Modelling pig and poultry production systems: computational and conceptual challenges

Modelling pig and poultry production systems: computational and conceptual challenges

M. Gilbert (& T. Van Boeckel)Université Libre de Bruxelles http://lubies.ulb.ac.be/Spatepi.html

T. RobinsonInternational Livestock Research Institute

Livestock Human population Spatial epidemiology & invasion ecology

Catherine Linard

Yann Forget Jean Artois

Clément TisseuilGaëlle Nicolas

Weerapong Thanapongtharm

Postdocs

PhD

http://lubies.ulb.ac.be/Spatepi.html

Intensified livestock production systems and the emergence of Highly Pathogenic Avian Influenza

Favour infections

High density & contacts

Genetic similarity

Living & health

condition

HPAI emergence mostly documented in intensive poultry production systems

Intensified livestock production systems and agricultural antimicrobial use

Favour infections

High density & contacts

Genetic similarity

Living & health

condition

Marginal gains due higher off-take rates do make a difference over large volume

(but see Graham et al. 2007)

Feed conversion

rate matters

Fast prod. cycles

High inputs /

high outputs

Higher use of antimicrobials in intensive systems (preventive, curative, feed additive)

Global trends in livestock numbers

0

500,000,000

1,000,000,000

1,500,000,000

2,000,000,000

2,500,000,000

1961

1965

1969

1973

1977

1981

1985

1989

1993

1997

2001

2005

2009

Hea

ds

Cattle Chicken (/10) Pork

Source: FAOSTAT (2010)

Global trends in livestock productivity

90

110

130

150

170

190

210

230

250

1961

1965

1969

1973

1977

1981

1985

1989

1993

1997

2001

2005

2009

Rela

tive

incr

ease

in O

utpu

t/In

put

(kg

outp

uts

head

-1 y

ear-

1)

Cattle (331) Chicken (1.73) Pork (60.9)

Source: FAOSTAT (2010)

Outline

Context

• Intensification has taken place rapidly in the past• Strong projected changes in demand will lead to further

intensification• Changes are structured geographically

Objectives

• Better document the geographic distribution of intensive livestock production

• Develop tools for making projections

Method

s

• Mapping the global distribution of livestock• Disaggregating in production systems

Livestock distribution: Gridded Livestock of the World (GLW 1.0)

• General principle• Collection of sub-national livestock

census data• Many variables correlated to livestock

farming are mapped at high resolution (e.g. land cover).

• Statistical models are based on high resolution GIS predictors and applied to downscale census values by pixel (stratified multiple linear regressions)

• Previous developments• GLW 1.0 published by FAO in 2007,

mostly based on census data < 2005 (Wint & Robinson 2007)

• Global extent, 5 km resolution

Livestock distribution: Gridded Livestock of the World (GLW 2.0)

• Recent developments• More recent & higher resolution census data• Spatial modelling @ 1km resolution• Automation of the methodology in R• Disseminated through the Livestock GeoWiki

• http://www.livestock.geo-wiki.org/

• New species division• Cattle• Pig• Chicken• Duck• Sheep• Goat

Robinson, T., W. Wint, T, G. Conchedda, T. P. Van Boeckel, V. Ercoli, E. Palamara, G. Cinardi, L. D’Aietti, & M. Gilbert (2014) Mapping the Global Distribution of Livestock. PLoS ONE 9(5): e96084. doi:10.1371/journal.pone.0096084

Livestock distribution: Gridded Livestock of the World (GLW 3.0)

• In progress…• New machine learning algoritm (Random Forest)

• Systematic evaluation (years of CPU time in 4 months)• 180 models for Asia chicken and Africa cattle

• Processing on ILRI cluster (parrallelized)• Full integration of metadata• Spatial modelling & dissemination @ 1 km & 10 km resolution• Toward global runs instead of continental tiles• Revision of predictor variable to include more anthropogenic

factors

Livestock distribution: Gridded Livestock of the World (GLW 3.0)

• In progress…• New machine learning algoritm (Random Forest)

• Systematic evaluation (years of CPU time in 4 months)• 180 models for Asia chicken and Africa cattle

• Processing on ILRI cluster (parrallelized)• Full integration of metadata• Spatial modelling & dissemination @ 1 km & 10 km resolution• Toward global runs instead of continental tiles• Revision of predictor variable to include more anthropogenic

factors

Livestock distribution: Gridded Livestock of the World (GLW 3.0)

RF (GLW 3.0) vs. Stratified regression (GLW 2.0) leave-out cross validation

Livestock distribution: Gridded Livestock of the World (GLW 3.0)

1 week / species

GLW 2.0 GLW 3.0

16h /species

1 month/ species

1-2 days / species

Outline

Context•Intensification has taken place rapidly in the past•Strong projected changes in demand will lead to further intensification•Changes are structured geographically

Objectives •Better document the geographic distribution of intensive livestock production•Develop tools for making projections

Methods •Mapping the global distribution of livestock•Disaggregating in production systems

Conceptual framework (1)

Conceptual framework (2)

Conceptual framework (3)

The % ext. chicken is predicted at national level by the GDP model

Ext. raised chicken are distributed equally across rural population

Intensively raised poultry is estimated by the difference with the total

Application to chicken

Extensively raised chicken

Intensively raised chicken

Validation: chicken extensive

Van Boeckel, T., W. Thanapongtharm, T. Robinson, L. D’Aietti & M. Gilbert (2012). Predicting the distribution of intensive poultry farming in Thailand. Agriculture, Ecosystem and Environment. Doi: 10.1016/j.agee.2011.12.019

Validation: chicken intensive

Van Boeckel, T., W. Thanapongtharm, T. Robinson, L. D’Aietti & M. Gilbert (2012). Predicting the distribution of intensive poultry farming in Thailand. Agriculture, Ecosystem and Environment. Doi: 10.1016/j.agee.2011.12.019

Application to pigs:extensive, semi-intensive, intensive

Application to pigs:extensive, semi-intensive, intensive

Pigs: extensive

Pig semi-intensive

Pig intensive

Disaggregating between extensive and intensive production systems

• Limitations• Uncertainty in the GDP model (& other important variables ?)• Ignore sub-national GDP variations• Assumption of equal number of Ext. Chicken / rural population

Validation: chicken extensive

Van Boeckel, T., W. Thanapongtharm, T. Robinson, L. D’Aietti & M. Gilbert (2012). Predicting the distribution of intensive poultry farming in Thailand. Agriculture, Ecosystem and Environment. Doi: 10.1016/j.agee.2011.12.019

Discussion (1)

People

• Number• Weatlh• Diet

Livestock

• Number• Production

systems

Impact

• Amonia pollution

• GHG emissions

• EIDs• Antimicrobial

resistance

Drivers of change in spatial distribution

Drivers of change in number

Demand

Discussion (2)

People

Livestock

Demography Wealth

# ConsumersDietary

preferencesUrbanization of

consumers

Change in stock

Change in productivity

Urbanization

Vertical integration and distribution of

inputs and demand

Future work (1)

Livestockproducts

Change in stock

Change in productivity

Vertical integration and concentration of

demand

• Methodological improvements• Using agricultural population• Using sub-national GDP where appropriate (e.g.

China, India)

• Forward and backward predictions

Future work (2)

2000log GDP per capita c. $ 2.9% extensive c. 83 %

2000

2030

2030log GDP per capita c. $ 3.8% extensive c. 18 %

Chicken production in China

Future work (3)

Livestockproducts

Change in stock

Change in productivity

Vertical integration and concentration of

demand

• Methodological improvements• Using agricultural population• Using sub-national GDP where appropriate (e.g.

China, India)

• Forward and backward predictions• GDP data & projections (national / sub-national)• Spatial concentrations (peri-urban, access to port,

founder effect)

Future work (4)

People

• Number• Weatlh• Diet

Livestock

• Number• Production

systems

Impact

• Amonia pollution

• GHG emissions

• EIDs• Antimicrobial

resistance