Transcript in malaria prediction

Using Seasonal Forecasts
Francisco J. Doblas-Reyes
[email protected]
ECMWF Training course
Reading, 27 April 2006
Forecasts are relevant for users
The user needs climate
information to take action
and mitigate the adverse
effects of climate
ECMWF Training course
Reading, 27 April 2006
Long-range forecast objective
“To utilize the ability to predict climate
variability on the scale of months to a year and
beyond to improve management and decision
making in respect to users’ needs at local,
regional, and national scales.”
ECMWF Training course
Reading, 27 April 2006
Long-range forecast objective
“To utilize the ability to predict climate
variability on the scale of months to a year and
beyond to improve management and decision
making in respect to users’ needs at local,
regional, and national scales.”
Requirements by the end user:
• predict climate variability: skilfully deal with uncertainties in
climate prediction
• seasonal-to-interannual time scales: coupled oceanatmosphere general circulation models
• variable spatial scale: downscaling
ECMWF Training course
Reading, 27 April 2006
A user strategy: the end-to-end approach
•
•
•
•
•
A broad range of forecast products might be
offered, but user requirements need to be defined.
End-to-end is based on collaboration and
continuous feedback.
End users develop their models taking into account
climate prediction limitations.
The level of forecast skill that provides added value
is defined by the application: user-oriented
verification. End users assess the final value of the
predictions.
Forecast reliability becomes a major issue.
ECMWF Training course
Reading, 27 April 2006
End-to-end: DEMETER
http://www.ecmwf.int/research/demeter/
•
•
•
•
•
Research project funded by the Vth FP of the EC,
with 11 partners.
Integrated multi-model ensemble prediction system
for seasonal time scales.
More than a multi-model exercise: seasonal
hindcasts used to assess the skill, reliability and
value of end-user predictions.
Applications in crop yield and tropical infectious
disease forecasting.
Officially finished in September 2003, but with an
operational follow up.
ECMWF Training course
Reading, 27 April 2006
DEMETER
Special Issue 2005
ECMWF Training course
Reading, 27 April 2006
Extremes for users: end-to-end
1
2
3
4
…………
62
63
Climate
forecast
1
2
3
4
…………
62
63
Downscaling
63
Application
model
1
2
3
4
…………
62
non-linear transformation
0
0
Probability of Precipitation
Probability of Crop Yield/Incidence
ECMWF Training course
Reading, 27 April 2006
Downscaling for s2d predictions
http://www.ecmwf.int/research/EU-projects/ENSEMBLES/news/index.html
ECMWF Training course
Reading, 27 April 2006
Downscaling for s2d predictions
•
•
•
•
•
Use dynamical and empirical/statistical methods.
Correct systematic errors of global models and
obtain reliable (statistical properties similar to the
observed data) probabilistic predictions (with only
relatively short, i.e., 15-30 years, training samples).
Deal with full ensembles, not a deterministic
prediction or the ensemble mean, maximising the
benefit of limited simulations with regional models.
Consider model and initial condition uncertainty.
Generate high-resolution (e.g. daily) time series of
surface variables (using, e.g., weather generators
with statistical methods).
ECMWF Training course
Reading, 27 April 2006
Examples of applications
•
•
•
Malaria incidence prediction in an epidemic region
(Botswana).
Crop yield prediction for Europe (wheat) and
western India (groundnut).
Seasonal streamflow prediction over tropical and
subtropical watersheds.
ECMWF Training course
Reading, 27 April 2006
Predictions for large agricultural areas
1-month lead spring (MAM)
T2m over Ukraine
ECMWF Training course
3-month lead early spring
(ASO) precipitation over
Eastern Australia
Reading, 27 April 2006
JRC’s CGMS
Crop Growth
Indicator
Meteo data
Statistical model
Seasonal forecast data
Yield
Meteo data
Jan Feb
ECMWF Training course
Aug
Reading, 27 April 2006
Wheat yield predictions for Europe
DEMETER multi-model predictions (7 models, 63 members, Feb starts) of
average wheat yield for four European countries (box-and-whiskers)
compared to Eurostat official yields (black horizontal lines) and crop results
from a simulation forced with downscaled ERA40 data (red dots).
France
Denmark
Germany
SIMULATION
WEIGHTED YIELD
ERROR (%)
± STANDARD ERROR
JRC February
7.1 ± 0.9
JRC April
7.7 ± 0.5
JRC June
7.0 ± 0.6
Greece
JRC August
5.4 ± 0.5
DEMETER (Feb.
start)
6.0 ± 0.4
From P. Cantelaube and J.-M. Terres, JRC
ECMWF Training course
Reading, 27 April 2006
Groundnut yield predictions with a LAM
Correlation between de-trended observed and DEMETER ensemble-mean
predicted groundnut yields for the period 1987 -1998
From Challinor et al. (2005)
ECMWF Training course
Reading, 27 April 2006
Malaria early warning systems
gathering cumulative evidence for early and focused response . . .
geographic/community focus
case surveillance alone = late warning
From M. Thomson (IRI)
ECMWF Training course
Reading, 27 April 2006
Malaria warning: meteorological factors
Limiting variables for malaria development as obtained with the MARA rulebased model and ERA40; white areas are influenced by all factors
The number of meteorological variables
required by the users is large and changes
with the region considered
From A. Jones (Univ. of Liverpool)
ECMWF Training course
Reading, 27 April 2006
Malaria warning: link to climate
Statistical relationship between DJF CMAP precipitation and
Botswana standardised log malaria incidence for 1982-2002
ECMWF Training course
Reading, 27 April 2006
Climate forecasts for malaria warning
Precipitation composites for the five years with the highest (top
row) and lowest (bottom row) standardised malaria incidence for
DJF DEMETER (left) and CMAP (right)
Areas with
epidemic
malaria
Quartiles define
extreme events
(epidemics) in
malaria
prediction
ECMWF Training course
Reading, 27 April 2006
Malaria warning with statistical model
Probabilistic predictions of standardised malaria incidence
quartile categories in Botswana with five months lead time
Very low
malaria
-- high malaria years
-- low malaria years
Available in
March
Available in
November
ROC Score
Precipitation
Event
DEMETER
CMAP
DEMETER
Very low
0.95
1.00
1.00
Very high
0.52
0.94
0.84
ECMWF Training course
Incidence
Very high
malaria
Reading, 27 April 2006
Dynamical malaria model: bias correction
Cumulative frequency
Daily precipitation as required by the Liverpool Malaria Model
Daily rainfall from the
CERFACS experiment (25°E,
22.5°S, November start
date, 1980-2001), correction
End users require
applied separately for dry
probabilistic models
and wetthat
days, with wet days
correct biases, downscale to
corrected with a ratio
Daily rainfall (mm)
the appropriate grid and are
ERA40 raw modelable
correct
model
to produce
daily time
Rainfall histograms
series with the correct
extremal properties
(CERFACS, all Botswana
grid points, November
start date, 1980-2001)
From A. Jones (Univ. of Liverpool)
ECMWF Training course
Reading, 27 April 2006
Malaria warning: nonlinearity
Malaria index for Botswana from Thomson et al. (2006) and incidence
simulated by the Liverpool malaria model (LMM) using ERA40
There is a disagreement between both
models for the year 2000: is it due to the
impact of extreme temperature or
precipitation? Interaction of climate
variables may affect the user predictions
rain (mm per month)
270.0
240.0
27.0
0.4
210.0
180.0
0.2
150.0
0.0
120.0
90.0
-0.2
60.0
-0.4
30.0
-2.0
28.5
0.6
0.0
01/98
01/99
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000
-0.6
25.5
24.0
22.5
21.0
19.5
18.0
16.5
15.0
01/00
01/01
Date
From A. Jones (Univ. of Liverpool)
Year
ECMWF Training course
Reading, 27 April 2006
TMax -5 (deg C)
Rainfall (mm per month)/Monthly Incidence
Malaria Anomaly
-1.0
30.0
LMM incidence
2.0
0.0
tmax-5
Incidence Anomaly
Malaria Index
1.0
incidence per month
300.0
Interacting factors in end-user systems
•
•
•
•
The predictions are designed to be included in an
early warning system (decision making).
Tropical disease incidence is an important factor
affecting food security in tropical/semi-arid areas
(socio-economic interaction).
The previous example deals with uncertainty in
malaria prediction using a probabilistic approach to
reduce forecast error and can easily be extended to
prediction of climate-related crop yields (uncertainty).
Seasonal prediction allows users to become familiar
with the use of climate information and understand
methods to mitigate the impact of and adapt to future
global change (climate change).
ECMWF Training course
Reading, 27 April 2006
Climate change and climate variability
•
•
•
The possibility of adaptation to climate change via a
learning process taking place at the interannual
time scale is an obvious way to achieve a high
degree of integration of climate time scales.
It implies:
Involvement of both climate scientists and end-users
That both scientists and end users/stakeholders
consider the whole range of time scales
As an example, crop managers see the adaptation
to long-term climate change as a process that takes
place on a yearly basis and that benefits from
predictions at various time scales.
ECMWF Training course
Reading, 27 April 2006
River basin predictions
Multi-model predictions of precipitation over river basins and
many other verification diagnostics
http://www.ecmwf.int/research/demeter/d/charts/verification/
ECMWF Training course
Reading, 27 April 2006
Combined/calibrated seasonal predictions
Forecast
Observations Multi-model Assimilation
• 3 DEMETER
coupled models
• 1-month lead time
DJF precipitation
• ENSO composites
for 1959-2001
• 16 warm events
• 13 cold events
From Coelho et al. (2006)
ECMWF Training course
r=0.51
r=0.97
r=0.28
r=0.82
(mm/day)
Reading, 27 April 2006
Calibrated downscaled predictions
PAGE agricultural extent
PAGE agroclimatic zones
ECMWF Training course
Reading, 27 April 2006
Calibrated downscaled predictions
Seasonal predictions of NDJ precipitation (3-month lead time)
Northern box
Southern box
Forecast
Correlation
BSS
Forecast
Multi-model
0.57
0.12
Forecast
Assimilation
0.74
0.32
Correlation
BSS
Multi-model
0.62
0.16
Forecast
Assimilation
0.63
0.28
From Coelho et al. (2006)
ECMWF Training course
Reading, 27 April 2006
Calibrated downscaled predictions
Seasonal predictions of NDJ precipitation (3-month lead time)
Forecast
Correlation
BSS
Parana
0.16
0.00
Tocantins
0.29
0.12
From Coelho et al. (2006)
ECMWF Training course
Reading, 27 April 2006
Summary
•
•
•
•
The multi-model ensemble has proven to be an
effective approach to reduce forecast error by
tackling both initial condition and model uncertainty.
The end-to-end approach has shown promising
results in seasonal forecasting, especially in a
probabilistic framework.
There is a clear need to link the research and
development carried out about climate variability at
different time scales and the users’ needs.
Seasonal-to-interannual forecasting can evolve into
a field where end-users learn to use (and verify)
climate information before developing adaptation/
mitigation strategies for global change.
ECMWF Training course
Reading, 27 April 2006
Questions?
ECMWF Training course
Reading, 27 April 2006