Transcript Walker Institute

Mechanistic crop modelling and
climate reanalysis
Tom Osborne
Crops and Climate Group
Depts. of Meteorology & Agriculture
University of Reading
Crop simulation models
Mechanistic
“
Process-based
“
NOT
Statistical
Dynamic evolution (daily time-step) of
 water balance,
 crop development,
 crop growth.
Relational diagram (Penning de Vries et al., 1989)
Atmospheric fields required:
• Rainfall
• Temperature (max, min, mean)
• Humidity (VPD, RH)
• Light (solar radiation)
Spatial scale
• Most crop models initially developed at field scale.
Use experimental station met data.
• GCM output downscaled to provide local met data.
• Few models developed for operation at larger
spatial scales
General Large Area Model for annual crops
Relationship between yield
and rainfall variability
observed at large spatial
scales.
Implies a potential to model
at such scales, leading to
the development of GLAM.
Relatively simple model
due to problems
associated with
parameter estimation
Correlation coefficient for detrended
groundnut yield and rainfall in India.
Temporal resolution
• Crop simulation models require daily data
• Interpolated/downscaled from monthly means
• Sub-seasonal variability important
- timing of events important e.g. flowering
Fruit set (%)
60
40
20
0
24
28
32
36
40
44
48
Flower bud temperature (oC)
Groundnut crop in Andhra Pradesh, India
1975
Total rainfall: 394mm
Yield = 1360 kg/ha
1981
Total rainfall 389mm
Yield = 901 kg/ha
Crop model able to capture difference between years
Use of reanalysis by crop simulation models
Extremely limited use of reanalysis for crop modelling
 CRUTS2.1 - monthly
0.5° global, 1901-2002 [pre, tmp, tmx, tmn, dtr, vap, cld, wet, frs]
 station data and/or downscaling commonly used
Sole example
 Challinor et al (2005) Simulation of Crop Yields Using ERA-40: Limits to
Skill and Nonstationarity in Weather-Yield Relationships J. Applied Met.
44,516-531.
Challinor et al (2005)
Indian summer monsoon rainfall
Despite a general over-prediction of tropical precipitation
in ERA-40, JJAS rainfall tends to be lower than
observations.
Fractional difference between ERA-40 and IITM data in (a) mean and (b) standard deviation
of JJAS precipitation. (c) The correlation in JJAS precipitation between the two datasets.
Challinor et al (2005)
Indian summer monsoon rainfall
• ERA-40: too many light rain-days, too few heavy rain-days.
• Skill in simulating onset varies
Fractional difference between ERA-40 and IITM in the mean (1966-89) number of
JJAS days with precipitation in the ranges shown. (c) The difference in days (ERA-40
-IITM) between the mean monsoon onset over the period of 1966-89.
Challinor et al (2005)
Seasonality
The ERA-40 and IITM seasonal cycle of precipitation for two grid cells in India.
Bars show the 1966-89 mean, and whiskers show one standard deviation.
Challinor et al (2005)
Crop simulations
Best performance in NW India
where climate signal is strongest
Comparison between simulated (control run) and observed yields for the period
of 1966-89: ratio of simulated to observed (a) mean and (b) standard deviation,
and (c) correlation between simulated and observed yields.
Challinor et al (2005)
Crop simulations
Mean bias correction
Interannual bias correction
Fractional changes in the rmse in yield, from the control run
baseline, for (a) bias correction B1, (b) bias correction B2,
Indication of potential importance of climate
model improvement for crop applications
Conclusion
 Limited use of reanalysis data as driving data for mechanistic
crop models.
 Reanalysis can fill the gaps where observational data is
sparse.
- Interannual variability (climatology in CRUTS)
- Finer temporal resolution
 When used, can provide an indication of maximum skill
attainable from GCM-crop model forecasting system.
- Biases in reanalysis limits skill of crop forecasts
- Indicates benefit of climate model improvement
Thanks for your time
E-mail: [email protected]
Visit www.walker-institute.ac.uk