An introduction to the crop
model GLAM
Sanai LI
Email: lisanai@apcc21.org
APEC Climate Center
12 Centum 7-ro, Haeundae-gu, Busan, 612-020, Republic of Korea
1
Introduction
Crop modelling methods
 Empirical and semi-empirical methods
+
+

Low input data requirement
Can be valid over large areas
May not be valid as climate, crop or management change
 Process-based
+


Simulates nonlinearities and interactions
Extensive calibration is often needed
skill is highest at plot-level
What is the appropriate level of complexity?

Depend on the yield-determining process on the spatial scale of
interest (Sinclair and Seligman, 2000)
General Large Area Model
for Annual Crops
(GLAM)
 Aims to combine:


Challinor et. al. (2004)
the benefits of more empirical approaches (low
input data requirements, validity over large spatial
scales) with
the benefits of the process-based approach (e.g.
the potential to capture intra-seasonal variability,
and so cope with changing climates)
 Yield Gap Parameter to account for the
impact of differing nutrient levels,
pests, diseases, non-optimal
management etc.
4
GLAM (General Large-Area Model for annual crops)
• Process based crop model
• Specifically designed for use on large spatial scales
- simulates climatic influences on crop growth and development
- low input data requirements
Typical climate model grid cell –
GLAM can be run on this spatial scale
GLAM – Inputs and outputs
INPUTS
Daily weather data:
OUTPUTS
- Rainfall
- Solar radiation
- Min temperature
- Max temperature

Soil water
balance
Soil type
\
GLAM
Planting date

Leaf canopy

Root growth

Biomass

Crop Yield
General Large Area Model for Annual
Crops (GLAM)
d(HI)/dt
Pod yield
Biomass
transpiration
efficiency
Root system
Development
stage
Transpiration
Leaf canopy
CYG
water
radiation
temperature
rainfall
RH
Soil water
stress
the flowchart of the GLAM model structure and the processes of yield and
biomass formation
General Large Area Model for Annu
al Crops (GLAM): some parameters
• Thermal duration: Determines development rate
o Predict weather extremes at sensitive stages (e.g. flowering)
• Transpiration efficiency to calculate biomass
o Yield changes under elevated CO2
• Maximum rate of change of LAI: determines growth of leaves
o Check model consistency by looking at Specific Leaf Area
• Yield gap parameter: time-independent site-specific parameter to
account for the impact of differing nutrient levels, pests, diseases,
non-optimal management etc.
o Process-based: acts on LAI to determine an effective LAI
o In practice, YGP can bias correct input weather data
o It is not the sole determinant of mean yield, however
Simulation of developmental stages
daily effective temperature (oC)
t TT 

ti 1
ti
(T eff  T b ) dt
i development stage
thermal time
(oCd)
Time (days)
base temperature (oC)
In GLAM, the simulation of developmental stages is controlled by
accumulated thermal time.
Once the thermal time accumulation tTT reaches the specified thermal time
for a given stage, next stage begins
Modelling crop growth:
biomass in GLAM
H2O CO2
Stomatal conductance
During the summer, cumulative
photosynthesis increases linearly with
cumulative transpiration
above ground biomass
transpiration efficiency
photosynthesis is not modeled directly, but
it is represented by transpiration efficiency
Maximum normalized transpiration efficiency
W
t
 ET

 T T min 
, E TN , max 
 V

daily actual transpiration
Vapor presser deficit
maximum transpiration efficiency
10
Modelling canopy in GLAM Leaf Area
Index
effective LAI
yield gap parameter
 S 
 L 
 L 
,1 
 C YG min 

  
  t  max
 t 
 S cr 
maximum LAI expansion rate
S 
TT
TTopt
the soil water stress factor
•Water stress factor reduces leaf
expansion
• decrease in leaf area index
affect radiation interception and
transpiration, and hence crop
yield
11
Modelling crop growth: Yield in GLAM
Yield=biomass*harvest index (HI)
Harvest index =yield/biomass
Biomass is allocated to yield by harvest index
from the beginning of gain-flilling, if there is no any stress, harvest index linearly
increase with time
12
Water Balance in GLAM model
Evaporation + Transpiration
Rain
net soil water input Sw=Trainfall-Ttranspiration-Tevaporation-Trunoff-Tdrainage
All of the non-runoff rain goes through the first soil layer first, then the total water
infiltration into the soil is distributed into NSL (No. of soil layers) vertical soil layers
13
Water Balance in GLAM
Runoff-US Soil Conservation Service method (USDASCS, 1964)
R is the runoff
P is the precipitation
S is the amount of water that can soak into the soil
S = ksat
ksat is saturated hydraulic conductivity of the soil
Kks is emperiacal constant
Infiltration rate=precipitation -runoff
All of the on-runoff rainfall goes through the first soil layer firstly. When the soil
water content is greater than the drainage limit, then the excess soil water is
infiltrated into the next soil layers and the soil water content in each layer is
simulated..
14
Water Balance in GLAM -drainage
Cd1, Cd2, Cd3 are empirical constants
θdul is drained upper limit
F accounting for simultaneous inflow from the layer above
Qi is the incoming water flux from the layer above
FD is the drainage rate
θ is soil water content
θs is the initial value of θ
If the soil moisture is greater than dul at the start of a timestep, the
incoming water from above is percolated to the lower layers
15
potential evapotranspiration rate-Priestley Taylor
RN - net all-wave radiation
G - soil heat flux, G = CGRNe−kL , CG constant, k-constant
λ - the latent heat of vaporisation of water
Δ = ∂esat/∂T - slope of the saturation-vapour pressure
versus the temperature curve
γ- ratio of the specific heat of air at constant
pressure to the latent heat of vaporisation of water
α -PriestleyTaylor coefficient, is a function of
VPD(vapour pressure deficit)
The energy-limited evaporation and transpiration rates (Ee and e , respectively) are
TT described using the simpler Priestley–Taylor equation (Priestly and Taylor, 1972)
16
Transpiration and evaporation
potential evapotranspiration rate
maximum possible energy-limited evapotranspiration
Is given by G=0
energy-limited
evaporation
energy-limited
transpiration
potentially extractable soil water
te(z) is the time of first root
uptake in layer z
kDIF is the uptake diffusion
coefficient
zmax is depth of soil profile
transpiration rate
evaporation rate
The available soil water is partitioned into transpiration and evaporation according to water demand where
necessary
17
Model calibration
GLAM Calibration-Yield gap parameter (YGP)
GLAM run with YGP, varying from 0.05-1 in step of 0.05. The optimal
value is chosen by minimizing the Root Mean Square Error (RMSR)
between observed yields and simulated yields
19
GLAM – Calibration
GLAM simulates the impact of weather on crop yields.
It does not directly simulate the impact of other factors such as nutrient
deficiencies, pests, diseases, weeds
The yield gap parameter is a time-independent site-specific parameter that
accounts for these factors.
Crop yield
It also acts to bias correct weather
1.00
Yield Gap Parameter = 0.80
0.05
GLAM – The Yield Gap Parameter (YGP) methods
You can choose how the yield gap parameter reduces simulated yields.
Options include acting on:
• EOS: end-of-season yield
•LAI: Leaf area index
• ASW: the available soil water
Roots
YGP = 0.2
YGP=1.0
Soil
Properties
Uptake of
water
YGP = 0.2 YGP =
1.0
YGP = 1.0
YGP = 0.2
Leaf Area
Index
Potential
rate of
transpiratio
n
Rate of
transpiratio
n
Soil Water
Stress
Factor
Biomass
Crop
Yield
Harvest
Index
Simulating the
floods effect on
wheat
Flooding effect
 There is increased risk of crop losses due to
flooding and excess precipitation
 crop damage from flooding and excess soil
moisture is not included or not well simulated
by some dynamic crop models
 GLAM- New schedule of surface water storage,
infiltration and waterlogging
23
Simulating the impact of flood on wheat in China
Correlation coefficient between observed wheat yield and rainfall in China
from 1985 to 2000
24
Water Balance in GLAM
-Surface water storage and runoff
PPTi
ETi
Surface storage
Runoff
Drainage
More frequent heavy rainfall
may increase surface water
storage and cause crop loss
due to excess soil moisture
Infiltration
 method : the infiltration capacity of the
soil is assumed to be affected the soil
water content
INF is infiltration rate (cm/day)
P is precipitation (cm/day)
SURFSTORAGE is surface water storage
SWsat is saturated volume soil water
SW is volume soil water content in the soil layer
SWdul is drained upper limit
DZ is the depth of soil layer
26
The response of transpiration to
waterlogging days for winter wheat
in China (Hu et al,2004)
•Waterlogging can result in the death of root cells, due to a reduction in oxygen
availability.
•Excessive soil moisture limits root growth and absorption of soil water,
consequently decreasing crop transpiration
Parameterizing the
flood effect
Method (Hu et al,2004): simulate flood effect by introducing a damage
function that limited the plant's transpiration and roots growth roots
when soil is greater than the field capacity
WSF is water stress factor
WSFC0 is the sensitivity of different crops to waterlogging
f(TW; PDT) is the response of transpiration to waterlogging days from empirical function of
experimental data
Kwl is the ratio of the lower limit of soil water content under waterlogging stress to field capacity
SW is soil water content
SWFC is field compacity
SWSAT is saturated soil water content
Comparison of soil water content in the first soil layer with and without
surface water storage in water balance model of GLAM
with surface water storage the simulated soil water content is slightly higher
than that without surface storage
Probability distribution function of correlation coefficient between observed
and simulated wheat yield in east China
Blue line : infiltration is calculated from original GLAM model without
flood
Green line: infiltration is calculated from original GLAM model with flood
Red line: infiltration is simulated to by the modified infiltration with flood.
the original model with waterlogging stress is better then without waterlogging stress. The modified
GLAM model was better than the original model
30
Simulating the flood effect
Comparison between observed yield and simulated yield with and without
flooding effect from 1985 to 2000 at 0.5◦ grid cell (31.75◦N; 120.25◦E)
the modified model improved yield predictions in years with serious flooding
damage year 1991 and 1998
Comparison of correlation coefficient between observed
and simulated yield at the 0.5◦ scale in east China from
1985 to 2000
With flooding effect, yield predictions showed a better agreement with
observed yield compared with no flooding effect
original
modified models
Model
performance
Evaluation of model consistency-RUE
Radiation Use Efficiency (RUE=biomass/radiation intercepted) of winter
wheat :1.58 g MJ-1, spring wheat: 1.34 g MJ-1
Measured RUE of 1.81 g MJ-1 in semi-arid environment by O’Connell et al.(2004)
34
Correlation between observed and simulated
yield at county(70-129km)/city(80-128km) level and
field level in China
0.8
Correlation(R)
0.7
0.6
Rainfedcounty(city)
Rainfed-field
0.5
0.4
0.3
0.2
Significant level
0.1
0
Guyuan
Guyang
Spring wheat
Huma
Zhengzhou Beijing
Winter wheat
35
Comparison of simulated and observed wheat
yield (kg/ha) at 0.5o scale across China
(a) Observations
(b) Simulations
36
Validation of GLAM-Wheat in China
-Difference between observed and simulated mean wheat yield
(%) in China (correlation r= 0.83,p<0.001)
37
Model skill for regional aggregated yield
Sensitivity to spatial scale of weather data ( Li and Tompkins, 2012)
In order to find the appropriate spatial scale of
climate data to run the regional crop model, the
GLAM model is run with the 0.5◦×0.5◦ aggregated
1◦×1◦ and 2◦×2◦ scale weather data respectively
GLAM model driven by the 0.5 resolution of
climate data gave a better fit of simulated yield
in the rainfed agriculture dominant regions
38
Importance of subseasonal temporal variability in
rainfall
Using 5 day pentad averages of rainfall, simulated yield is very similar to that
using the original daily dataset for the selected sites in China.
Driving GLAM with 10 and 30 -day average rainfall resulted in a larges
difference as using the original daily dataset
39
Importance of subseasonal temporal
variability in temperature
The use of averages temperature over 5 days and 10 days in GLAM
has a minor impact compared with simulations with daily temperature
at all sites. Only the use of a 30 day running mean had great impact
on the yield
40
Impacts of extremes –temperature
Grain-set fraction
short periods of exposure to high daily maximum temperatures can
also exert a dramatic impact on crop yield
1.2
1
0.8
0.6
0.4
Observations
GLAM-Spring
GLAM-winter
0.2
0
24
26
28
30
32
34
36
Maximum temprature(oC)
38
Temperature at anthesis ~ time of
flowering
Also observed for rice (above)
and groundnut
Summary of observed and modeled increase
in wheat yield in response to elevated CO2
W
t
 TT m in(
ET
V
, E TN ,m ax )
Increase
in CO2
ppm
Increasee in
yield (%)
Methods
Source
330-660
37
Glasshouse or growth Kimball, 1983
chambers
350-700
31
Estimated by cubic
equation from
multiple experiments
Amthor, 2000
350-700
28
Linear extrapolation
of FACE experiment
Easterling et
al., 2005
370-550
7 -23
FACE experiment
Kimball,2002
330-660
25
CERES for C3 crops
Boote,1994
350-700
16-30
GLAM model in
China
Case studies for
lab class
GLAM lab class
First download GLAM
www.see.leeds.ac.uk/research/icas/climate_change/glam/glam.html
GLAM lab class
Then choose case study:
• Ghana
• China
For details see
https://www.see.leeds.ac.uk/redmine/public/projects/glam/wiki
(unique username and password needed)
And decide whether to use a text edit or the beta version GUI
See https://www.see.leeds.ac.uk/redmine/public/projects/glam/wiki/blabla
Step 1: Decide on the grid cells/regions GLAM will be run for.
Depends on:
• Available input data
• Scale of relationship between weather and
yield
• Aim of the project
Thailand– GLAM will be run on 0.5°x 0.5°
gridcells
Step 2: Collect and organise input data
Daily weather data
Rainfall, min and max temperature, solar radiation.
thailand- PRECIS output for 1981-2012, 2041-2050
Step 3: Collect and organise input data
Soil data
Soil hydrological properties (can be found from soil texture):
R. Evans et al 1996
Saturation limit:
maximum amount of
water in the soil.
Drained upper limit:
water held after thorough
wetting and drainage
Lower limit:
any remaining water
can not be extracted.
Thailand– Soil texture information from FAO soil map of the world
– Data averaged onto model grid
Step 4: Collect and organise input data
Planting date information
Specify planting date or start of ‘intelligent sowing window’
Observed yield data
Crop yield (kg/ha) = Production (kg)
Cultivated area (ha)
800
Original
600
500
400
300
200
100
Year
2005
2003
2001
1999
1997
1995
1993
1991
1989
1987
0
1985
Thailand– provincial yield data is converted to
grid cells by ArcGIS
– Remove ‘technology trend’
Crop yield (kg/ha)
700
Step 5: Check parameter values are appropriate for local cultivars.
Step 6: Run GLAM in ‘calibration mode’ to find the yield gap parameter (YGP) for
each grid cell.
Crop yield
Yield Gap Parameter = 1.00
Yield Gap Parameter = 0.80
Yield Gap Parameter = 0.05
Yield (kg/ha)
Step 7: Run GLAM using these YGPs – compare simulated yields to observed
yields.
1000
900
800
700
600
500
400
300
200
100
0
Observed
Yields
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Year
Conclusions
 The GLAM-wheat model was able to capture a large fraction
of interannual variability in observed yields in the rain-fed
agricultural regions
 Spatially averaging the driving climate data show that the
use of the highest spatial resolution maximized model skill
at reproducing yield
 5 day (pentad) averaged rainfall can be used to drive crop
model, however 10 day or longer averaging rainfall strongly
impacts crop yield
 temperature, a smoother field, could be averaged for 10
days or even monthly for crop modeling
51
Conclusions
 There is an increase in frequency of heavy precipitation
events in many parts of the world. It often cause large
agriculture losses and great economic costs
 By including the waterlogging damage and surface water
storage, the modified GLAM-wheat model was able to
capture most serious flooding damage to wheat yield in
China
 It is critical to consider the yield loss due to waterlogging
when crop model is used for crop forecasting and climate
impacts studies in region with high flooding probability
52
Thank you
for your
attentation
53