Winter Season Forecasting Using the Winter Disruptiveness Index

1
Winter Season Forecasting Using the Winter
Disruptiveness Index

• Methodology and results of the 2002/2003 Winter
  season forecast.

Dan Swank Meteo 497 Long Range Forecasting
2
The Winter Disruptiveness Index (WDI)

• A quantitative measure of winter season severity.
• Defined over the time period November through
  March
• Designed to be applicable everywhere
• A numeric scale Higher values denote cold and
  snowy winter seasons
• Negative values for mild winters
• Values are the sum of seven components

3
WDI Value scale
4
The 7 WDI Components What can be forecasted
with WDI?

• Average NDJFM Temperature
• Total NDJFM Snowfall
• NDJFM days with gt 1 snowcover
• Abnormally cold days
• Abnormal daily snowfall
• Abnormal daily rainfall
• Ice storms

5
(No Transcript)
6
(No Transcript)
7
How can we forecast the next winters WDI?

• Use Analog forecasting. Currently two
  experimental methods.
• Compare global 500 mb height anomalies during
  summer/autumn months for years when the WDI is
  between a certain range (i.e. gt 12)
• Correlate average monthly values of
  oceanic-atmospheric indices, before November, to
  that seasons WDI value.

8
Averaged 500 Mb Height anomaliesAnalog
forecasting method

• This method can gives a general idea of the
  likely outcome of the coming winter, but does not
  give an exact value for the WDI
• Take the average height anomaly over 3 months,
  such as August, September, and October. Average
  them over all years where the WDI is within a
  given range
• Maps shown on the next slide are composites of 4
  years where the WDI was between a set range.
• These maps would be different if WDI value at
  another location are used
• Interestingly, teleconnection nodes tend to show
  u in the averaged 500 mb analyses

9
WDI gt12
6 to 12
0 - 6
lt-4
10
2002 August to October (ASO) 500 mb height
anomalies
11
500 mb analog method

• Often anomaly comparisons may be inconclusive.
  Unless patterns similar to the extreme cases are
  present, go near normal.
• The 2002/03 pattern best matched the harsh (6 to
  12) regime. Although vaguely.
• This method can also be applied with any of the 7
  WDI components, to forecast likely temperature
  and precip trends.
• Other month ranges (besides ASO) can be used,
  however months closer to November will probably
  be more reliable.

12
Oceanic Atmospheric Indices Analog forecasting
method

• Correlate the WDI to averaged monthly values of
  Ocean/Atmospheric indices such as the NAO, SOI,
  and PNA using various lag/span computations.
• For example Each years average April through
  August NAO correlated with the value of WDI for
  the following winter, starting in November.
• Much more complicated then the 500 mb method, but
  gives an exact forecast value for the WDI.
• Used to make the 2002/03 forecast

13
Oceanic Atmospheric Indices method

• Must use a computer program to calculate the
  millions of correlation possibilities.
• Output from the program can be accessed via a web
  form http//pasc.met.psu.edu/PA_Climatolo
  gist/WDI/correlform.html
• Use the best 4 predictors that can be found. The
  values of indices must be taken over months
  before the winter occurs, in order to be useful
  for forecasting.

14
Correlation coefficients

• The WDI correlation calculations were done with
  the Pearson Correlation Coefficient (R).
• 1 for a perfect correlation, 0 for no
  relationship what-so-ever.
• gt 0.6 indicate a good relationship exists between
  the two datasets
• 0.2 to 0.6 represents a weak relationship.
• The best WDI correlations fall between 0.4 and 0.6

15
1 Find the best predictors - Example EPO
WDI correlations in State College
EPO is undefined in August and Sept.
16
For the 2002/03 winter forecast, the following
predictors were used
17
Next steps

• Make a table of values, listing the WDI, PNA,
  EPO, AO, and SOI values for each year where data
  is available.
• Obtain the index values for the current year.
• Make a list of analog years where the current
  index values match previous years
• Also keep track of how many indices each analog
  year matched
• Take a weighted average of the analog years WDI
  values.

18
Analog Years
19
2002-03 Analog years
Listing of analog years, which matched one
(single) index two (double) atmospheric/oceanic
indices
Taking the weighted average of each analog
winters WDI (the double match years are double
weighted), gives the value of roughly WDI
-0.5 Rounded to the nearest 0.5
20
Oceanic/Atmospheric index method summary

• More specific and calculation intensive then the
  500 mb method
• Correlation values may be too low to be
  dependable
• Predicted a normal to slightly mild winter for
  2002-03
• The two methods should be compared to see if they
  agree
• Other methods not involving the WDI should be
  incorporated into the final forecast.

21
The 2002/03 winter forecast

• WDI forecasted to be from 0 to 2, after
  considering other techniques
• When the WDI is in this range, the typical
  conditions usually occur, typical of an average
  winter season in this area
• -0.8 to 0.7 degree departure from average NDJFM
  temperature.
• 34-43 inches of snowfall
• 32 to 44 days with 1 snowcover
• 1 major (12) snowstorm, or 2 moderate snowfalls
• 1 storm with minor ice accumulation

22
STC Verification

• WDI 8.0
• Components, statistics and averages
• Tmean 1.37 (30.7 F, AVG 32.8 )
• Smean 2.39 (75.1, AVG 41)
• SCmean 1.61 (60 days SCgt1, AVG 36 days)
• Tdaily 0.20 (16 DCDs, AVG of 15)
• Sdaily 2.5
• Rdaily 0 Idaily 0
• Winter as much colder and snowier then expected.
  Total snowfall was nearly twice the average.
• However, the predication did not indicate a mild
  winter, which is what most people are adjusted to
  because of the past few years

23
Insight explanations

• Analog forecasts are subject to error because of
  the relatively short period of record of existing
  weather data
• The WDI definition was changed since the forecast
  was made, values were amplified
• Weak correlation values
• Perhaps a component-wise analog method would be
  more accurate, this would also give more insight
  into the temperature and precipitation breakdowns
• Weather patterns can change drastically over the
  period of 5 months
• A few more methods should be developed which use
  the WDI to make a seasonal forecast. 2 methods
  may not be enough