Transcript IMPACT Data-Model Philosophy

Introducing IMPACT 3: Modeling
Philosophy and Environment
Sherman Robinson
Daniel Mason-D’Croz
Shahnila Islam
1
Global Futures and IMPACT
• Objective: Use IMPACT for ex-ante analysis of potential agricultural
technologies to help policy makers prioritize agricultural
investments
• Phase 1: IMPACT Developments:
–
–
–
–
Welfare Module
Benefit-Cost Analysis
Technology Adoption Module
Tracking progress against MDGs
• Challenges identified in Phase 1:
–
–
–
–
Insufficient geographic disaggregation
Need to model more CG-mandate crops
2000 base year outdated
Model needed to be recoded to allow for better integration with new
modules under development (water, livestock, fish, biofuels)
2
What is IMPACT 3?
• More than a new FAO download and cleaner
code
• A modeling-data platform built on modularity
and interoperability
– Harmonized Data
– Data driven
model specification
– More flexible to
meet user needs
3
Why Data Harmonization?
FAO Data Collection
• IMPACT integrates
various models, which
often use similar input
data
• Better data sharing,
common definitions,
and clear responsibility
of data processing
removes redundancy
and improves quality
control
Bulk Download
Data Cleaning
Livestock
Production
Crop Production
Commodity
Demand and Trade
Data Processing
Spatial disaggregation
Balance Demand, and Trade
with Production
IMPACT 3 FAO Database
4
Why Data Harmonization?
FAO Data Collection
• IMPACT integrates
various models, which
often use similar input
data
• Better data sharing,
common definitions,
and clear responsibility
of data processing
removes redundancy
and improves quality
control
SPAM
Bulk Download
Data Cleaning
Livestock
Production
Crop Production
Commodity
Demand and Trade
Data Processing
Spatial disaggregation
Balance Demand, and Trade
with Production
IMPACT 3 FAO Database
5
Why Data Harmonization?
FAO Data Collection
• IMPACT integrates
various models, which
often use similar input
data
• Better data sharing,
common definitions,
and clear responsibility
of data processing
removes redundancy
and improves quality
control
SPAM
IMPACT
Bulk Download
Data Cleaning
Livestock
Production
Crop Production
Commodity
Demand and Trade
Data Processing
Spatial disaggregation
Balance Demand, and Trade
with Production
IMPACT 3 FAO Database
6
IMPACT Data-Model Environment
Shared Data
FAO
Climate
Data
Data Processing
Data Users
SPAM
IMPACT
Models
Land-Use
Model
Exogenous
IMPACT
Parameters
Geospatial
and
Subnational
Data
Hydrology
Crop
Models
7
Share Data
• FAO
–
–
–
–
–
Crop Production
Livestock Production
Supply-Utilization
Food Balance Sheets
Water Stress
• Climate Data
– GCMS
– Generated Weather
• Geospatial and
Subnational Data
–
–
–
–
Irrigation
Subnational Statistics
Crop suitability maps
Population Density
• Exogenous IMPACT
Parameters
–
–
–
–
–
Yield, Area Growth
Elasticities
Prices (AMAD)
Population
GDP
8
Models
• SPAM - Spatial
Production Allocation
Model
• Land-Use Model
• DSSAT Crop Models
• Biofuel Model
• Hydrology Model
• Water Basin
Management Model
• Water Stress Model
• Food Model
–
–
–
–
Crops
Livestock
Sugar
Oilseeds
9
FAO Data
FAO
IMPACT
Climate
Data
SPAM
Exogenous
IMPACT
Parameters
FAO:
Estimation
• Food
• Water Stress
• Water Demand
Geospatial
and
Subnational
Data
Shared Data
Direct Users of FAO
Using Processed FAO
10
Processing FAO Data
Source Data
(FAO, SPAM)
• FAO Bulk Download for
3-year average around
2005 (04-06)
Priors on values and
Feedback to
estimation errors of
data source
• Harmonized
production, demand,
and trade
SPAM/IMPACT
commodity, and
New information to
Estimation by Crosscorrect identified
geographic definitions
Entropy Method
problems
• Bayesian Work Plan
– Iterate with new
information
Check results against
priors and identify
potential data problems
11
Data Harmonization and Quality
• Too many cooks
– Climate change is modeled in Water and Crop
models for IMPACT
– Need to use same initial and processed climate
data
– Ensure crop shocks and water shocks are
compatible
12
Climate Change Consistency
FAO
IMPACT
Climate
Data
Crop
Models
Exogenous
IMPACT
Parameters
Hydrology
• Food
• Water Demand
• Water Stress
Geospatial
and
Subnational
Data
Shared Data
Users of Climate Data
Use Aggregated
Processed Climate Data
13
Data Harmonization and Quality
• Building common geographical definitions
• Standardize mapping of data
• Share data (initial and processed)
14
Geospatial Data Users
FAO
SPAM
IMPACT
Climate
Data
Land-Use
Model
Exogenous
IMPACT
Parameters
Geospatial
and
Subnational
Data
Shared Data
Hydrology
• Food
• Water Demand
• Water Stress
Crop Models
Users of Geospatial and
Subnational Data
Use Aggregated Outputs
from direct users
15
Modularity – Data Partitioning
• IMPACT model is now data driven
– General code built on specific data structures
• Each dataset has unique problems
– Detox drivers vs. self-driving car
– Data Processing
is source-specific
– Model Inputs are
model-specific
16
Modularity – IMPACT Partitioning
• IMPACT model is now data driven
– General code built on specific data structures
• Each dataset has unique problems
– Detox drivers vs. self-driving car
– Data Processing
is source-specific
– Model Inputs are
model-specific
17
Benefits of Data Independence
• Cleaner Model Code
– Facilitate model transfer and training
• Data Processing and Model design are
independent tasks
• Model can run different data sources and
aggregations without modification
18