WunderBlog Archive » Category 6™

Category 6 has moved! See the latest from Dr. Jeff Masters and Bob Henson here.

Early 2011 Atlantic Hurricane Season Forecasts

By: Dr. Jeff Masters, 6:22 AM GMT on April 07, 2011

Hi everybody, this is Dr. Rob Carver filling in for Dr. Masters. 

A continuation of the pattern of much above-average Atlantic hurricane activity we've seen since 1995 is on tap for 2011, according to the latest seasonal forecast issued April 6 by Dr. Phil Klotzbach and Dr. Bill Gray of Colorado State University (CSU). They are calling for 16 named storms, 9 hurricanes, and 5 intense hurricanes. An average season has 10 named storms, 6 hurricanes, and 2 intense hurricanes. The new forecast is nearly identical to their forecast made in December, which called for 17 named storms, 9 hurricanes, and 5 intense hurricanes. Only six seasons since 1851 have had as many as 17 named storms; 19 seasons have had 9 or more hurricanes. The 2011 forecast calls for a much above-average chance of a major hurricane hitting the U.S., both along the East Coast (48% chance, 31% chance is average) and the Gulf Coast (47% chance, 30% chance is average). The Caribbean is forecast to have a 61% chance of seeing at least one major hurricane (42% is average.) Five years with similar pre-season November atmospheric and oceanic conditions were selected as "analogue" years that the 2011 hurricane season may resemble: 2008, 1999, 1996, 1955, and 2006.  The first four years listed all had neutral to La Niña SST's during hurricane season, while 2006 had El Niño SST's.  The average activity for these years was 12.6 named storms, 7.8 hurricanes, and 4.8 major hurricanes.

This year, the forecasters have introduced a new statistical model for their  April forecasts.  There are four components in this model:

1. Average sea-level pressure in March around the Azores in the subtropical Atlantic.

2. The average of January through March sea-surface temperatures (SST's) in the tropical Atlantic off the coast of Africa.

3. Average sea-level pressure in February and March for the southern tropical Pacific ocean west of South America.

4. Forecasts of September's SST in the tropical Pacific using a dynamical model from the European Centre for Medium-Range Weather Forecasts (ECMWF) 

The first two components are loosely linked together.  Statistical studies have shown that a weaker subtropical high near the Azores, combined with warmer SST's off the coast of Africa in March are associated with weak winds near the surface and aloft from August to October.  This decrease in wind speeds reduces wind shear which can disrupt forming storms.  These March conditions also are associated with warmer SST's in August to October, which is also favorable for more tropical storms.   For this forecast, the first component is strongly favorable for increased hurricane activity, while the second component is weakly negative.

The last two components represent the changes in sea-surface temperature and sea-level pressure that are the El Niño-Southern Oscillation (ENSO).  Briefly speaking,  El Niño conditions (warm sea-surface temperatures) are not favorable for Atlantic hurricanes.  For more info on ENSO and hurricanes, Jeff has this article.

Using the ECMWF model as guidance (see Figure 1), the CSU group believes that SST's in the tropical Pacific will be neutral (less than 0.5°C from normal).  This would have a small negative effect on hurricane activity.  However, the tropical Pacific sea-level pressure shows that the atmosphere looks like a La Niña event is still going on.  This is strongly favorable for Atlantic hurricane activity in the CSU group's model.

Figure 1. Forecasts of El Niño conditions by 20 computer models, made in March 2011. The ECMWF forecast used by the CSU group is represented by the dark orange square.  The forecasts for August-September-October (ASO) show that 5 models predict El Niño conditions, 7 predict neutral conditions, and 5 predict a weak to moderate La Niña. El Niño conditions are defined as occurring when sea surface temperatures in the Equatorial Pacific off the coast of South America ( the "Niño 3.4 region) rise to 0.5°C above average (top red line). La Niña conditions occur when SSTs in this region fall to 0.5°C below average. Image credit: Columbia University.

How accurate are the April forecasts? While the formulas used by CSU do well in making hindcasts--correctly modeling the behavior of past hurricane seasons--their April hurricane season forecasts have had no skill in predicting the future. This year's April forecast is using a new system and has not yet produced a verified forecast.  The scheme used in the past three years successfully predicted active hurricane seasons for 2008 and 2010, but failed to properly predict the relatively quiet 2009 hurricane season. A different formula was used prior to 2008, and the April forecasts using that formula showed no skill over a simple forecast using climatology. CSU maintains an Excel spreadsheet of their forecast errors ( expressed as a mathematical correlation coefficient, where positive means a skilled forecast, and negative means they did worse than climatology) for their their April forecasts. For now, these April forecasts should simply be viewed as an interesting research effort that has the potential to make skillful forecasts. The next CSU forecast, due by June 1, is the one worth paying attention to. Their early June forecasts have shown considerable skill over the years.


Figure 2.
Accuracy of long-range forecasts of Atlantic hurricane season activity performed by Phil Klotzbach and Bill Gray of Colorado State University (colored squares) and Tropical Storm Risk, Inc. (colored lines). The CSU team's April forecast skill is not plotted, but is less than zero. The skill is measured by the Mean Square Skill Score (MSSS), which looks at the error and squares it, then compares the percent improvement the forecast has over a climatological forecast of 10 named storms, 6 hurricanes, and 2 intense hurricanes. TS=Tropical Storms, H= Hurricanes, IH=Intense Hurricanes, ACE=Accumulated Cyclone Energy, NTC=Net Tropical Cyclone Activity. Image credit: TSR.

2011 Atlantic hurricane season forecast from Tropical Storm Risk, Inc.

The  British  private  forecasting  firm  Tropical Storm Risk, Inc.  (TSR),   issued  their  2011  Atlantic hurricane season forecast on April 5. They are also calling for  a  very  active  year: 14. 2 named storms, 7.5 hurricanes, and 3.6 intense hurricanes. We would round that to 14 named storms, 8 hurricanes, and 4 intense hurricanes.   This  compares to their forecast issued in December of 15.6 named storms, 8.4 hurricanes,   and intense hurricanes. TSR predicts a 55%  chance  of  an  above-average  hurricane season, 28% chance of a near-normal season, and only a 17%  chance  of  a  below normal season. TSR bases their April forecast on predictions  that  sea  surface temperatures this fall in the tropical  Atlantic  will  be  above  about  0.08°C above average, and trade  wind  speeds  will  be  about 0.2  m/s  slower  than average.  The decrease in the trade wind speeds is favorable for enhanced hurricane activity, while the forecast SST's are expected to be neutral for hurricane activity.

TSR puts their skill level right next to the forecast numbers: 13% skill above chance at forecasting the number of named storms, 11% skill for hurricanes, and 10% skill for intense hurricanes. That's not much skill, and really, we have to wait until the June 1 forecasts by CSU, NOAA, and TSR to get a forecast with reasonable skill.

Rob's critiques of the April forecasts
I have to note that Jeff and I wrote this article together.  He wrote the general framework before the forecasts were issued, while I wrote the details based on the actual forecasts.  So the preceding text is a joint production.  However, I have a few observations to make that are my responsibility alone.

First, I am disappointed that the CSU group has changed forecast models only after three seasonal forecasts.  This makes it very difficult to assess the skill of the current forecast using past performance.  This is very important for forecast users, and they do it everyday.  For example, I tend to discount a forecast of rain if it comes from a source that over-forecasts rain (The boy who cried wolf problem).

In the documentation that came with the April forecast, the CSU group argue that the hindcasts show the new forecast model has skill.  However, I think hindcasts are a poor substitute for real forecasts in understanding the skill of a statistical forecast model, like that of the CSU's group.  As Jeff noted, the previous forecast model did well with the hindcasts and yet had mixed results with the actual forecasts.  This does not give me confidence that the new forecast model will be superior to the previous model.

From a philosophical viewpoint, I am inherently cautious about statistical forecast models like the one used by the CSU group.  Essentially, they look at what happened in the past and use that to predict the future.  However, for making forecasts, we assume that the relationships in space and time between the predictors (such as the average March sea-level pressure around the Azores) and the predictands (Atlantic hurricane activity) does not change as we move forward in time.  In a world with climate change, that's a tricky assumption to make.

In any event, it is customary in the meteorological community to continue running older forecast guidance models after the introduction of newer models.  This allows forecasters and forecast users to leverage their knowledge of the forecast skill of the older model and gain insight into the forecast skill of the new model.  The CSU group really should have included the forecast from the previous statistical forecast system in this forecast.     

I am uneasy with some of the methodology choices made in implementing the forecast model.  Data for the first three predictors was obtained from the Climate Forecast System Reanalysis (CFSR), NOAA's newest and most advanced reanalysis product.  However, CFSR data for 2010 and 2011 has not been released yet, so the CSU group used NCEP/NCAR Reanalysis (NNR), NOAA's first-generation reanalysis, to fill in the gaps.  Due to differences in design, resolution, etc., CFSR and NNR can have different depictions of the state of the atmosphere.  So using NNR's March 2011 average SLP instead of CFSR's could alter the forecast in unexpected ways.  It would be interesting to see how CFSR's 2010-2011 data changes the results. 

In any event, we will have to wait and see what the Atlantic hurricane season of 2011 brings.

Hurricane

The views of the author are his/her own and do not necessarily represent the position of The Weather Company or its parent, IBM.