For roughly the last decade, a special version of multiple linear
regression has been used by Dr. William Gray and his team to make forecasts
separately for each of several parameters of tropical storm activity in
the Atlantic Basin (Gray et al. 1992). The least absolute deviation (LAD),
rather than the least squared deviation (where "deviation"
represents the error), is used as the criterion to develop the prediction
model for these forecasts. LAD regression was discussed briefly in the
December 1992 issue and is described more fully in Gray et al. (1993) and
references therein. Forecasts for each year's storm season are made at
three times: in late November of the preceding calendar year, in early
June, and finally in early August at the beginning of the storm season.
Forecast parameters include (1) named storms, (2) named storm days, (3)
hurricanes, (4) hurricane days, (5) intense hurricanes, (6) intense hurricane
days, (7) hurricane destruction potential, (8) net tropical cyclone activity,
and (9) maximum potential destruction. The predictors used, and the skill
expected, at each of the three forecast times have been revised a number
of times as new opportunities for skill increases have emerged. For example,
in 1994 prediction of the ENSO condition expected for the forthcoming fall
storm season was largely objectified and incorporated into the regression
equation (Gray et al. 1994b). In fact, the present forecast includes a
cluster of Atlantic Ocean regional predictors that has not been used previously
(Landsea et al. 1997). Expected predictive skill is improved with the inclusion
of the new predictors when compared with skills using only the previous
set of predictors. The following list identifies the six predictor clusters,
their expected influence on Atlantic tropical storm activity, and their
status regarding the 1996 storm season. More detail on the influences of
these predictor groups is found in Gray et al. (1993, 1994a).
(1) The Quasi-Biennial Oscillation (QBO) at 50 and 30 mb in the
northern tropics expected at the onset of the hurricane season: Westerly
QBO winds enhance storm activity, while easterly winds suppress it (Gray
et al. 1992). During fall 1996 the QBO is easterly, which is expected to
suppress storm activity this season.
(2) El Nino/Southern Oscillation (ENSO): Warm east-central equatorial
Pacific sea surface temperature, (SST), or the warm phase of ENSO, reduces
storm activity, while anomalously cool SST enhances it. Going into fall
we currently have a near-zero to weak negative SST anomaly in the Nino
3.4 region. This should very slightly enhance hurricane activity.
(3) African rainfall: Intense hurricane activity is enhanced when
the Western Sahel receives above normal rainfall in June-July immediately
preceding the current hurricane season, and when the Gulf of Guinea region
had been wet during the August-November period of the previous year. Conversely,
activity is suppressed when precipitation in those two regions/periods
is below average (Landsea et al. 1993). This June-July 1996 features weak
drought conditions (-0.6 standard deviations) in the western Sahel. Gulf
of Guinea rainfall for August through November of 1995 was near to very
slightly above average. The net effect of these two indicators is from
very slightly suppressing to near neutral.
(4) West Africa west-to-east surface pressure and temperature gradients:
Above average west-to-east surface pressure and (often associated) east-to-west
surface temperature gradients from February to May are associated with
enhanced hurricane activity later that year. For February through May of
this year, both the east-to-west temperature and west-to-east pressure
gradients were slightly negative (-0.3 standard deviations), indicating
a slight suppressing influence on storm activity.
(5) Caribbean basin sea level pressure anomaly (SLPA) and upper
tropospheric (12 km) zonal wind anomaly (ZWA): Negative anomalies of
either one of these in Jun-Jul weakly imply enhanced storm activity, while
positives weakly associate with reduced activity. For Jun-Jul both SLPA
was somewhat higher than normal (suppressing storm activity) but ZWA was
slightly below average, tending to enhance storm activity. The net indication
is for average 1996 storm conditions.
(6) Atlantic Ocean Regional Predictors (ONR, SSTA-MATL, SSTA-TATL):
When the previous year's October-November northeast Atlantic (30-40N, 20-30W;
near Azores) subtropical ridge (ONR) is weaker than normal, seasonal hurricane
activity is enhanced (see Landsea et. al 1997 for physical reasoning).
The SLP anomaly in Oct-Nov 1995 was quite low (-1.8 standard deviations),
indicating high storm activity for 1996. The Atlantic SST anomalies in
two regions (MATL: 30-50N, 10-30W; TATL: 6-22N, 18-82W) in the April to
June period is positively correlated with storm activity several months
later through the enhancement of deep oceanic convection. The anomalies
in these regions were slightly positive, encouraging slightly positive
storm activity for 1996. Together, the three predictors indicate a somewhat
active 1996 storm season.
The LAD multiple regression predictions, made first in November 1995 and
then in early June and early August 1995, are shown in Table 1 for each
tropical storm parameter. (The interim April forecasts are not shown here.)
The August forecasts are shown both as objective results of the LAD regression
equations ("obj") and as a final forecast ("fnl") which
includes a qualitative human/intuitive adjustment. The earlier forecasts
also include such an adjustment. The August forecasts are the sum of the
1996 storm activity observed before August 1 and the model-predicted amount
of activity expected from August 1 through the balance of the 1996 storm
season. The mean values over the 1950-95 period are shown. The right-hand
column provides a 50% confidence interval for the forecasts--i.e., the
25 and 75 percentile points of the estimated probability distribution of
the final August forecast. This distribution is estimated by empirical
examination of 1950-95 hindcast errors, and not from a theoretical calculation
based on a statistical skill score. Regarding the latter, the first column
shows expected skill for the forecasts in terms of percent variance explained.
The first coefficient is based on the development (dependent) sample, and
as such it is an overestimate of the skill expected for a forecast for
a future season or for any year outside of the development sample. The
second coefficient, based on exhaustive resampling simulations, is a best
estimate of the skill expected on truly independent data (Mielke et al.
1995). The decrease from the dependent to independent sample skill occurs
because of the fitting of a fairly large number of predictors to the predictand
using only a small to moderate sample (46) of hurricane seasons. The dependent
versus independent sample skill differences have encouraged the authors
to choose a set of predictors judiciously, resulting in retention of only
the most valuable predictors for each tropical storm parameter. The number
of elemental predictors now used ranges from 4 to 7 out of the pool of
12 candidates (composing the six clusters described above), depending on
the predictand. By economizing on the number of predictors, the penalty
for overfitting on the development sample is minimized.
Near to very slightly above average tropical storm activity is predicted
for 1996, as it has been since the late November 1995 forecast. The best
analog years to 1996 are 1954, 1956, 1970, 1974, 1979, and 1989.
Table 1. Expected forecast skill (agreement coefficient, or percentage
variance explained, for early August forecasts) and predictions for the
1996 Atlantic tropical storm season, as of late November 1995, early June
and early August 1996. The last two columns show the 1950-90 climatology
and the 50% confidence interval for the final August forecast. The number
in parentheses after the forecast parameter shows the number of predictors
used for that parameter.
Expected Skill |
Forecasts |
|||||||
Depnt/Indepnt |
Nov. 1995 |
June 1996 |
Aug. 1996 |
1950-90 |
50% |
|||
Forecast Parameter |
(% Variance |
Fcst |
Fcst |
Fcst |
Mean |
Conf. Int. |
||
(Number of Predictors) |
Explained from |
(obj,fnl) |
||||||
1950-94) |
||||||||
Named Storms (5) |
0.53/0.36 |
8 |
10 |
11,11 |
9.3 |
10.3-12.0 |
||
Named Storm Days (4) |
0.57/0.42 |
40 |
45 |
56,50 |
46.6 |
46.7-57.3 |
||
Hurricanes (5) |
0.54/0.37 |
5 |
6 |
7,7 |
5.8 |
6.4-8.2 |
||
Hurricane Days (6) |
0.63/0.48 |
20 |
20 |
24,25 |
23.9 |
22.0-27.3 |
||
Intense hurricanes (7) |
0.64/0.50 |
2 |
2 |
3,3 |
2.3 |
2.7-3.4 |
||
Intense hurricane days (4) |
0.61/0.48 |
5 |
5 |
5,4 |
4.7 |
2.7-4.6 |
||
Hurricane destruction potential (6) |
0.67/0.54 |
50 |
60 |
68,70 |
71.2 |
64.5-86.1 |
||
Net tropical cyclone activity (4) |
0.66/0.54 |
85 |
95 |
104,105 |
100.0 |
92.6-119.3 |
||
Maximum potential destruction (4) |
0.65/0.53 |
55 |
60 |
75,65 |
66.0 |
65.0-78.5 |
References
Gray, W.M., C.W. Landsea, P.W. Mielke, and K.J. Berry, 1992: Predicting
Atlantic seasonal hurricane activity 6-11 months in advance. Wea. Forecasting,
7, 440-455.
Gray, W.M., C.W. Landsea, P. Mielke and K. Berry, 1993: Predicting Atlantic
basin seasonal tropical cyclone activity by 1 August. Wea. Forecasting,
8, 73-86.
Gray, W.M., C.W. Landsea, P. Mielke and K. Berry, 1994: Predicting Atlantic
basin seasonal tropical cyclone activity by 1 June. Wea. Forecasting,
9, 103-115.
Gray, W.M., J.D. Sheaffer, P.W. Mielke, K.J. Berry and J.A. Knaff, 1994:
Predicting ENSO 9-14 months in advance. Proceedings of the 18th Annual
Climate Diagnostics Workshop, Boulder, Colorado, November 1-5, 1993, 390-393.
Landsea, C.W., W.M. Gray, P.W. Mielke and K.J. Berry, 1993: Predictability
of seasonal Sahelian rainfall by 1 December of the previous year and 1
June of the current year. Preprints, 20th Conference on Hurricane and
Tropical Meteorology, AMS, San Antonio, Texas, 473-476.
Landsea, C.W., W.M. Gray, K.J. Berry and P.W. Mielke, Jr., 1997: Revised
Atlantic basin seasonal tropical cyclone prediction methods for 1 June
and 1 August forecast dates. Wea. Forecasting, 12, in preparation.
Mielke, P.W., K.J. Berry, C.W. Landsea and W.M. Gray, 1995: Artificial
skill and validation in meteorological forecasting. Wea. Forecasting,
11, 153-169.