26 May 2015
Towards the simulation of global crop productivity in
IMPACT with geospatial data
Michael MarshallWorld Agroforestry CentreUnited Nations Ave Gigiri
PO Box 30677-00100Nairobi Kenya
+254207224244mmarshallcgiarorg
Global Crop Model Approaches
Local-scale models (eg DSSAT and APSIM) aggregated via multiple-site calibration and averagingminus Complexminus High-input data requirementminus Locally accurateminus Site specificity (over-fitting)minus Non-linearity
Large-area models (eg GLOPEM and GLAM)minus Simpleminus Low-input data requirementminus Locally less accurateminus Generalizable
ldquoEssentially all models are wrong but some are usefulrdquo
Global crop model for IMPACT
Opti-LU
Opti-WU
εmax=006 molmol-1 (C4)
Model Calibration Sensitivity Analysis and Validation
Identify most important parameters remove redundant or insignificant parameters and potentially integrate new parametersminus Model performance statisticsminus Monte Carlo Simulationminus Residual analysis
Multiple scales of validationminus Plot (eddy covariance flux tower data)minus Field (non-destructive biomass transects)minus Landscape rarr global (high moderate and
coarse resolution remote sensing)
Eddy Covariance Flux Towers
Three towers (2009-2014) Alfalfa rice and
citrus orchard 30-minute energy
balance and meteorological data
FAPAR (MODIS subset tool httpdaacornlgov)
Daily GPP RECO and NPP
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Global Crop Model Approaches
Local-scale models (eg DSSAT and APSIM) aggregated via multiple-site calibration and averagingminus Complexminus High-input data requirementminus Locally accurateminus Site specificity (over-fitting)minus Non-linearity
Large-area models (eg GLOPEM and GLAM)minus Simpleminus Low-input data requirementminus Locally less accurateminus Generalizable
ldquoEssentially all models are wrong but some are usefulrdquo
Global crop model for IMPACT
Opti-LU
Opti-WU
εmax=006 molmol-1 (C4)
Model Calibration Sensitivity Analysis and Validation
Identify most important parameters remove redundant or insignificant parameters and potentially integrate new parametersminus Model performance statisticsminus Monte Carlo Simulationminus Residual analysis
Multiple scales of validationminus Plot (eddy covariance flux tower data)minus Field (non-destructive biomass transects)minus Landscape rarr global (high moderate and
coarse resolution remote sensing)
Eddy Covariance Flux Towers
Three towers (2009-2014) Alfalfa rice and
citrus orchard 30-minute energy
balance and meteorological data
FAPAR (MODIS subset tool httpdaacornlgov)
Daily GPP RECO and NPP
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Global crop model for IMPACT
Opti-LU
Opti-WU
εmax=006 molmol-1 (C4)
Model Calibration Sensitivity Analysis and Validation
Identify most important parameters remove redundant or insignificant parameters and potentially integrate new parametersminus Model performance statisticsminus Monte Carlo Simulationminus Residual analysis
Multiple scales of validationminus Plot (eddy covariance flux tower data)minus Field (non-destructive biomass transects)minus Landscape rarr global (high moderate and
coarse resolution remote sensing)
Eddy Covariance Flux Towers
Three towers (2009-2014) Alfalfa rice and
citrus orchard 30-minute energy
balance and meteorological data
FAPAR (MODIS subset tool httpdaacornlgov)
Daily GPP RECO and NPP
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Model Calibration Sensitivity Analysis and Validation
Identify most important parameters remove redundant or insignificant parameters and potentially integrate new parametersminus Model performance statisticsminus Monte Carlo Simulationminus Residual analysis
Multiple scales of validationminus Plot (eddy covariance flux tower data)minus Field (non-destructive biomass transects)minus Landscape rarr global (high moderate and
coarse resolution remote sensing)
Eddy Covariance Flux Towers
Three towers (2009-2014) Alfalfa rice and
citrus orchard 30-minute energy
balance and meteorological data
FAPAR (MODIS subset tool httpdaacornlgov)
Daily GPP RECO and NPP
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Eddy Covariance Flux Towers
Three towers (2009-2014) Alfalfa rice and
citrus orchard 30-minute energy
balance and meteorological data
FAPAR (MODIS subset tool httpdaacornlgov)
Daily GPP RECO and NPP
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Transects
60 m
10 quadrats per frame (2011-2012) Alfalfa cotton
maize and rice
Biophysical data
Ground spectra
Destructive aboveground wet biomass
Empirical (non-destructive) model-building
CIMIS
1
2
3
5
4
6
10
7
8
9
Marshall and Thenkabail (2015)
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Tasseled Cap Transformation
Weighted linear combinations of satellite reflectanceminus Brightnessminus Greennessminus Wetness
Fraction of vegetation bare soil and water
Density
Kauth and Thomas (1976)
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Index of Agreement (yellow = perfect match)
Biomass
Dark Soil
Light Soil
PlantsPlants
Reference Data
AVHRR (5 km) MODIS (500 m) Landsat (30 m) IKONOS (4 m) WorldView (185 m) and GeoEye
(165 m)
Tasseled Cap
Biomass Estimation for Validation at Remote Sensing Scales
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Muchow (1990)
GPP rarr Crop Yield
Prince et al (2001) expressed yield in terms of GPP
Where SOS is start of season EOS is end of season RS is the root-to-shoot ratio HI is the harvest index and m0 is grain moisture content
RS HI and m0 are constant by crop type
(Xin et al 2013)
OR
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Need for systems simulation models in Agroforestry Systems
Lloyd et al (1990)
Agroforestry systems large payoffs but complex
Systems simulation minus Can handle
complex systemsminus Provide targeted
and effective intervention
minus Tunable for ldquofuturerdquo climates
Not widely used in SSA because of data scarcity
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
GPP and crop yield model validated at multiple spatial scales in California
minus Thomas Gumbricht MODIS-era GPP modeling tool for California
minus Patricia Masikate Agro-forestry module (APSIM) and estimates of RS HI and m0
minus Erick Okuto SOS and EOS simulation
Next year (2016) Global GPP and crop yield model development calibration and validation
minus 5 km resolution 1982 ndash 2014 (+30 years)
Summary
Thank You
Thank You