. However, Proenza was making his QuikSCAT accuracy arguments in March, two months before the Zapotocny study he cited had been accepted for publication.
The numbers pushed by Proenza have led to some potentially serious misconceptions about QuikSCAT. The Congressional Record has this to say about QuikSCAT:
"A single plane gathering data is like a tiny fishing line collecting data only along the single strand of the line. The satellite, on the other hand, provides rich, detailed data horizontally from one side of the storm to the other side, and vertically, from the ocean surface to the top of the storms swirling winds. The QuikSCAT is like a detailed MRI."
Well, QuikSCAT is not like an MRI, it just measures the ocean surface winds. In a letter written by Representatives Melancon and Klein in support of H.R. 2531, there are comments that data from the reconnaissance aircraft are inferior to the data from the QuikSCAT:
"Short-term options for replacing QuikSCAT include hurricane hunter aircraft, buoys, and foreign satellites--all of which will collectively produce inferior data."
There is not a hurricane forecaster anywhere that would trade hurricane hunter data for QuikSCAT. Lawmakers may start cutting aircraft reconnaissance with misconceptions like this. That would be a disaster.
I would hate to lose the QuikSCAT satellite, and have been calling for a replacement since before Mr. Proenza came on the job. QuickSCAT data is invaluable in identifying weak systems and in defining storm structure, particularly of outer wind radii of 34 knots and 50 knots. This is particularly true outside of the Atlantic, where there are no Hurricane Hunter flights, and in the Atlantic beyond where the Hurricane Hunters can reach. Track forecasts for tropical cyclones in the Pacific and Indian Oceans may benefit from QuikSCAT data, since Hurricane Hunter information is not available. QuikSCAT also helps identify when a tropical depression or tropical storm is intensifying.
Besides hurricanes, the QuickSCAT data is invaluable to the Ocean Prediction Center, which now issues hurricane force wind warnings for extratropical storms in the Atlantic and Pacific. Search and rescue missions, and the U.S. Navy also greatly benefit from QuikSCAT. QuikSCAT should be replaced, but not due to a rush knee-jerk reaction that will get us a replacement with old technology. NHC needs a "next-generation" scatterometer, one that has greatly improved capabilities to help tackle the structure and intensity problem. We should take our time, and deal with a gap in coverage, if it gets us an instrument that has higher resolution, higher saturation speed, and is not adversely affected by rain. Such a gap would not put the public at risk.
It greatly troubles me that the most visible and admired member of my profession has failed to use good science in his arguments for funding a replacement of the QuikSCAT satellite. The Director of the National Hurricane Center needs to be an able politician and good communicator, but being truthful with the science is a fundamental requirement of the job as well. Mr. Proenza has misrepresented the science on the QuikSCAT issue, and no longer has my support as director of the National Hurricane Center.
Other critical concerns--lost in the hubbub?
We strongly support many of the valid concerns Proenza has raised. Of particular concern are the slashing of critical research funding for the Joint Hurricane Testbed (JHT) from $1.7 million to $1 million, and the lack of adequate yearly increases to the National Hurricane Center budget. Both of these important concerns still remain to be addressed; they were quickly overshadowed by a frantic campaign by lawmakers to fund a new QuikSCAT satellite. The JHT provides the means for promising research to be tested in the NHC operational environment, usually resulting in a successful transition to an operational product at NHC. This program has been extremely successful, and its budget should have been increased, not slashed. As hurricane activity has increased dramatically over the last twelve years, NHC's budget should have increased accordingly, but it did not.
Proenza also raised legitimate concerns about NOAA's effort to promote their "Corporate Identity" by renaming the National Hurricane Center and National Weather Service. The new organizations would be called the "NOAA Hurricane Center" and the "NOAA Weather Service". He also justly complained about NOAA's plan to spend between $1.5 million and $4 million on a "bogus" 200-year NOAA anniversary celebration.
While wanting to take a neutral stand as to whether to call for Proenza's dismissal, Senior NHC Hurricane Specialist Lixion Avila clearly shares the concerns that have been put forth by the other senior forecasters Richard Pasch, James Franklin, and Richard Knabb, and former director Max Mayfield. Avila noted, "If I [was] the director of the hurricane center, I would not spend my time fighting for QuikSCAT--I would be fighting to make sure that the reconnaissance planes are always there." That leaves a vacationing Jack Beven as the only senior hurricane forecaster to not comment publicly on the issue. Max Mayfield has refrained from making public comments on the deteriorating situation these past months, but all of his comments in the Miami Herald article lend support for the hurricane forecast staff. Given his previous experience in the position of NHC Director and his successful tenure, his feedback counts tremendously.
With the busiest part of hurricane season just a few weeks away, expect a decision on Bill Proenza's tenure to be made soon.
Jeff Masters and Margie Kieper
Having lost the support of most of his senior forecasters, and having misrepresented the science on the importance of the QuikSCAT satellite on hurricane forecasts, it would be best for Mr. Proenza to step down as director of the National Hurricane Center.