Introduction to seasonal climate forecasting

1
Introduction to seasonal climate forecasting

• Scott Power
• Bureau of Meteorology
• Ackowledgements
• D. Jones, P. Reid, R. Fawcett, NCC
• W. Drosdowsky, N. Nicholls, L. Chambers, O.
  Alves, BMRC
• T. Kestin formerly BMRC now NCAR

2

• Empirical or statistical prediction
• relatively straightforward, cheap, transparent
• crucial to have long continuous good quality
  records
• use relationships between variables in historical
  record
• relationships must be lagged, physically
  plausible
• assumes that past relationships will hold in
  future
• chaos ? imperfect relationships ? probabilities
  used in forecasts
• avoiding artificial skill, importance of
  independent data
• beware complex implausible schemes
• statistical forecasts will be important for
  years to come
• statistical forecast systems provide valuable
  benchmark

3
For a particular location the wet season rainfall
distribution curves may look like
July SOI lt -5
-5 lt July SOI lt 5
July SOI gt 5
For example During El Niño events, this location
receives lower rainfall more often than normal or
higher rainfall
4
The Rainfall Chocolate Wheel
July SOI lt -5
-5 lt July SOI lt 5
July SOI gt 5
For this particular location, if July SOI lt -5 ,
then there is a 50 chance of receiving below
average rainfall, a 25 chance of average
rainfall, and a 25 chance of higher than
average rainfall But the wheel must still be
spun!
5
For a small number of spins, the results might
almost be random but over a large number of
spins those odds will emerge. Climate
forecasting through probabilities is the same -
long term use of the odds will inevitably produce
a long term skill, but in the short term
errors will be perceived. Careful
communication of the seasonal climate outlooks
is therefore essential.
6
There must be a relationship between the
variables used to predict and the thing being
predicted ahead of time
7
SST phases of years exceeding median rainfall
8
SOI phases of years exceeding median rainfall
9
Artificial Skill

• danger of too many inputs
• danger of trying too many inputs
• independent data
• cross-validation
• importance of supporting evidence
• simple plausible hypothesis
• climate models
• process studies

10
(No Transcript)
11
Bureau Operational scheme

• Four inputs
• SST1 last month and 2 months before that
• SST2 last month and 2 months before that
• used to predict rainfall and surface temperature
  across Australia over next 3 months

12
(No Transcript)
13

• Numerical Prediction
• solve mathematical equations representing the
  physics of the atmosphere-ocean system
• seen as future of climate forecasting
• e.g. have potential to explicitly include CO2
  increases
• research ongoing, have not reached potential,
  havent replaced statistical forecast methods,
  verification initialisation data limited
• hybrid methods promising
• chaos - uncertainty - probabilities
• long-term quality data remains crucial for
• initialisation
• verification
• parameterisation, improving models
• hybrid methods e.g. downscaling

14
The Formula
DSxT where D distance travelled,
Sspeed Ttime taken
DST
15
How predictable is ENSO?Sensitivity experiments
Sensitivity of NINO4 index to small initial
nudges
NINO4
Chaos limits predictability
Time (Years 1 to 4))
16
UNCERTAINTY BEGINS..
The cruise control is faulty Varies between
98 and 102 kph How far will the car travel in 2
hours?
17
Just where is that car?
Without precise information on how the cruise
control is varying we cannot give a precise
answer. But we can say between 196 and 204
km. This is an error band of 8 km.
18
Probability icebergs
19
The Weather Equations
Two horizontal equations of motion The
hydrostatic equation The continuity equation
The thermodynamic equation The equation of
state The potential temperature equation
20
Mathematical modelling of the weather
Weathersmart image
21
Climate Model Forecasts

22
(No Transcript)
23
Coupled Model Dynamical Forecasts
Strength of El Nino
24
NEC SX-5 Supercomputer
16 Vector processors Processing speed of 128
Gflops/sec 1024 Gbytes of bandwith Read Big
and Fast
25

• Helping people use climate predictions prudently
• raising awareness of forecasts
• communicating probabilities uncertainty
• presenting information

26
Communication

• Deliver services
• Raise awareness of services available
• Get feedback on existing services
• Get ideas for new services
• Help people use our services
• Explain services
• Explain underpinning science
• Explain underpinning technology

27
Communication channels

• radio, TV, newspapers, the web, field days, phone
  calls, emails, public lectures, regular scheduled
  meetings, fax services, brochures, atlases,
  books, ...

28
Helping people to use our services

• educational glossies, field days, public
  addresses,
• media interviews, web-based information,
  conferences, articles in rural magazines,
• Staff training (marketing, media liaison,
  communication)

29
National Weather Service Seasonal Climate
Outlook Over SE Australia, there is a 60 to
70 chance that rainfall will be less than the
median over the next 3 month period. So with
climate patterns like the current, about 6 or 7
seasons out of 10 are expected to be drier than
average in this area, with about 3 or 4 out of
10 being wetter. Media headline EXPERTS TIP
RETURN OF THE DROUGHT
30
Impediments to the use of seasonal climate
predictions

• limited skill
• users locked in
• not relevant to users decisions
• decisions mae using superior method
• dont fully understand or trust information
• User resistance or misuse (eg., user
  conservatism)
• Cognitive illusions (eg., probability illusions)

31
People often have trouble using uncertain
information
32
use consistent terminology in forecasts - use
numbers- words can be vague
33
Framing effect

• If doctors were told there is a mortality rate of
  7 within 5 years for a certain operation, they
  hesitated to recommend it to their patients.

34
Framing effect

• If doctors were told there is a mortality rate of
  7 within 5 years for a certain operation, they
  hesitated to recommend it to their patients.
• BUT, if they were told it had a survival rate
  after 5 years of 93, they were more inclined to
  recommend it to their patients.

35
Framing effect

• Your country faces an outbreak of an unusual
  disease which is expected to kill 600 people. Two
  alternative programs to combat the disease have
  been proposed. Which program (A or B) would you
  favour?
• If A is adopted, 200 people will be saved
• If B is adopted there is a 1/3 probability that
  600 people will be saved and a 2/3 probability
  that nobody will be saved

36
Framing effect

• Your country faces an outbreak of an unusual
  disease which is expected to kill 600 people. Two
  alternative programs to combat the disease have
  been proposed. Which program would you favour?
• If A is adopted, 200 people will be saved
• If B is adopted there is a 1/3 probability that
  600 people will be saved and a 2/3 probability
  that nobody will be saved
• Now, which of these alternatives (C or D) would
  you favour?
• If C is adopted 400 people will die
• If D is adopted there is a 1/3 probability that
  nobody will die, and 2/3 probability that 600
  people will die

37
Framing effect(results of surveys in red)

• Your country faces an outbreak of an unusual
  disease which is expected to kill 600 people. Two
  alternative programs to combat the disease have
  been proposed. Which program would you favour?
• If A is adopted, 200 people will be saved 72
• If B is adopted there is a 1/3 probability that
  600 people will be saved and a 2/3 probability
  that nobody will be saved 28
• Which of these alternatives would you favour?
• If C is adopted 400 people will die 22
• If D is adopted there is a 1/3 probability that
  nobody will die, and 2/3 probability that 600
  people will die 78

38
Availability

• Which of the following causes more deaths in the
  USA each year?
• Stomach cancer
• Motor vehicle accidents

39
Availability

• Which of the following causes more deaths in the
  USA each year?
• Stomach cancer
• Motor vehicle accidents
• Most respondents select motor vehicle accidents,
  but stomach cancer causes twice as many deaths.

40
Availability

• Which of the following causes more deaths in the
  USA each year?
• Stomach cancer
• Motor vehicle accidents
• Most respondents select motor vehicle accidents,
  but stomach cancer causes twice as many deaths.
• The availability of media stories about motor
  vehicle deaths biases our perception of the
  frequency of events.

41
Overconfidence

• Select from each pair the most frequent cause of
  death (and decide how confident you are)?
• All accidents, or heart attacks?
• Homicide, or suicide?
• Breast tumour, or diabetes?

42
Overconfidence

• Select from each pair the most frequent cause of
  death (and decide how confident you are)
• All accidents, or heart attacks?
• Homicide, or suicide?
• Breast tumour, or diabetes?
• First alternative is selected with great
  confidence by most respondents.
• The correct answer is the second in each pair.

43
Decision regret

• Regret associated with a loss incurred by an
  action tends to be more intense than the regret
  associated with inaction or a missed opportunity
• Worldly wisdom teaches that it is better for
  reputation to fail conventionally than to succeed
  unconventionally (Keynes, 1936)

44
Belief persistence

• Primacy effect. Asch (1946) gave subjects a list
  of adjectives describing a person, such as
• intelligent, industrious, impulsive, critical,
  stubborn, envious, or
• envious, stubborn, critical, impulsive,
  industrious, intelligent.
• The impressions of the person were more
  favourable given the first list than the second!
• Inertia effect. Lord et al (1979) gave subjects
  (who had previously indicated that they favoured,
  or opposed, capital punishment) two reports, one
  purporting to show that capital punishment was
  effective in deterring crime, the other
  purporting to show that it was ineffective.
  Subjects beliefs strengthened, regardless of
  whether they favoured or opposed capital
  punishment!

45
Which statement gives you a clearer understanding
of the climate forecast? A. Based on the rising
SOI phase in the last 2 months, the probability
of above median rainfall in the spring is 80.
B. The SOI has been in a rising phase in the
last 2 months. There are 30 other years in the
historical record when this has occurred, at this
time of year. In 24 of these (i.e., 80) there
was above median rainfall in the next three
months.
Coventry, W. L., 2000. Unpublished thesis, Univ.
of Queensland.
46
Which statement gives you a clearer understanding
of the climate forecast? A. Based on the rising
SOI phase in the last 2 months, the probability
of above median rainfall in the spring is 80.
29 B. The SOI has been in a rising phase in
the last 2 months. There are 30 other years in
the historical record when this has occurred, at
this time of year. In 24 of these (i.e., 80)
there was above median rainfall in the next three
months. 71
Coventry, W. L., 2000. Unpublished thesis, Univ.
of Queensland.
47
Care needed in describing forecasts!
48
Hindsight

• Prior to Richard Nixons trips to China and
  Russia in 1972, Fischhoff and Beyth asked
  students to consider 15 possible outcomes such
  as
• Nixon will meet Mao Tse-tung at least once
• Nixon will see Soviet demonstrators
• The students were asked to assign a probability
  to each possible outcome.

49
Hindsight

• Prior to Richard Nixons trips to China and
  Russia in 1972, Fischhoff and Beyth asked
  students to consider 15 possible outcomes such
  as
• Nixon will meet Mao Tse-tung at least once
• Nixon will see Soviet demonstrators
• The students were asked to assign a probability
  to each possible outcome.
• After the trip, the students were asked, in
  hindsight, to remember their original (pre-trip)
  probabilities
• When the interval between the tests was two
  weeks, 67 thought their original estimates were
  closer to the truth than they really were.
• When a 4-8 month interval had elapsed, 84 showed
  hindsight bias.

50
Improving discription of climate predictions

• Talk to users. How do they interpret
  probabilities (worded and numerical)?
• Use consistent terminology, say things in more
  than one way, refer to multiple versions of
  forecasts or analogs, use numbers and relative
  frequencies, talk things over, independent
  review!

See Nicholls, N., 1999. Cognitive illusions,
heuristics, and climate prediction. Bull.
American Meteorological Society, 80, 1385-1397.
for more details on biases people have