Scaling up Crop Model
Simulations to Districts for Use in
Integrated Assessments:
Case Study of Anantapur District in
India
K. J. Boote, Univ. of Florida
Co-Coordinator, Crop Modeling
Activities – Crop Modeling Team

Activity 3 – Calibrate multiple models at field and regional
scale, accounting for regional weather, soils, cultivars, &
management. Use published experiments, variety trials, &
historical regional yields within regions. TWO STEPS!!!
1.
2.



Calibrate cultivars: Site-specific experiments with known
soils and management (time-series data, Platinum) (end-ofseason data – variety trials, etc., Silver)
Bias-Adjustment for Regional district-level yields, which
lack site-specific information. Upscale to predict regional
yields, accounting for bias and variability associated with
lack of knowledge on soils, irrigation, sowing date, sowing
density, fertilization, cultivar, pests, & technology.
Regional Calibration Teams (soil, crop, climate experts)
Activity 4 – Predict impact of baseline & climate change
scenarios on agricultural production for regions, with climate
team.
Activity 5 – Evaluate strategies of genetic improvement and
management (sowing date, fertilization, irrigation, etc.) for
adaptation to climate change.
Two relevant scales and appropriate
methodologies
Crop model calibration
against site-specific
experimental data
sets.
But is this
representative of
region?
Regional yield estimates
must account for
uncertain distribution of
weather, soils, cultivars,
sowing dates, fertility for
region
From the point
to the region
Case Study: Scaling up Crop Model Simulations
for Anantapur District of India
Data Available: Lacked “on-farm” surveys. Had one
sentinel experiment station site. 28 years of aggregated
groundnut yield for Anantapur District from 1980 to 2007.

Activity 3 (point to region) – Predict district-level peanut pod
yields for Anantapur, accounting for weather sites, soils,
cultivars, sowing date, & management of the region.




FIRST, calibrate cultivar life cycle and yield traits for TMV-2
cultivar with site-specific studies with known soils (measured
neutron probe) and management (6 years of time-series and
end-of-season data, Platinum/Silver sites).
SECOND, simulate district-level yields over 28 years, using 3
sowing dates (auto-plant), 3 representative soils, and 9 weather
sites. Gives n=81. Compute simulated mean yield per year.
Plot observed district-level yields (per year) versus simulated
mean annual yields. Compute bias (ratio or slope with zero
intercept). Plot bias-adjusted yields and observed yields over
historical time. Evaluate deviations from observed.
Use calibrated model for climate impact assessments.
Case Study: Scaling up Crop Model Simulations
for Anantapur District of India

Activity 4 – Predict impact of baseline & climate change
scenarios on yield for Anantapur region, with climate team.




Create uncertainty distributions – weather, mgt, soils
Interpret results.
Link outputs to economic teams
Activity 5 – Evaluate strategies of genetic improvement and
management for adaptation to climate change. DO
ADAPTATION EVALUATION FOR BASELINE AND CLIMATE
SCENARIOS!!!


What sowing dates, residue management, irrigation, and fertility
management are best for adaptation?
Can genotypes can be modified to improve productivity under
climate change?
1.4
1.2
1
0.8
Series1
0.6
0.4
0.2
0
1960
1970
1980
1990
2000
2010
Anantapur district peanut yields (metric ton/ha) over historical time.
Used only 1980 to 2007. De-trend? No trend from 1980 to 2007.
Calibrating Cultivars for Anantapur Exp. Sta
(Multi-Model: DSSAT, APSIM, INFOCROP)
Best results if no N and water deficit stresses
• Estimate life cycle-dependent traits first - Most Important
– Thermal time to anthesis and to maturity.
• Estimate growth, partitioning, and yield traits next.
– 1. Is final biomass correctly predicted? Is SOC correct?
Initial NO3 and NH4? If site has high N fertilization and
model over-predicts, then reduce RUE, or reduce SLPF for
unknown P, pH, micronutrient deficiencies.
– 2. Grain yield. Set grain size first. Then grain number. Is
HI correct? APSIM: Set rate of HI-increase
• Time-series data: dry weights & leaf area are helpful.
• Use your knowledge. No blind statistical methods.
Experimental Data for Anantapur Site
Year
Datasets
1986
1987
1988
1989
1990
1993
12
4
4
6
6
4
Treatments on TMV-2 Cultivar
Sowing date X Irr X Plant density
Sowing date X irrigation
Same
Same
Same
Same
Total 36 data sets/treatments
Simulated Total Biomass and Pod over Time in 1987 at
Anantapur after calibration of DSSAT-Groundnut model.
6000
5000
4000
Sim. Tops wt (kg/ha)
Sim. Pod wt kg/ha
Obs. Tops wt kg/ha
Obs. Pod wt kg/ha
3000
2000
1000
0
20
40
60
Days after Planting
80
100
Comparison of observed versus simulated pod yield at
Anantapur after calibration of DSSAT-Groundnut model.
3500
y = 1.0242x - 39.692
R2 = 0.6881
Observed Pod wt (kg /ha)
3000
2500
2000
1500
1000
500
0
0
500
1000
1500
2000
Simulated Pod wt (kg/ha)
2500
3000
Anantapur – Groundnut-1987
Info Crop
7000
Dry matter (kg/ha)
simulated TDM
6000
Observed TDM
5000
Simulated Pod
Yield
4000
3000
2000
1000
0
0
20
40
60
Days after sowing
80
100
120
Groundnut pod yield (1987-89)-InfoCrop
APSIM-Groundnut
• Calibration completed for TMV 2 cultivar
– Anthesis and Maturity Phenological stages
– Pod Yield
– Total Dry matter
– Harvest Index (HI)
Anthesis
DAS
Mean
Median
Maturity Pod Yield Total Dry
DAS
(kg/ha)
Matter
HI
Sim
29
99
1698
3997
0.29
Obs
28
94
1654
3773
0.29
Sim
27
97
1709
3967
0.27
Obs
27
93
1663
3722
0.31
Comparison of observed district yields versus DSSATsimulated pod yield (aggregated over 9 weather sites,
3 soils, & 3 sowing dates). Slope is bias-adjustment.
-1
Observed dist yield (kg ha )
1600
1400
1200
1000
800
600
400
y = 0.6952x
2
R = 0.1778
200
0
0
500
1000
1500
2000
-1
Simulated mean yield (kg ha )
2500
DSSAT-simulated & District yield, unadjusted
Simulated mean pod yield (n=81)
Observed district yield
2000
1500
1000
500
Year
2006
2004
2002
2000
1998
1996
1994
1992
1990
1988
1986
1984
1982
0
1980
Pod yield (kg ha-1)
2500
Observed historical district yields versus DSSAT-simulated
yield (after bias-adjustment and aggregation) at Anantapur.
1600
1200
1000
800
600
400
200
0
19
80
19
82
19
84
19
86
19
88
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
Pod yield (kg ha-1)
1400
Adjusted simulated mean yield (n=81)
Observed district mean yield
Year
3000
Observed and simulated district-level peanut pod yield for
Anantapur from 1980 to 2007, APSIM Model – bias unadjusted.
2500
Observed Yields
Simulated Yields
Pod Yield, kg/ha
2000
1500
1000
500
0
1978
1983
1988
1993
1998
Year of Production
2003
2008
2000
Observed and simulated district-level peanut pod yield for
Anantapur from 1980 to 2007, APSIM model, after bias adjustment
Observed Yields
Pod Yield, kg/ha
1500
Simulated Yields
1000
500
0
1978
1983
1988
1993
1998
Year of Production
2003
2008
Summary: Two-Step Process for Scaling up Crop
Model Simulations for Regions if no Survey Data
What Data is Available: Do you have “on-farm” surveys?
How many sentinel-site experiments? Are they
representative of Region? Do you have district yield over
historical time? Can you describe the range of distribution
of weather, soils, sowing dates, fertilization inputs needed
for simulating aggregated yield for the region?
1.
2.
Sentinel Site Experimental Data – Calibrate thermal times for
cultivars (time to anthesis and maturity) & partitioning/yield traits
Predict District-level Yields and do Bias-adjustment
–
Collect district-level historical yields and de-trend.
–
Determine range of distribution of soils, weather stations, sowing
dates, fertilization, soil organic carbon for the region
–
Simulate district-level yields over the range of distributed inputs and
compute simulated mean yield per year.
–
Aggregate and plot observed district-level yields (per year) versus
simulated mean annual yields. Compute bias (ratio or slope with
zero intercept).
Summary: Process for Scaling up Crop Model
Simulations for Regions, Linking to Economics,
and Adaptation
3.
4.
–
–
5.
–
Impact assessment: Simulate with baseline and climate
scenarios, using distribution of weather sites, soils, and
management inputs.
Crop model outputs to economic models to simulate for same
regions, with management inputs and economic cost inputs.
Yield variability (probability) caused by soils & management
variability, seasonal weather variability.
Economic variability includes additional aspects from economic
inputs.
Adaptation (do it for baseline and climate change scenarios).
What management inputs, genetic improvement, government policy
interventions can be used to improve food security now, as well as
under future climate.