NHC Track Forecasts Were Best Ever in 2017; No Improvement in Intensity Forecasts

The National Hurricane Center (NHC) made their all-time best track forecasts during the 2017 Atlantic hurricane season, setting new records for track forecast accuracy at all forecast times. According to their annual forecast verification report released on May 9, the 2017 NHC track forecasts were up to 15% better than the previous record-best forecasts. Over the past fifteen years, 1- to 5-day NHC track forecast errors have declined by about a factor of two—an extraordinary accomplishment that has undoubtedly led to a huge savings in lives, damage, and emotional angst for the people living in Hurricane Alley.

The average track errors for official NHC forecasts in 2017 were 37 miles for a 1-day forecast, 65 miles for a 2-day forecast, 101 miles for 3 days, 131 miles for 4 days, and 179 miles for 5 days. The official track forecasts tended to have an east-southeasterly bias of 43 - 58 miles for 3 - 5 day forecasts (i.e., the official forecast tended to fall to the east or southeast of the verifying position). NHC made much better track forecasts for hurricanes and strong tropical storms (by about 30 – 40%) than for weak tropical storms and tropical depressions. The NHC track forecasts for Harvey, Irma, and Maria were particularly excellent.

NHC intensity forecasts in 2017: no improvement over the 5-year average

Official NHC intensity forecast errors did not improve in 2017 compared to forecasts from the past five years, with errors for 12-hour, 1-day, and 2-day forecasts 10% - 15% worse than the 5-year average error. Intensity errors for 3-day, 4-day, and 5-day forecasts were near the 5-year mean. Mean intensity forecast errors in 2017 were 10 mph at 24 hours, and 13 – 16 mph for 2-day, 3-day, 4-day, and 5-day forecasts. The official forecasts were biased too low at most forecast times, underpredicting a storm’s strength by an average of 2 – 3 mph for 2-day through 5-day forecasts. Part of the reason for the lack of better intensity forecasts was the fact that rapidly intensifying hurricanes are the hardest to make intensity forecasts for, and there were more of these storms than usual in 2017. The report noted that in particular, NHC did a poor job forecasting the rapid intensification of Hurricane Irma near the Leeward Islands and the subsequent weakening due to land interaction of the hurricane with Cuba.

Figure 1. Verification of official NHC hurricane intensity forecasts for the Atlantic, 1990 - 2017. Intensity forecasts have shown little improvement over the past five years, but some modest improvement over the past fifteen years. Image credit: 2017 National Hurricane Center Forecast Verification Report.

Best track model in 2017: the European

As usual, in 2017 the official NHC track forecasts for Atlantic storms were better than those of any individual computer model at all forecast time periods, although the European Center (ECMWF) model was very close in accuracy to the official forecast. The UKMET model made a big jump in accuracy compared to its performance in 2014 - 2016, and was the second-best model in 2017. NOAA’s GFS model, which had been very close to the European model in accuracy from 2014 – 2016, fell into the third tier, with an accuracy similar to that of the HWRF and CMC (Canadian) models. The simple BAMM model did quite well, outperforming two sophisticated dynamical models, the HMON and COAMPS-TC, at many forecast times. Here is a list of some of the main hurricane forecast models used by NHC:

Euro: The European Center for Medium-range Weather Forecasting (ECMWF) global forecast model
GFS: The National Oceanic and Atmospheric Administration (NOAA) Global Forecast System model
UKMET: The United Kingdom Met Office's global forecast model
HMON: Hurricanes in a Multi-scale Ocean-coupled Non-hydrostatic regional model, initialized using GFS data
HWRF: Hurricane Weather and Research Forecasting regional model, initialized using GFS data
CTC: COAMPS-TC regional model, initialized using GFS data
CMC: The Canadian GEM global model
BAMM: The very old Beta and Advection Model (Medium layer), which is still useful at longer ranges

If one averages together the track forecasts from two or more of the first six of these models, the NHC official forecast will rarely depart much from it. These six models are used as input to what are called “consensus” models, which you will often see referenced in the NHC discussions for a storm.

Figure 2. Skill of computer model forecasts of Atlantic named storms in 2017, compared to a "no skill" model called "CLIPER5" that uses just climatology and persistence to make a hurricane track forecast (persistence means that a storm will tend to keep going in the direction it is currently going). OFCL=Official NHC forecast; GFS=Global Forecast System model; HMON=Hurricanes in a Multi-scale Ocean-coupled Non-hydostatic regional model; HWRF=Hurricane Weather Research Forecasting model; Euro=European Center for Medium-range Weather Forecasting (ECMWF) model; UKMET model=United Kingdom Met Office model); CMC=Canadian Meteorological Center (GEM) model; BAMM=Beta Advection Model (Medium depth).

Best intensity model in 2017: the HWRF

In 2017, the official NHC intensity forecast outperformed the five main intensity models at all forecast times, save for 4-day forecasts, which the COAMPS-TC model did a better job at. The five main intensity models are the regional/dynamical HWRF, HMON, and COAMPS-TC models (which subdivide the atmosphere into a 3-D grid around the storm and solve the atmospheric equations of fluid flow at each point on the grid), and the statistics-based LGEM and DSHP models (DSHP is the SHIPS model with inland decay of a storm factored in). Two of our top-performing global dynamical models for hurricane track, the GFS and European (ECMWF) models, are typically not considered by NHC forecasters when making intensity forecasts. These models made poor intensity forecasts in 2017 (with the GFS model having some skill, and the European model, no skill). In 2017, the HWRF model was the best-performing intensity model for 1-day to 3-day forecasts, with the other four main intensity models close behind. The LGEM statistical model was the best one at longer-term intensity forecasts of 5 days. 

Model improvements for 2018

At last month’s American Meteorological Society’s 33^rd Conference on Hurricanes and Tropical Meteorology, I heard about a number of encouraging model improvements coming for the 2018 hurricane season. All of the models will benefit from high-resolution data from the new GOES-16 satellite, which will provide the models better and more frequent Atmospheric Motion Vectors (AMVs)—estimates of wind speed and direction based on cloud movement. The models will also benefit from a new polar orbiting satellite that was launched in November, NOAA-20.

The HWRF model, which was the best-performing intensity model in 2017 (and roughly the 3^rd-best track model), will benefit from getting more data from the Hurricane Hunters in real-time. The NOAA jet now has a Doppler radar, and data from this radar, as well as the Doppler radar on NOAA’s P-3 hurricane hunter aircraft, will be assimilated in real-time into the HWRF model. Not all of the GPS dropsonde data from the Hurricane Hunters was being used by the HWRF model last year, but will be utilized this year; the model will also incorporate all of the flight-level high-density observations (HDOBs) for the first time. The model will also use all of the surface wind data from the Step Frequency Microwave Radiometer (SFMR) instruments carried by both the NOAA and the Air Force Hurricane Hunters. The HWRF model used three nested grids zoomed in on hurricanes with resolutions of 18 km, 6km, and 2 km last year; the resolutions of the three grids improves to 13.5 km, 4.5 km, and 1.5 km this year.

A new version of the GFS model called the FV3, using the finite volume method to make its computations, was tested last year, and did an excellent job predicting the hurricanes of 2017. The FV3 will eventually become NOAA's new unified model, replacing the current version of the GFS. You can view plots of the experimental GFS FV3 model at tropicaltidbits.com.

Sources of model data

Tropical Tidbits, which has become the best source of free model data on the web.
Wunderground’s wundermap, with the “Model Data” layer checked: GFS and hi-res Euro
NOAA excellent HFIP page, which compares 7-day forecasts from the GFS, Euro, UKMET, CMC, NAVGEM, and new GFS FV3 models.
Longer ten-day ECMWF forecasts, available from the ECMWF web site.
FSU’s model page (CMC, ECMWF, GFS, GFS FV3, HWRF, HMON, and NAVGEM models)
NOAA’s HWRF model page
NOAA’s HWRF and HMON model data page
The Navy’s COAMPS-TC model data page
Experimental HFIP models