African Sustainable Livestock 2050

Aishwarya Venkat

EcoHealth Net Research Exchange Intern, Summer 2017

About Me

About Me

African Sustainable Livestock (ASL) 2050

• FAO project funded by USAID
• Goals:
  • identify opportunities and threats associated with the long-term development of livestock
  • agree upon priority reforms and investments, and the capacity needed for their implementation, to ensure sustainable development of the livestock sector in the next three or four decades.
• Countries: Burkina Faso, Egypt, Ethiopia, Kenya, Nigeria and Uganda

EHA & ASL 2050

• Metaflu (Hosseini et al, 2013) package to simulate propagation and impacts of seeded avian influenza outbreaks
  • Kate: variable farm sizes, growth, culling
  • John: presence/absence of markets, effect on households and farms
  • Me: spatial data collection, developing pipeline for 'real world' data input into Metaflu

Key Actors

Households

• Key exposure points for humans
• 60-85% of livestock sector in ASL countries is driven by household/backyard poultry farming
• Easy targets for interventions and incentives

Commercial Farms

• Operations of scale, most impacted by avian influenza outbreaks
• Most ASL2050 countries aim to expand commercial poultry operations in next decade-incentive for improved biosecurity

Markets

• Daily, weekly markets frequented by majority of people
• Locations of high chicken movement between actors

Knowns & Unknowns

Small World Networks

Watts, Duncan J., and Steven H. Strogatz. Collective dynamics of 'small-world' networks, Nature 393.6684 (1998): 440

Goals & Methods

1. Ground Metaflu simulations in reality
   • Country-wide poultry sector information collection
   • Spatial data processing
     • Probability surface development and spatial sampling for households and farms
     • Assignment of 'farm size' based on extensive/intensive production rasters
2. Apply Small World Network principles to Random Spatial Networks
   • Generate network representations of spatial data
   • Define how each actor is connected to the others
   • Parametrize network to allow for testing of intensification and connectivity
3. Identify limitations of existing data, and generate models and questions to share with FAO theme leaders

Livestock Survey Data

• Primary data source: Country-level Livestock Survey data aggregated by FAO

##    HHID    EAID Latitude Longitude Household.Weight Chickens
## 1 87095 1912049    9.324    38.592             3971        8
## 2 86375 1270896   10.394    38.225             4752        2
## 3 84926 1391416   15.392    39.212             3230        4

• Surveys provide approximate location information

• Representative households are actually spread out across survey enumeration areas

Approach: Households

• Household/Backyard chicken production density from Marius et al, 2015

• Use this as probability surface to distribute households within country

Approach: Commercial Farms

• Get number and sizes of commercial farms from literature, poultry sector reports (Ethiopia, Kenya, Uganda), or scraped from OIE & FAO EMPRES-I outbreak data

• Fit lognormal model to farm size, sample randomly from distribution until value adds up to total estimated intensive chicken population

Aproach: Commercial Farms

• Distribute points based on intensive poultry production raster

Approach: Markets

• Ethiopia, Uganda, Kenya: Intergovernmental Authority on Development (IGAD)

• Burkina Faso, Egypt, Nigeria: Populated Places data from OpenStreetMap and SEDAC GRUMP

Combining Data

Parametrize connectivity

Parametrize connectivity

Parametrize connectivity: households

Parametrize connectivity: households

Connecting poultry sector actors

\[p_{u,v} = min\Bigg(\kappa_u\kappa_v\frac{f(d_{uv})}{\rho\langle\kappa\rangle},1\Bigg)\]

$u,v = $ two nodes (any of households, markets, farms)
$ p = $ probability of connection between nodes \(u\) and \(v\)
$f(d_{uv}) = $ exponential decay kernel connecting nodes, defined by a distance (rate) at which 50% of nodes in country are connected
$\kappa = $ expected degree of connections per node (Poisson distributed across nodes)
$\langle\kappa\rangle = $ average degree of connections for all nodes
$\rho = $ density of nodes within country

Lang, John, et al. Random Spatial Networks: Small Worlds without Clustering, Traveling Waves, and Hop-and-Spread Disease Dynamics. arXiv:1702.01252 (2017).

Progress

• Identified data sources for all six countries, validation and review in progress
• Pipeline for analysis and input into Metaflu defined
• Experiments to test kappas and distances (rates) for reproducibility across countries

Next Steps

• Network generation for all ASL countries
• Break down network into modules for metaflu simulations
• Develop risk maps and outbreak probability analyses
• Writing and publication of results

Skills learned

• GIT!
• Parallel processing
• Network analysis in igraph
• Sparse matrices
• Geographically weighted principal components analysis (PCA)
• Raster stack manipulation and raster PCA
• Random forest models and boosted regression trees

Acknowledgements

• Noam Ross

• Cale Basaraba

• Allison White

• Carlos Zambrana-Torrelio

• Modeling & Analytics team

• Orsolya Mikecz

• Antonio Mele

• Ugo Pica-Ciamarra