[Previous Article] [Next Article]
LAD Multiple Linear Regression Forecasts of Atlantic
Tropical Storm Activity for 1998
contributed by William Gray1, Christopher Landsea2, John Knaff1,
Paul Mielke3, and Kenneth Berry3
1Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado
2Hurricane Research Division, AOML, NOAA, Miami, Florida
3Department of Statistics, Colorado State University, Fort Collins, Colorado
For more than a decade now, 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 of this Bulletin 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, 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
following list identifies the predictor clusters, their expected influence on Atlantic tropical storm
activity, and their status regarding the 1998 storm season. The original predictor groups for
forecasts made in December consisted of only the QBO and African rainfall. The present scheme,
however, includes additional predictors. More detail on the influences of these predictor groups is
found in Gray et al. (1993, 1994), and more about the 1998 forecasts is provided in the web site
http://tropical.atmos.colostate.edu/forecasts/index.htmll.
a) Quasi-biennial Oscillation (QBO)
The easterly and westerly phases of stratospheric QBO zonal winds over the Tropics influence
Atlantic tropical cyclone activity (Gray, 1984). Typically, there is 50 to 75 percent more hurricane
activity (depending on the hurricane index considered) when stratospheric QBO winds between
30 mb and 50 mb are anomalously westerly and, consequently, when the vertical wind shear (i.e.,
the variation of wind speed with height) between these two levels is small. Activity tends to be
reduced when the QBO is in the easterly phase and the wind shear between 30 and 50 mb is large.
The QBO winds in fall 1998 are projected to be from a relatively easterly direction with large
shear, which should be an inhibiting influence on next year's hurricane activity, especially major
hurricanes.
b) African Rainfall
As discussed by Landsea and Gray (1992) and Gray et al. (1992), Atlantic hurricane activity is
influenced by rainfall anomalies in (1) the western Sahel in Aug-Sep of the previous year, and (2)
the Gulf of Guinea in Aug-Sep-Oct-Nov of the previous year. In both (1) and (2), wetness
indicates enhanced hurricane activity the following year. This is largely based on year-to-year
persistence of the rainfall anomalies, and the fact that there is a contemporaneous relationship
with hurricanes that is in the same sense as the lagged relationship. The rainfall for the West Sahel
during Aug-Sep 1997 was -0.73 standard deviations below average. We believe that a portion of
this decrease was due to the very strong 1997 El Niño. It has been documented that western Sahel
rainfall is reduced in El Niño years. Aug-Sep-Oct-Nov Gulf of Guinea rainfall influences intense
hurricanes in particular. The 1997 Gulf of Guinea rainfall was below average (-0.43 standard
deviations), again believed due to the anomalously strong El Niño of 1997. Because we expect
the warm ENSO situation to end or even reverse by late summer 1998 (see below), this year's
below average Gulf of Guinea rainfall would likely be better indicator of a reduced influence of
next year's hurricane activity if it had occurred in the absence of such an intense El Niño.
Therefore, we do not weight it as heavily as we otherwise might.
c) ENSO
The El Niño/Southern Oscillation (ENSO) is one of the principal global scale environmental
factors affecting Atlantic seasonal hurricane activity. Hurricane activity is usually suppressed (e.g.,
1997) when anomalously warm SSTs are present in the equatorial eastern and central Pacific.
Conversely, activity is usually enhanced during seasons with cold (or La Niña) SST conditions.
We expect current conditions to rapidly cool during the spring of 1998, resulting in near average
to cool conditions during the key months of August through October. This ENSO outlook is
based on a consensus of the many ENSO outlooks presented in this Bulletin (particularly the
climatology/persistence [CLIPER] ENSO forecast [Knaff and Landsea, 1997] which calls for SST
anomalies of -0.54C for Jun-Jul-Aug 1998, and -0.75C for Sep-Oct-Nov), and also on specific
diagnostic features of the present climate. These features resemble past events that dissipated
rapidly once boreal winter was over, as opposed to a more protracted lingering into the following
year as occurred after winter 1982-83 and 1986-87.
d) Strength of the Oct-Nov Atlantic Subtropical Ridge (Azores High) Located within 20-38N,
20-30W
High surface pressure in this eastern North Atlantic location reflects stronger east Atlantic trade
winds, which enhance upwelling of cold water off the northwest African coast. Colder SST is thus
created; this, in turn, can cause higher surface pressures to develop in the spring which is a
self-enhancing response. Such positive feedback can continue through the following year's
hurricane season, and would reduce the hurricane activity, while lower than normal pressure
would enhance activity. The ridge strength during Oct-Nov 1997 was significantly below average
(-1.37 standard deviations). This should be an enhancing influence on 1998 hurricane activity.
e) Others
(1) The configuration of SST anomaly patterns over much of the low and high latitude Atlantic.
Warm SST anomaly patterns in these regions during the summer and fall are usually associated
with an enhancement of hurricane activity the next summer/fall, and negative SST anomalies
patterns with a reduction of activity. This summer and fall's SST anomaly patterns have all been
anomalously warm, which should enhancing hurricane activity for 1998.
(2) Middle latitude circulation patterns during Sep-Oct-Nov. When middle latitude westerly
oceanic wind patterns are more zonal, and both the Aleutian low and the Icelandic low are
stronger (e.g.., blocking action in Atlantic is reduced), then hurricane activity during the following
summer is typically reduced. With the opposite conditions (when westerly circulation and the
Aleutian low pressure is weaker and more blocking action is present in the North Atlantic), the
next year's hurricane activity is typically enhanced. No data on this parameter are given for the
current hurricane outlook.
The LAD multiple regression predictions for 1998 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 a 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.
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 forecast, which includes an
upward qualitative (human) adjustment due largely to the expected dissipation of the
currently very strong El Niño, is for a slightly suppressed 1998 storm season. This represents
a result of the present and recent past states of the various predictors that are highlighted above.
Table 1. Predicted Atlantic tropical storm historical hindcast skills and forecasts for the 1998 season as of early December, 1997. Using the newest statistical scheme, skill expected for independent (real-time) forecasts made in December (shown after the slash in bold in the first column) has been estimated using resampling simulations. These are lower than the hindcasting skill levels (shown before the slash) obtained using a partial cross-validation design. Some human judgement was used to modify the objective statistical forecasts to obtain the final forecasts. | |||
ATLANTIC TROPICAL CYCLONE PARAMETER |
Skill (% variance) |
Final Forecast Dec 1997 | 1950-90 Mean |
Named storms | .52/33 | 9 | 9.3 |
Named storm days | .55/.37 | 40 | 47 |
Hurricanes | .49/.30 | 5 | 5.7 |
Hurricane days | .54/.36 | 20 | 24 |
Intense hurricanes | .44/.20 | 2 | 2.2 |
Intense hurricane days | .42/.16 | 3 | 4.7 |
Hurricane destruction potential | .49/.29 | 50 | 71 |
Net tropical cyclone activity (% avg) | .53/.35 | 90 | 100% |
Verification of 1997 Tropical Storm Forecasts
The 1997 hurricane season was characterized by reduced levels of tropical cyclone activity. There
was a total of 7 named storms (average is 9.3) and 3 hurricanes (average is 5.8) which persisted
for a total of only 10 days (average is 24). There was 1 major (intense) hurricanes of
Saffir/Simpson category 3-4-5 (average is 2.3 intense hurricanes) with but 2 intense or major
storm days (average is 4.7). The seasonal total of named storm days was 28 or only 60 percent of
the long-term average. Net tropical cyclone (NTC) activity was 54 percent of the average years of
1950-1990. Despite having a reduced level of tropical cyclone activity the 1997 season was still
more active in terms of NTC than 10 other of the last 30 hurricane seasons.
Table 2 summarizes the forecasts that we issued at the usual four times of the year, and the
observations.
Table 1. Verification of 1997 forecasts | ||||||
ATLANTIC TROPICAL CYCLONE PARAMETER |
Dec
1996
Forecast |
Apr
1997
Forecast |
Jun
1997
Forecast |
Aug 1997 Forecast |
Observed |
1950-90 Mean |
Named storms | 11 | 11 | 11 | 11 | 7 | 9.3 |
Named storm days | 55 | 55 | 55 | 45 | 28 | 47 |
Hurricanes | 7 | 7 | 7 | 6 | 3 | 5.8 |
Hurricane days | 25 | 25 | 25 | 20 | 10 | 24 |
Intense hurricanes | 3 | 3 | 3 | 2 | 1 | 2.3 |
Intense hurricane days | 5 | 5 | 5 | 3 | 2 | 4.7 |
Hurricane destruction potential | 75 | 75 | 75 | 60 | 70.6 | |
Net tropical cyclone activity (% avg) | 110 | 110 | 110 | 100 | 54 | 100% |
Maximum potential destruction | 70 | 70 | 70 | 60 | 61.7 |
Our early 1997 forecasts called for a slightly above average hurricane season, and the 6 August
forecast was for an average season. This overestimation of activity was a result of our inability to
anticipate that we would experience the most extreme El Niño event ever to be recorded during
the period of June through October. This unprecedented El Niño dominated other normally
hurricane-enhancing forecast signals, leading to a larger reduction in this year's hurricane activity
than would have occurred if this El Niño had not been so intense. Had we only experienced a
strong El Niño like 1972 and 1982 (as we anticipated in late July), it is likely that our forecasts
would have been close.
Gray, W.M., 1984: Atlantic seasonal hurricane frequency: Part I: El Niño and 30 mb
quasi-biennial oscillation influences Mon. Wea. Rev., 112, 1649-1668.
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.
Knaff, J. A. and C. W. Landsea, 1997: An El Niño-Southern Oscillation CLImatology and
PERsistence (CLIPER) Forecasting Scheme. Wea. Forecasting, 12, 633-652.
Landsea, C.W., and W.M. Gray, 1992: The strong association between Western Sahel monsoon
rainfall and intense hurricanes. J. Climate, 5, 435-453.