For roughly the last decade, a special version of multiple linear regression has been used by Dr. 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 four times: in early December of the preceding calendar year, in early April, 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). Just recently, the forecasts have begun including certain Atlantic Ocean regional predictors for forecasts made at certain lead times (Landsea et al. 1997). Expected predictive skill is improved with the inclusion of the new predictors as 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).

The predictors used, and the skill expected, at three of the four forecast times were originally presented in the September 1993 issue of this Bulletin. However, the prediction system has since been changed/improved. Extended range predictions of the QBO, the ENSO (Gray et al. 1994), and western tropical African rainfall for the upcoming storm season have become increasingly objective. The current forecast for 1997 cyclone activity is based on both the original scheme that used two predictor clusters as in Gray et al. (1992) (see below), and a newer technique in which a predictand-specific set of the 3 to 7 best predictors out of a pool of 10 candidate predictors is used for each of the 9 predictands. The final issued forecast is a combination of both schemes, plus a final qualitative adjustment based on human intuition and judgement.

The following list identifies the two predictor clusters for the original, basic forecast scheme, their expected influence on Atlantic tropical storm activity, and their status regarding the 1997 storm season. More detail on the influences of these predictor groups is found in Gray et al. (1993).

(1) The Quasi-Biennial Oscillation (QBO) at 50 and 30 mb in the northern tropics (near 10^oN) expected at the onset of the hurricane season: Westerly QBO winds enhance storm activity, while easterly winds suppress it (Gray et al. 1992). The amount of absolute shear between the two levels is also important: large shear inhibits storm activity. Larger shears tend to occur with easterly QBO. In fall 1997 the QBO is expected to have switched from easterly to near-neutral (having little effect on storm activity) and the magnitude of the shear is expected to be low. The net effect on 1997 storms is slightly positive.

(2) African rainfall: Intense hurricane activity is enhanced when the Western Sahel and Gulf of Guinea regions in West Africa have above average precipitation the previous late summer and fall (implying favored chances for the same anomaly sign for the upcoming late summer and fall, which is the actual "predictor"), and is suppressed when that precipitation is below average (Landsea et al. 1993). Conditions in summer/fall 1996 were about average in the Western Sahel and mildly below average in the Gulf of Guinea, together indicating neutral to slightly below normal storm activity in fall 1997.

In the newly developed prediction technique a set of 10 potential predictors is introduced for each predictand, and the set of up to 7 of these that maximize the dependent sample explained variance is used. The following table shows which predictors were selected for each predictand.

---------------------- P R E D I C T A N D -------------------------------

NS NSD H HD IH IHD HDP NTC MPD

#predictors==>(3) (6) (5) (5) (4) (3) (5) (4) (7)

PREDICTOR

Zonal wind, 50mb . X X X X . X . X

Zonal wind, 30mb X X . X . . X . .

|30mb-50mb shear| . . X . X X . X X

Aug-Nov Guinea X X X X X X X X X

rain

Aug-Sep West . X X X . X X X X

Sahel rain

Atlantic ridge X X X X X . X X X

May-July Darwin . . . . . . . . .

SLPA

Niño 4 trend: . . . . . . . . X

(Aug-Oct)-(May-Jul)

Aug-Oct SOI . . . . . . . . X

SOI Trend:

(Aug-Oct)-(May-Jul) . X . . . . . . .

Warm eastern equatorial Pacific sea surface temperature (SST), and/or a low Southern Oscillation Index (SOI), both associated with El Niño, reduces storm activity, while oppositely signed anomalies enhances activity. Although the current state of the Pacific SST is cool, Dr. Gray & Company's ENSO prediction calls for a continuation of slightly cool SST during the fall 1997 tropical cyclone season, which would slightly encourage storm development. Note that the last four of the 10 candidate predictors for the new prediction scheme are closely related to the ENSO situation. The strength of the Atlantic SLP ridge near the Azores, which is inversely related to storm activity, is currently well above its average. This will influence the 1997 prediction downward.

The LAD multiple regression predictions for 1997 are shown in Table 1 for each tropical storm parameter. Percentages of the 46-year (1950-95) average are shown to the right. The first column shows expected real-time forecast skill for late November forecasts in terms of an agreement coefficient, using the new 3 to 7-predictor forecast system. Shown are both the skills on the dependent (training sample) data set, followed by roughly estimated skills for independent data (since these statistics have not yet been determined for these particular predictors). The latter skills are considerably lower than those realized on the training sample, and more correctly reflect the skill expected in real-time forecasts. The next three columns in Table 1 show the forecasts from the original regression equation, then from the improved 3 to 7-predictor forecast, and finally after the qualitative adjustment. The original scheme predicts a generally near-normal storm season, while the newer system predicts a slightly inactive season. The final forecast, including the qualitative adjustment, is for a slightly enhanced 1997 storm season. Two reasons for this positive adjustment from the objective guidance are (1) the cold 100 mb temperature at Singapore, which is not yet in the prediction equation, and (2) the difference between SSTAs in the North vs. South Atlantic basins, also not yet in the equations, which has been increasing over the last couple of years presumably due to a stronger thermohaline circulation.

Table 1. Historical late November hindcast skills for Atlantic tropical storm predictions, and forecasts for the 1997 season. Percentages of the 1950-95 mean for the final (qualitative) forecasts are listed in the last column. Note: the skill estimates for independent data (following the slash mark), which are of greatest interest, are only rough estimates at this time.

Expected Skill, --------Forecasts----------

Depnt/Indepnt Basic Newer Final Final:

Forecast (% Variance Method Method (After % of

Parameter Explained, (QBO, (3-7 Pre- Human 1950-90

(No. of predictors) from 1950-94) ENSO) dictors) Adjustmt) Mean

Named storms (3) .52/.35 10.2 8.5 11 (118%)

Named storm days (6) .55/.38 50.2 36.9 55 (118%)

Hurricanes (5) .49/.34 6.3 5.3 7 (121%)

Hurricane days (5) .54/.39 23.6 17.1 25 (105%)

Intense hurricanes (4) .44/.28 1.9 1.9 3 (130%)

Intense hurricane days (3) .42/.27 3.8 3.8 5 (106%)

Hurricane destruction .49/.32 60.4 47.0 75 (105%)

potential (5)

Net tropical cyclone (4) .53/.38 91.1 73.1 110 (110%)

activity

Maximum potential (7) .66/.54 ---- 54.9 70 (106%)

destruction

Verification of 1996 Tropical Storm Forecasts

Table 2 shows the results for the 1996 tropical storm season, including Dr. Gray's November, June and August predictions, the observations, and the long term averages for each of the eight storm parameters. The forecasts were for a near-average storm season, with progressive increases in the predicted storm activity between the forecasts issued in November and August. These increases were in the right direction, as the 1996 storm activity turned out to be well above average (but not as active as 1995). Dr. Gray and his colleagues attribute the underestimate to the low probability with which two very active years occur in succession, and the easterly QBO winds and high Lower Caribbean Basin sea level pressure anomaly going into the 1996 season--both of which tend to reduce activity.

Table 2. Verification of the 1996 Atlantic tropical storm season. The forecasts issued in April are not listed.

---Forecast Issue Times--- 1950-90

Late Early Early Long-term

Forecast Nov. June August 1996 Mean/ 1996

Parameter 1995 1996 1996 Observed Percentage

Named storms 8 10 11 13 9.3/140

Named storm days 40 45 50 78 46.6/167

Hurricanes 5 6 7 9 5.8/155

Hurricane days 20 20 25 45 23.9/188

Intense hurricanes 2 2 3 6 2.3/261

Intense hurricane days 5 5 4 13 4.7/277

Hurricane destruction 50 60 70 135 71.2/190

potential

Maximum potential 55 60 65 92 66.0/139

destruction

Net tropical cyclone 85 95 105 198 100/198

activity

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., 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.

[Previous Article] [Next Article]