toward next generation crop models...toward next generation crop models james w. jones crop modeling...
TRANSCRIPT
Toward Next Generation Crop
Models James W. Jones
Crop Modeling for Agriculture and Food Security
International Crop Modeling Symposium
Berlin, Germany
15-17 March 2016
Source: Monica Ozores-Hampton
Comments on history of crop modeling
• What fueled/limited progress?
• What are we learning?
Example Efforts Contributing to Next Generation
Models
• Model Intercomparison Projects (MIPs)
• Data, big data
• Genetics, gene-based modeling
What are we learning?
Laying the foundation for next generation
models
Outline
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Early innovators/innovations, visionaries
Fueled by technological development
Funding inconsistencies
Not fully embraced by traditional agronomic
researchers
Families or “tribes” of modelers
Long period of limited progress, stagnation
…
Nature Climate Change paper by Rotter et al. 2011.
titled “Crop-Climate Models Need an Overhaul”
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
History - Characterized by:
Modeling linkage between transpiration and growth (C. T. de
Wit, Wageningen University, 1958.
Modeling evapotranspiration & soil water dynamics (H.
Penman, J. Monteith, J. T. Ritchie, 1972, others)
Photosynthesis and growth (R. S. Loomis, W. G. Duncan, F.
Penning de Vries, de Wit et al.*)
Phenological development (J. Hesketh, H. Nix, J. Ritchie)
Soil nutrient dynamics and uptake (H. van Keulen, D.
Godwin, W. J. Parton)
Integrated models (SUCROS, CERES models, EPIC, AFRC-
Wheat, DSSAT-CSM, CENTURY, APSIM, STICS, …)
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Early Innovators / Innovations
*Bouman et al., 1996. (C. T. de Wit, H. van Keulen, F. Penning de Vries, R. Rabbinge, J. Goudriaan, M. Kropft, M. van Ittersum, etc.)
My first paper, 1972, Jones, Hesketh, Kamprath. N balance for crop models: A first approximation. Crop Science.
Mathematical modeling in other fields
Development of computers, enabling solutions
to nonlinear dynamic system models
Development & proliferation of personal
computers
Remote sensing, satellite technologies
Internet, communication technologies, www
{Molecular genetics technologies}
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Development Fueled by Technological Progress
Incremental progress to new plateaus funded by immediate
crises or initiatives • Unexpected large Soviet Union wheat purchase in 1972
• US Soil & Water Conservation act (1980)
• Overuse of pesticides; funding for integrated pest management (e.g.,
national CIPM project in US)
• USAID-funded IBSNAT project to develop systems approaches for
technology transfer (1983-93); DSSAT foundation continues now
• Systems Analysis and Simulation for Rice Production (SARP), 1984
• Climate change assessment of impacts and adaptation, starting in late
1980s (IPCC assessments)
• Australia support of APSIM, starting in 1991 continuing now
• EU research funding for policy support (SEAMLESS project, 2005-9)
• AgMIP and MACSUR crop modeling initiatives; starting in 2010
But, slow, incomplete embracing of crop modeling by
“mainstream” agricultural researchers/experimentalists
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
… also by Funding & Institutional Initiatives
Uncertainty, understanding and reducing it
Open, discoverable, accessible, and usable
data
Incorporating modern genetic information in
crop models
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Brief Overview of Three AgMIP Efforts
Uncertainty, understanding and reducing it
Open, discoverable, accessible, and usable
data
Incorporating modern genetic information in
crop models
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Brief Overview of Three AgMIP Efforts
Asseng, Ewert et al. 2013.
Nature Climate Change
Ensemble of models
predicted yields
accurately even if
uncalibrated (given
only phenology info.)
No individual model
predicted all sites
accurately.
Crop models are more uncertain than most would
have thought. Why? • Model structure, functions vary and contribute considerably
• Parameter definitions and values vary, difficult to harmonize
• A number of high impact journal articles are showing new ways of
quantifying uncertainties due to model structure, parameters,
Inputs, measurement errors (e.g., Wallach et al. 2016. Estimating
model prediction error … Submitted).
Most crop models have been developed using a limited
range of conditions, e.g., for weather & soils in region
where model was developed
More extensive datasets are needed to develop and test
models for more reliable simulations
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
What Are We Learning?
Uncertainty, understanding and reducing it
Open, discoverable, accessible, and usable
data
Incorporating modern genetic information in
crop models
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Brief Overview of Three AgMIP Efforts
Data: the Foundation of the Knowledge Chain
12
Modeling system knowledge chain. Infrastructure includes technical, institutional,
and organizational aspects (adapted from Lokers & Janssen 2014).
13
There is a Data Gap
• More extensive data are necessary to achieve robust models
• Major gap between potential value of data collected in
research and the value currently obtained through their use
• Typically, data collected in agronomic experiments are used
for the original research purpose only
• Vastly greater value can be obtained if data were combined
across locations, time, and management conditions
• Data are needed to provide the science base for next
generation models of agricultural systems and decision
support systems
• US federally-funded projects are required to make data
openly available; no existing process
• NARDN objectives:
• Create distributed network for harmonized crop, livestock data
• Devise common metadata for those systems
• Develop tools for discovering, accessing, and using the data
• Develop tools, procedures for researchers to contribute data
• Develop plan for long-term network operation
National Agricultural Research Data Network, NARDN
Characteristics of NARDN Project
15
• Emphasis on core sets of data defined by researchers;
main portal at the US National Agricultural Library (NAL)
• Uses ICASA Data Standards for crops (~30 years
experience) as a Data Dictionary (White et al., 2014)
• New definition of livestock core data, data dictionary
• Demonstrated by AgMIP to work for different crop
models (e.g., running APSIM and DSSAT models with
same inputs, assumptions)
• Active contributions by researchers, initially 13 core
states; open to all
• ARS endorsement and support for data portal at NAL
• Endorsed by international data initiatives and private
sector collaborators
16
Virtual Research Lab: Network of Networks
Uncertainty, understanding and reducing it
Open, discoverable, accessible, and usable
data
Incorporating modern genetic information in
crop models
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Brief Overview of Three AgMIP Efforts
Genetic Coefficients – genotype-specific parameters (GSPs) for
simulating variations among cultivars/hybrids
First used in CERES maize and wheat models (J. T. Ritchie and
others)
Key GSPs defined for depicting phenological development
differences (phase durations, daylength sensitivities) and yield
formation
Most crop models now have these, BUT there are major
differences among models in definitions/use
Difficult to estimate, limiting applicability of models
No direct link to modern genetic information, and field data are
required to parameterize GSPs for each cultivar
Next generation crop models must provide a better way to
include gene effects (gene-based crop model)
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Genetics and Crop Models
Current approaches – develop relationships between GSPs
and QTLs (e.g., White and Hoogenboom, 1996, 2003;
Messina et al., 2006; etc.)
Why not continue this?
• Current models do not include GSPs for all processes and traits that
we now know are under genetic control (examples from this study)
• May need to modify environmental effects, interactions, in the model
• Current crop models are not ideally structured to make all of the
changes that are needed.
• Major changes may be needed, but code may be reusable
• Although some existing crop models are modular, new modules are
needed that are designed based on what we are now learning about
genetic control of processes and so that new modules can be easily
modified as more is learned, fine granularity
Need for a gene-based model
MR is a GSP (genetic coefficient)
g(P,G) includes two GSPs (slope and critical day length)
f(T) has been assumed to be the same for all genotypes
(e.g., base and optimal temperature thresholds did not
vary across genotypes)
RF thus depends on G; E={T,P} for a particular day t
Geneticists and bioinformatics specialists in UF’s NSF
gene-based modeling project told us we were wrong
They were correct; we had developed this model over 20
years ago
Incorporating Genetic Effects RF(t|G,T,P) = MR(G) * f(T) * g(P,G)
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Two Models for Rate of Development
Toward First Flower
Hwang et al., Agricultural Systems (in review)
1. In the DSSAT CROPGRO-Bean Model
RF(t|G,T,P) = MR(G) * f(T) * g(P,G)
𝑅𝐹 𝑡 = 0.029 + 7.5 · 10−4 𝑇𝑀𝐸𝐴𝑁 𝑡 − 21.35 − 7.3 · 10−6 𝑆𝑅𝐴𝐷 𝑡 − 18.31 − 2.2 · 10−3 𝐷𝐿 𝑡 − 12.7
−3.3 · 10−4 𝑇𝑀𝐸𝐴𝑁 𝑡 − 21.35 𝐷𝐿 𝑡 − 12.7
+9.8 · 10−4 · 𝑇𝐹1 + 1.7 · 10−3 · 𝑇𝐹2 − 3.9 · 10−4 · 𝑇𝐹3 + 2.0 · 10−4 · 𝑇𝐹4
−1.5 · 10−4 · 𝑇𝐹5 + 8.9 · 10−4 · 𝑇𝐹6 − 5.3 · 10−4 · 𝑇𝐹7 − 3.1 · 10−4 · 𝑇𝐹8
−3.4 · 10−4 · 𝑇𝐹9 − 9.7 · 10−5 · 𝑇𝐹10 + 2.6 · 10−4 · 𝑇𝐹11 − 6.6 · 10−5 · 𝑇𝐹12
+𝑇𝐹2 −3.6 · 10−5 𝑇𝑀𝐸𝐴𝑁 𝑡 − 21.35
+𝑇𝐹3 6.7 · 10−5 𝑇𝑀𝐸𝐴𝑁 𝑡 − 21.35 − 1.1 · 10−3 𝐷𝐿 𝑡 − 12.7
+𝑇𝐹5 5.5 · 10−5 𝑇𝑀𝐸𝐴𝑁 𝑡 − 21.35
+𝑇𝐹7 −2.6 · 10−4 𝐷𝐿 𝑡 − 12.7
+𝑇𝐹12(−6.4 · 10−6 𝑆𝑅𝐴𝐷 𝑡 − 18.31 − 3.9 · 10−4 𝐷𝐿 𝑡 − 12.7 ) [7]
2. Mixed effects dynamic model, geneticists
Phaseolus Vulgaris L. (Common Bean)
184 Recombinant Inbred Lines, 5 locations
Mixed Effects dynamic development rate model
Equation (2) from last slide
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
GxE Model for Time to First Flower
Significant temperature – daylength interactive effects on
development rate/time of first flower
Significant effects of 12 quantitative trait loci (QTL) on
development variations among genotypes
From 75 to 80% variability among 184 genotypes grown in 5
environments were explained by GxE dynamic model
And concluded:
Need to revise functional form of development rate model
Need to develop fine granularity modules for dynamic
processes (as traits)
Develop an evolutionary pathway toward Next Generation
models as knowledge is gained
More collaboration among crop modelers, geneticists, and
bioinformatics may lead to transformations in models
We found that
New activity to compare different gene-based phenology
models and gene incorporation into 12 rice crop models
Co-led by T. Li (IRRI), X. Yin (WUR), and T. Lefarge
(CIRAD)
Data on about 300 lines grown across different
environments are available from IRRI
Additional data from CIRAD/IRRI to be made available
Evaluate phenology components of existing models, using
genetic information to inform each, and a new non-linear
mixed effects dynamic model
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
AgMIP Rice Modeling Team
Intercomparison of models for specific crops
Harmonizing databases, model inputs
Model improvement; gene-based modeling
Incorporating pest and disease damage in crop models
Crop rotations, residue management
Expanding models to include under-served crops (e.g.,
vegetables, fruit, pasture, trees)
Incorporation of nutrient composition in crop models (ILSI)
Incorporation of ozone effects
Spatial modeling (e.g., for regional & global assessments)
Coupling crop, livestock & household economic models for
farming systems analyses
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
AgMIP, MACSUR Contributions to Next Generation
Crop Modeling
Early crop modelers can feel pride in getting this far, but
humility and a new generation of models are needed.
Open, discoverable, accessible, and usable data are
essential; More extensive datasets are needed to develop and
test models for more reliable, robust simulations
Technologies continue to enable new models, new
capabilities and application opportunities
• Molecular genetics
• Low powered sensors, rapid phenotyping platforms
• Internet of things
User demands for information derived from models will grow
for specific use-cases
Should establish requirement for publicly-funded agronomic
research to include modeling component and make data
openly available
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016
Prospects for Next Generation Crop Models
Adopt a data culture across the research to application chain
Include math and modeling courses in degree programs
Include interdisciplinary experience in graduate agricultural
science research programs
Emphasize transdisciplinary research, teams incorporating
geneticists, field experimentalists, modelers, pathologists,…
Provide funding opportunities & incentives for incorporating
modeling components into research
Continue to build community of science; but not silos
Embrace diversity of approaches
Laying Foundation for Next Generation Models
Thank You
Crop Modeling for Agriculture and Food Security Berlin, Germany: 15-17 March 2016