Prsentation PowerPoint

1
AGCM simulations of subseasonal characteristics
of rainfall and implications for crop yield
prediction in Central West Africa
Sylwia Trzaska, Liqiang Sun, Pauline Dibi-Kangah,
Andrew Robertson IRI The International
Research Institute for Climate and Society The
Earth Institute at Columbia University, Palisades,
NY 100964-8000
2
What does really matter?
high resolution well informed input e.g. Niamey
supersite Specific pond Research plot
AMMA
Application (Agric.,Water Ress. Mangt., Health,
Fisheries etc.)
Large scale clim info (obs, GCM)
dynamical
dynamical
Climatic constraint e.g. daily rainfall
model
downscaling
statistical
statistical
?
Subseasonal PDF rainfall amount /Dry spell
length Generation of Sequences (weather
generators, HMM)
Larger scale scarce input e.g. station data for a
district
Represantativity of the data used for training
3
Example

• CERES-Maize model to simulate maize yields
  10locations in summer cropping season
  March-August for 74  years (19231996)(one of
  the crop models in DSSATv3.5)  
• Disaggregation from daily regional averages

Robertson et al. subm
4
Sorghum Yields in Central and Northern Burkina
Faso

• DATA
• Daily rainfall in 14 stations 1931-1998
• (monthly up to 2004)
• Annual yield for staple and cash crops in 10
  districts 1984-2004
• Monthly gridded global SST 1951-present
• NCEP Reanalysis 1951-present
• ECHAM4.5 historical runs 1951-present
• ECHAM4.5 forecats (persisted SST)
• 1951-1996

5
Sorghum Yields and intraseasonal characteristics
of observed rainfall

• Detrended yields at district level vs
• seasonal totals,
• number of rainy days,
• number of short (1-3 d) dry spells,
• number of long (gt5 or 10d) dry spells,
• mean dry spell length,
• drought index (assoc short and long dry spells)
• etc
• at each station

?Linear relationships poor (high scatter) and
highly diverse e.g Fada Ngourma - seasonal
totals, number of rainy days, mean dry spell
length Dori long dry spells (gt10d)
drought index Ouahigouya number of rainy
days short spells but NOT seasonal totals
6
Index NC Burkina Faso
Consistency of daily rainfall variability in
stations N of 12N ? Regional index based on 7
stations
Regional occurrence index rainy (dry) day if
rainy (dry) day in 5 stations/7
7
Index NC Burkina Faso
r0.75
8
Statistical forecast of aggregated yield using
drought index
Linear regression, cross validated -1
Yield (std)
year
Using more predictors does not significantly
improve the forecast
9
Subseasonal characteristics vs global SST
Correlations JAS 1984-98
10
Yield vs global SST
11
Yield vs JAS ECHAM 4.5 (obs SST) total rainfall
JAS
12
Correlations subseasonal characteristics and
yield vs ECHAM MSLP JAS
13
Correlations subseasonal characteristics and
yield vs ECHAM U850hPa JAS
14
Correlations subseasonal characteristics and
yield vs ECHAM V850hPa JAS
15
Correlations sorghum yield vs ECHAM U200hPa
JAS
Sorghum Yield vs JAS ECHAM 4.5 (obs SST) U200
16
Summary

• What is needed?
• Evidence of climate information from historical
  data
• (not always the major limiting factor)
• Adequate area and scale
• Importance of subseasonal characteristics vs raw
  daily rainfall
• Look for GCM predictors ?empirically defined?
• Need for multiple applications/subregions
• in order to identify robust predictors