At the FAO Workshop, held in Panama City on August 6 - 8th the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), Head of Research James Hansen gave a presentation on CRAFT tool. More info: http://ow.ly/ocIqJ
1. 1
The CCAFS Regional Agricultural
Forecasting Toolbox (CRAFT)
James Hansen, Theme 2 Leader
International Research Institute for Climate and Society
Herramientas para la Adaptación y Mitigación del Cambio Climático
en la Agricultura en Centroamérica
Panamá, 6-8 de Agosto 2013
Date
2. 2
What is CRAFT?
• Software platform to support within-season forecasting of crop
production; secondarily, risk analysis and climate change impacts
• Functions:
• Manage spatial data, crop simulation (currently DSSAT)
• Integrate seasonal forecasts (CPT)
• Spatial aggregation
• Probabilistic analysis
• Post-simulation calibration
• Visualization
• Analyses: risk, forecast,
hindcasts, climate change
• Current version preliminary
3. 3
What is CCAFS?
• Strategic partnership of international agriculture (CGIAR)
and global change (Future Earth) research communities
4. 4
What is CCAFS?
• Strategic partnership of international agriculture (CGIAR)
and global change (Future Earth) research communities
• World’s largest research program addressing the
challenge of climate change and food security
Mechanism for organizing, funding
climate-related work across CGIAR
Involves all 15 CGIAR Centers
$67M per year
5. 5
What is CCAFS?
• Strategic partnership of international agriculture (CGIAR)
and global change (Future Earth) research communities
• World’s largest research program addressing the
challenge of climate change and food security
• 5 target regions across the developing world
6. 6
What is CCAFS?
• Strategic partnership of international agriculture (CGIAR)
and global change (Future Earth) research communities
• World’s largest research program addressing the
challenge of climate change and food security
• 5 target regions across the developing world
• Organized around 4 Themes:
• Adaptation to progressive change
• Adaptation through managing climate risk
• Pro-poor climate change mitigation
• Integration for decision-making
8. 8
Why CRAFT?
Meets an unmet need
• Platform to facilitate research, testing and implementation
of crop forecasting methods
• Target researchers and operational institutions in the
developing world
• Accessible: free, open-source (eventually)
• Adaptable: support multiple crop model families
9. 9
Basics of yield forecasting:
Uncertainty
• Consider yields simulated
with monitored weather
thru current date, then
sampled historic weather
• Uncertainty diminishes as
season progresses
• Model error the non-
climatic component
• Relative contribution of
climate, model uncertainty
changes through the
season
forecast
date harvest
season
onset
Time of growing season
1
2
.
.
.
n
T
Weatherdatayear
monitored
weather
historic
weather
0.5
1.0
1.5
2.0
Grainyield,Mg/ha
1May 1Jun 1Jul 1Aug Harvest
Forecast Date
90th
75th
50th
25th
10th
1989 climatology-based Qld.
Australia wheat forecast. Observed,
and forecast percentiles. Hansen et
al., 2004. Agric. For. Meteorol.
127:77-92
Uncertainty
planting anthesis harvest
Time
model uncertainty
climate uncertain
SIMULATION
◄——— PREDICTION ———
Hansen, J.W., Challinor, A., Ines, A.V.M, Wheeler,
T., Moron, V., 2006. Climate Research 33:27-41.
10. 10
Basics of yield forecasting:
Reducing uncertainty
Uncertainty
planting anthesis harvest
Time
model uncertainty
climate uncertainty
Uncertainty
planting anthesis harvest
Time
2b. N-limited
3. Actual
1. Potential
pests, disease,
micronutrients,
toxicities
H, T,
crop
charac-
teristics
water2a. Water-limited
??????
soil N dynamics,
plant N use,
stress response
photosynthesis,
respiration,
phenology
water balance,
transpiration,
stress response
Level of production Processes
nitrogen
after Rabbinge, 1993
• Reduce model error:
• Improve model
• Improve inputs
• Assimilate monitored state
• Greatest benefit late in season
• Reduce climate uncertainty
• Incorporate seasonal forecasts
for remainder of season
• Greatest benefit early in season
Uncertainty
planting anthesis harvest
Time
model uncertainty
climate uncertainty
Uncertainty
planting anthesis harvest
Time
11. 11
Incorporating seasonal forecasts:
Queensland wheat study (2004)
• WSI-type crop model
• PC1 of GCM (ECHAM4.5)
rainfall, persisted SSTs
• Yields by cross-validated
linear regression with
normalizing transformation
• Probabilistic, updated
• Demonstrated yields more
predictable than rainfall
• One of several potential
methods tested 200 0 200 400 km
Correlation
< 0.34 (n.s.)
0.34 - 0.45
0.45 - 0.50
0.50 - 0.55
0.55 - 0.60
0.60 - 0.65
> 0.65
Rain
Yield
Hansen, Pogieter, Tippett, 2004.
Agric. For. Meteorol. 127:77-92
N
200 0 200 400 km
1 July
1 June
1 August
1 May
Correlation
<0.34 (n.s.)
0.34-0.45
0.45-0.55
0.55-0.65
0.65-0.75
0.75-0.85
> 0.85
0.5
1.0
1.5
2.0
Grainyield,Mg/ha
1May 1Jun 1Jul 1Aug Harvest
Forecast Date
0.5
1.0
1.5
2.0
1May 1Jun 1Jul 1Aug Harvest
Forecast Date
1982 Queensland, Australia wheat yield forecast.
climatology only + GCM forecast
Forecast date
Grainyield(Mgha-1)
12. 12
Linking crop simulation models and
seasonal climate forecasts statistically
forecast
date
harvest
model
initialization
fittedstatisticalmodel
yn,1
yn,2
yn,3
.
.
.
yn,n-1
Time of year
1
2
.
.
.
n
}
Weatherdatayear
ˆky
13. 13
Versions:
• Windows 95+
• Linux batch
• Windows batch (for CRAFT)
Incorporating seasonal forecasts:
CPT
Climate Predictability Tool (CPT) is an easy-to-use software
package for making tailored seasonal climate forecasts.
14. 14
Why CPT?
Address problems that arose in RCOFs:
• Slow production made pre-forum workshops expensive
and prohibited monthly updates
• Multiplicity, colinearity, artificial skill, lack of rigorous
evaluation made forecasts questionable
• Little use of GCM predictions
(http://www.wmo.int/pages/prog/wcp/wcasp/clips/outlooks/climate_forecasts.html)
15. 15
What CPT does
• Statistical forecasting
• Statistical downscaling
coarse resolution
fine resolution
statistical model
dynamical model
16. 16
What CPT does
• Statistical forecasting
• Statistical downscaling
• Designed to use gridded data (GCM output and SSTs)
as predictors
• Uses principal components (PCs, or EOFs) as predictors
• Rigorous cross-validation to avoid artificial skill
• Diagnostics and evaluation
• New multi-model support
17. 17
CPT: Principal Components (PCs)
• Explain maximum amounts of variance within data
• Capture important patterns of variability over large areas
• Uncorrelated, which reduces regression parameter errors
• Few PCs need be retained, reducing dangers of “fishing”
• Corrects spatial biases
First PC of Oct-Dec 1950 -1999
sea-surface temperatures
18. 18
CPT: Canonical Correlation Analysis
(CCA)
July (top) and December (bottom)
tropical Pacific sea-surface
temperature anomaly, 1950-1999
December
July
22. 22
Steps: yield forecast run
• Step 1 – Prepare/Review Data Sets
• Step 2 – Create Project & Run
• Step 3 – Link Data Sources
• Step 4 – Enter Crop Management Data
• Step 5 – Setup & Execute Crop Model Run
• Step 6 – View Crop Model Run Results
• Step 7 – Seasonal Forecast Run
• Step 8 – View Forecast Yield Results
23. 23
Home Page
• Main Menu
• Connect
application to
desired database,
and test
• Lists 5 most recent
projects
• Project Name link
will direct the user
to the current state
of the workflow
24. 24
Data Upload
• CCAFS pre-loaded
with default data
(currently South
Asia).
• Users can upload
data: crop mask,
cultivar, fertilizer,
field history, planting,
irrigation mask &
management, soil.
• These data are input
data to DSSAT and
CPT engines during
run.
• Version control of
user-supplied data.
26. 26
Project
• Search or select
existing project,
or create new
project
• Navigate to data
source form for
active run of
active project
27. 27
Data Sources
• Customize run
by selecting
uploaded or
default data
sources
• Drop-down lists
of previously
uploaded data
• This screen is
not shown if
Default is
selected when
creating Project
and Run.
28. 28
Apply Inputs
• Tabbed details of
available Levels,
and a grid to show
the applied levels
for specified ‘Type
of Data’
• Green = input
applied,
• Pink = input not
applied
• Mandatory fields
must be applied
29. 29
Run Project
• Executes current
active project with
configured data
sources and applied
inputs
• On successful
execution, prompts
to save run results,
view result maps
• 2 steps:
• Crop simulation
• Seasonal forecast
31. 31
User interface summary
SPATIAL
DATA
MANAGEMENT
DATA
PROJECT &
RUN SETUP
RUN
PROJECT
RESULTS
Import
default data
sets – admin
only
Import
gridded user
data sets
Export default
data sets
Export
gridded user
data sets
Define Cultivar
Define Planting
Dates
Define Irrigation
Application
Define Fertilizer
Application
Define Field
History
Create a
project
Search and
Select Project
& Run
Create Run(s)
Identify data
sources
Apply UI based
inputs
Run crop
model
Run
Seasonal
Forecast
module
Run
Calibration
module
Single Project Run
• Select project
• Select outputs to
view
• View/Export
Results
Compare Project
Runs
• Select the two
projects
• Select outputs to
compare
• View/Export
Results
32. 32
Major planned enhancements
• Generalize locations, grid schemes, user inputs
• Crop model interoparability (AgMIP)
• Additional crop models:
• APSIM
• AquaCrop
• ORYZA2000
• SARA-H
• InfoCrop
• …
• Hindcast analysis and validation statistics
• De-trending and post-simulation calibration