The Norwegian experience with Seasonal Forecasting

1
The Norwegian experience with Seasonal Forecasting

• R.E. Benestad

2
Seasonal forecasting

• Motivation
• Scientific challenge
• Media pressure
• Preliminary empirical studies and prognoses from
  ECMWF
• Examples
• Challenges
• Outreach do the users understand the message?

3
Reasons to believe that seasonal forecasting is
possible?

• The old 'Newtonian' school of thought perfect
  knowledge.
• State is a consequence of a
• series of events? (predictable)
• Chaotic? (not predictable)
• Conditional chaos?
• (some predictability)
• Evidence for external forcings
• Geographical differences
• Ice ages
• The annual cycle

2m temperature anomalies w.r.t. latitudal mean.
4
A trivial example the seasonal cycle in
temperature and precipitation are affected by
annual variations in insolation. Despite 'chaos'
the statistics is shifted by changes in external
factors
5
Seasonal forecasting at the Norwegian
Meteorological Institute
Collaboration with Natsource-Tullett (power
trading). Benchmarking against MeteoConsult
SMHI.
Two approaches empirical-statistical forecasts
dynamical (ECMWF) Forecasts.
6
Some early examples statistical-empirical
methods SST, SLP from DNMI analysis past
observations. CCA -gt r0.44 (Bergen, Feb-April)
7
Simple statistics past observations. Clear
stratification!
8
Errors in the data non-stationary
relationships Destroy empirical relationships.
Solution random sub-sampling?
9
Problem short records of continually updated
values, Long historical records that are frozen
merging of different data sets. Trivial but not
so trivial
10
Quick fix PC3 set to zero (not a god solution)
11
ECMWF prognoses adapted at met.no
Adjusting anomalies to the 1961-90 climatology
absolute values give unrealistic results (due to
drifts?).
12
Oslo Forecasts based on interpolation of ECMWF
anomalies
Oslo ensemble spread for Sept.-Nov. 2005
High correlation due to climate trend! 1961-90
climatology used consistently for reports.
13
Sesongvarsel for juni - august 2005 Utstedt
18.05.2005
Sesongkart for juni august 2005 basert på
observasjoner Utgitt 01.09.2005
Utjevnet temperaturavvik i oC fra normalen for
sesongen
Utjevnet temperaturavvik i oC fra normalen for
sesongen
Normalperioden er 1961-1990
14
What do the users think the forecasts represent?
15
Probability density functions.
16
Forecast evaluation
Correlation scores RMSE Contingency tables (hit
ratio) Brier scores for probs. Cross-validation.
Actual forecasts.
17
Worst years 1995 2001
Sea Surface Temperature
Interesting why different to T(2m)?
18
Climate change seasonal forecasting
The day(s) after tomorrow...

• If there is an ongoing climate change, then
  there will be a great need for good seasonal
  forecasts in order to be ready for unexpected
  events (e.g. Summer 2003, autumns 2000, 2002,...)
• Synergy effects seasonal forecasting will draw
  from the experience gained from climate studies
  climate research will benefit from seasonal
  forecasting research (e.g. Coupled models for
  ENSO prediction).

19
Extra slides
20
Common EOF analysis DNMI_sst ECMWF SST
!
21
(No Transcript)
22
Many attempts in the past, but still no success.
Why do we believe that we now can do any better?

• More observations longer records, improved
  coverage (global, oceans, stratosphere), more
  elements (snow, ice, stratosphere, vegetation,
  sea level).
• Greatly improved computational capacity.
• New methods and models.
• Improved understanding.
• Improved infrastructure.

23
Some early examples statistical-empirical
methods SST, SLP, soil moisture, humidity from
NCEP re-analysis, SOI, past observations.
Regression -gt r0.03 (Jul)
24
The way forward a combination of three
approaches

• Data possibly the limiting factor hard and
  expensive to get good reliable data (for the
  past). Errors?
• Models Improving, but still crude. Important
  tool for study and for making forecasts
• Statistics For testing and exploring hypotheses
  and for searching for precursory signals. Simple
  forecasts.

Data
Stats
GCMs
Extensive collaborations required a 'Manhattan'
type project. Dialgoue between various
disciplines. Trust.
25
Seasonal forecasting activity at met.no

• Evaluation of ECMWF products for Norway
• Empirical-statistical month-seasonal forecasts.
• Data processing, testing, merging, decomposition.
• Data DNMI analysis, NCEP reanalysis, ECMWF
  analysis, ECMWF seasonal forecasts, station
  observations.
• Reports, Proposals (NFR, IPY)
• Trial error

26
Status search for data mismatch. (SST sea
surface temperature)
?
?
?
?
27
How good are the seasonal forecast products at
ECMWF?
Worst years 1988 1995 1997 1999 2001
Positive/negative/near zero 8 / 5 / 4
Temperature
28
Worst years 1988 1991 1995 1999 2002
Positive/negative/near zero 8 / 5 / 3
Precipitation
29
How to present the forecasts?
30
Vardø