Livestock & Grazing

Overview

The livestock module models animal product production (meat, dairy, eggs) through two distinct production systems:

  • Grazing-based: Animals feed on managed grasslands

  • Feed-based: Animals consume crops as concentrated feed

Animal Products

The model includes seven major animal product categories configured in config/default.yaml:

animal_products:
  include:
  - meat-cattle
  - meat-pig
  - meat-chicken
  - dairy
  - eggs
  - dairy-buffalo
  - meat-sheep
  # Feed conversion efficiency mode (how much feed is required per unit product)
  # Source: Wirsenius (2000) regional feed energy requirements
  # Options:
  #   - List of regions: average efficiencies across those regions (all countries use same values)
  #   - null: use country-specific regional efficiencies based on geographic mapping
  # Available regions: East Asia, East Europe, Latin America & Caribbean,
  #   North Africa & West Asia, North America & Oceania, South & Central Asia,
  #   Sub-Saharan Africa, West Europe
  feed_efficiency_regions:
  - North America & Oceania
  - West Europe
  # Ruminant net-to-metabolizable energy conversion efficiency factors
  # Used to convert net energy (NE) requirements to metabolizable energy (ME) requirements
  # Based on NRC (2000) typical values for mixed diets
  # ME_required = NE_m/k_m + NE_g/k_g (+ NE_l/k_l for dairy)
  # TODO: Should check the reference for this.
  net_to_metabolizable_energy_conversion:
    k_m: 0.60  # Maintenance efficiency
    k_g: 0.40  # Growth efficiency
    k_l: 0.60  # Lactation efficiency (dairy)
  # Carcass-to-retail meat conversion factors
  carcass_to_retail_meat:
    meat-cattle: 0.67  # kg boneless retail beef per kg carcass (OECD-FAO 2023)
    meat-pig: 0.73     # kg boneless retail pork per kg carcass (OECD-FAO 2023)
    meat-chicken: 0.60 # kg boneless retail chicken per kg carcass (OECD-FAO 2023)
    eggs: 1.00         # No conversion needed (whole egg = retail product)
    dairy: 1.00        # No conversion needed (whole milk = retail product)
    meat-sheep: 0.63   # kg boneless retail lamb per kg carcass (slightly lower than beef)
    dairy-buffalo: 1.00 # No conversion needed (whole milk = retail product)
  # FAOSTAT QCL item names to aggregate for each model product.
  # First item is the primary product; additional items are proxied species
  # whose production is lumped into the model product.
  faostat_items:
    dairy:
      - "Raw milk of cattle"
      - "Raw milk of goats"       # proxy: goat milk → dairy
      - "Raw milk of sheep"       # proxy: sheep milk → dairy
      - "Raw milk of camel"       # proxy: camel milk → dairy
    meat-cattle:
      - "Meat of cattle with the bone, fresh or chilled"
      - "Meat of buffalo, fresh or chilled"   # proxy: buffalo → cattle
    meat-pig:
      - "Meat of pig with the bone, fresh or chilled"
    meat-chicken:
      - "Meat of chickens, fresh or chilled"
      - "Meat of ducks, fresh or chilled"            # proxy: duck → chicken
      - "Meat of turkeys, fresh or chilled"           # proxy: turkey → chicken
      - "Meat of pigeons and other birds n.e.c., fresh, chilled or frozen"
    eggs:
      - "Hen eggs in shell, fresh"
    dairy-buffalo:
      - "Raw milk of buffalo"
    meat-sheep:
      - "Meat of sheep, fresh or chilled"
      - "Meat of goat, fresh or chilled"              # proxy: goat → sheep
  feed_proxy_map:
    dairy-buffalo: dairy
    meat-sheep: meat-cattle
  feed_efficiency_calibration:
    enabled: true
    generate: false
    source: "data/curated/feed_efficiency_calibration.csv"
    max_multiplier: 2.0
    scenario: "default" # Which scenario's solved model to generate calibration from
  residue_crops:
  - banana
  - barley
  - chickpea
  - cowpea
  - dry-pea
  - dryland-rice
  - foxtail-millet
  - gram
  - maize
  - oat
  - pearl-millet
  - phaseolus-bean
  - pigeonpea
  - rye
  - sorghum
  - sugarcane
  - wetland-rice
  - wheat

grazing:
  enabled: true
  isimip_utilization_rate: 0.50 # Applied to ISIMIP yields in merge step
  forage_overlap_subtraction_alpha: 0.0 # Subtract alpha * modeled forage-crop supply from LUIcube grass supply by country
  forage_overlap_crops:
  - alfalfa
  - silage-maize
  - biomass-sorghum
  grassland_forage_calibration:
    enabled: true
    generate: false
    source: "data/curated/grassland_forage_calibration.csv"
    scenario: "default"

Each product can be produced via either production system, with different feed requirements and efficiencies.

Production Systems

Grazing-Based Production

Concept: Animals graze on managed grasslands, converting grass biomass to animal products.

Inputs:

  • Land (per region and resource class, similar to cropland)

  • Managed grassland yields from ISIMIP LPJmL model

Process:

  1. Grassland yields (t dry matter/ha/year) are computed per region and resource class

  2. Feed conversion ratios translate grass biomass → animal products

  3. Land allocation to grazing competes with cropland expansion

Configuration: Enable/disable grazing with grazing.enabled: true

Feed-Based Production

Concept: Animals consume crops (grains, soybeans, etc.) as concentrated feed.

Inputs:

  • Crops from crop production buses

  • Feed conversion ratios (kg crop → kg animal product)

Process:

  1. Crops are allocated to animal feed (competing with direct human consumption)

  2. Feed conversion links transform crop inputs to animal products

  3. Multiple crops can contribute (e.g., maize + soybean for poultry)

Grassland Yields

Grazing supply is determined by managed grassland yields from the ISIMIP LPJmL historical simulation.

Data Source

Dataset: ISIMIP2b managed grassland yields (historical)

Resolution: 0.5° × 0.5° gridded annual yields

Variable: Above-ground dry matter production (t/ha/year)

Processing: workflow/scripts/build_grassland_yields.py

Aggregation follows the same resource class structure as crops:

  1. Load grassland yield NetCDF

  2. Aggregate by (region, resource_class) using area-weighted means

  3. Output CSV with yields in t/ha/year

Pasture Utilization

The model assumes that only a portion of the total grassland biomass production is available for grazing livestock. This reflects the need to leave biomass for regrowth, soil protection, and ecosystem function (“take half, leave half” principle). The correction is applied upstream in the merge_grassland_yields step so that the yield column in the merged output is already effective feed yield:

  • LUIcube rows: yield is multiplied by the per-cell grazing_intensity from LUIcube data.

  • ISIMIP rows: yield is multiplied by a fixed utilization rate (default 50%).

  • Parameter: grazing.isimip_utilization_rate in configuration (applied to ISIMIP fallback yields).

This value is consistent with the GLOBIOM model, which assumes a 50% grazing efficiency for grass in native grasslands [3]. While intensive dairy systems can achieve higher utilization (up to 70-80%), global rangeland management guidelines typically recommend utilization rates below 50% to prevent degradation.

Managed grassland yield potential

Global distribution of managed grassland yield potential (tonnes dry matter per hectare per year) from ISIMIP LPJmL historical simulations

Feed Conversion

The model uses feed conversion ratios to link feed inputs to animal outputs, with explicit categorization by feed quality to enable accurate CH₄ emissions tracking.

Feed System Architecture

The feed system uses seven distinct feed pools that combine animal type with feed quality:

  • Ruminant pools: ruminant_roughage, ruminant_forage, ruminant_grain, ruminant_protein

  • Monogastric pools: monogastric_low_quality, monogastric_grain, monogastric_protein

This categorization enables the model to:

  1. Differentiate methane emissions using GLEAM feed digestibility classes (roughage/forage vs. grain/protein)

  2. Route crops, residues, and processing byproducts to appropriate feed pools based on nutritional properties

  3. Model production system choices (e.g., roughage-dominated beef vs. high-grain finishing rations)

  4. Distinguish between grazing (grassland) and confinement feeding systems for nitrogen management

Feed Properties (Generated from GLEAM)

Feed properties (digestibility, metabolizable energy, protein content) are automatically generated from GLEAM 3.0 data during workflow execution. The workflow produces two files in processing/{name}/:

  • ruminant_feed_properties.csv: Properties for all feeds used by ruminants

  • monogastric_feed_properties.csv: Properties for all feeds used by monogastrics

Each file contains:

  • feed_item: Item name (e.g., “maize”, “wheat-bran”)

  • source_type: Either “crop” or “food” (byproduct)

  • digestibility: Digestible fraction (0-1)

  • ME_MJ_per_kg_DM: Metabolizable energy (MJ per kg dry matter)

  • CP_pct_DM: Crude protein (% of dry matter)

  • ash_pct_DM: Ash content (% of dry matter)

  • NDF_pct_DM: Neutral detergent fiber (% of dry matter)

These properties are extracted from the GLEAM 3.0 supplement using data/curated/gleam_feed_mapping.csv to map between model feed items and GLEAM feed categories.

Feed quality categories (assigned based on nitrogen content and digestibility):

  • Ruminant feeds:

    • Protein: High nitrogen content (>50 g N/kg DM) - protein meals such as rapeseed-meal, oilseed-meal, soybean meal (assigned by N content; takes precedence over digestibility)

    • Roughage: Low digestibility (<0.55), high-fiber forages (crop residues, straw)

    • Forage: Medium digestibility (0.55-0.70), improved forages and grassland (silage maize, alfalfa, pasture)

    • Grain: High digestibility (0.70-0.90), energy concentrates (maize, wheat, barley)

  • Monogastric feeds:

    • Protein: High nitrogen content (>35 g N/kg DM) - protein meals such as soybean meal, fish meal, rapeseed-meal (assigned by N content; takes precedence over energy)

    • Low quality: Low metabolizable energy (<11 MJ/kg DM), bulky feeds and byproducts

    • Grain: Medium energy (11-15.5 MJ/kg DM), cereal grains

    • Energy: High energy (>15.5 MJ/kg DM), fats and high-energy feeds

Categorization logic: Both ruminant and monogastric feeds prioritize nitrogen content to identify protein feeds, ensuring high-protein oilseed meals are correctly classified regardless of digestibility. For feeds below the nitrogen threshold, ruminants use digestibility ranges while monogastrics use metabolizable energy thresholds.

Byproducts from food processing (with source_type=food) are automatically excluded from human consumption and can only be used as animal feed.

Feed Conversion Efficiencies

Feed conversion efficiencies (tonnes retail product per tonne feed DM) are generated automatically from Wirsenius (2000) regional feed energy requirements combined with GLEAM 3.0 feed category energy values.

Proxy Products: Buffalo milk (dairy-buffalo) and sheep meat (meat-sheep) use cattle feed requirements as proxies, since Wirsenius (2000) does not provide separate regional estimates for these products. Buffalo milk inherits dairy cattle parameters, while sheep meat inherits beef cattle parameters with adjustments for the different carcass-to-retail conversion factor (0.63 for sheep vs 0.67 for cattle).

In this calculation, we have to account for the following units: * Feed inputs: Dry matter (tonnes DM) * Animal product outputs: Fresh weight, retail meat (tonnes fresh weight)

  • For meats: retail/edible meat weight (boneless, trimmed) - NOT carcass weight

  • For dairy: whole milk (fresh weight)

  • For eggs: whole eggs (fresh weight)

Wirsenius (2000) [1] provides feed requirements per kg carcass weight (dressed, bone-in). We apply carcass-to-retail conversion factors to obtain feed requirements per kg retail meat, from OECD-FAO Agricultural Outlook 2023-2032, Box 6.1 [2]:

  • Cattle meat: 0.67 kg boneless retail per kg carcass

  • Sheep meat: 0.63 kg boneless retail per kg carcass

  • Pig meat: 0.73 kg boneless retail per kg carcass

  • Chicken meat: 0.60 kg boneless retail per kg carcass

  • Eggs, dairy, & buffalo milk: 1.00 (no conversion, already retail products)

Generation workflow:

  1. Regional feed energy requirements from Wirsenius (2000) provide MJ per kg carcass output for eight world regions

  2. Carcass-to-retail conversion: Convert MJ per kg carcass → MJ per kg retail meat

    • For meats: ME_retail = ME_carcass / carcass_to_retail_factor

    • For dairy/eggs: No conversion (already retail products)

  3. Energy conversion for ruminants: Net energy (NE) requirements converted to metabolizable energy (ME) using NRC (2000) efficiency factors:

    • k_m = 0.60 (maintenance)

    • k_g = 0.40 (growth)

    • k_l = 0.60 (lactation)

  4. Feed category energy content from GLEAM 3.0 provides ME (MJ per kg DM) for each feed quality category

  5. Efficiency calculation: efficiency = ME_feed / ME_retail (tonnes retail product per tonne feed DM)

Output: processing/{name}/feed_to_animal_products_uncalibrated.csv with columns:

  • country: ISO 3166-1 alpha-3 country code

  • product: Product name (e.g., “meat-cattle”, “dairy”)

  • feed_category: Feed pool (e.g., ruminant_forage, ruminant_grain, monogastric_grain)

  • efficiency: Feed conversion efficiency (t product / t feed DM)

  • notes: Description with inverse feed requirement

Configuration: The feed_efficiency_regions setting controls how feed conversion efficiencies are assigned:

animal_products:
  # Option 1: Average specific regions (all countries use same values)
  feed_efficiency_regions:
  - North America & Oceania
  - West Europe

  # Option 2: Use country-specific regional values (set to null)
  # feed_efficiency_regions: null

Available regions (from Wirsenius 2000): East Asia, East Europe, Latin America & Caribbean, North Africa & West Asia, North America & Oceania, South & Central Asia, Sub-Saharan Africa, West Europe

When feed_efficiency_regions is null, each country uses the feed conversion efficiencies from its geographic region. The mapping from countries to Wirsenius regions is defined in data/curated/country_wirsenius_region.csv.

Example efficiencies (North America & Oceania + West Europe average, with carcass-to-retail conversion):

  • Cattle meat from forage: ~0.026 t/t (~38 t DM feed per tonne retail beef)

  • Cattle meat from grain: ~0.035 t/t (~28 t DM feed per tonne retail beef)

  • Dairy from forage: ~0.480 t/t (~2.1 t DM feed per tonne milk)

  • Pig meat from grain: ~0.110 t/t (~9.1 t DM feed per tonne retail pork)

  • Chicken meat from grain: ~0.226 t/t (~4.4 t DM feed per tonne retail chicken)

Note: Carcass-to-retail conversion increases feed requirements per kg retail meat by ~33-50% compared to per kg carcass, reflecting bone removal and trimming losses.

This structure allows modeling different production systems for the same product (grass-fed vs. grain-finished beef, pasture vs. intensive dairy, etc.).

Regional Feed Energy Requirements

Feed requirements vary significantly by region due to differences in production systems, genetics, and environmental conditions. Wirsenius (2000) [1] provides estimated feed energy requirements per unit of commodity output:

Feed energy requirements per unit of animal product output (Wirsenius 2000, Table 3.9)

Commodity

Unit

East Asia

East Europe

Latin America & Carib.

North Africa & W. Asia

North America & Oc.

South & Central Asia

Sub- Saharan Africa

West Europe

Cattle milk & cow carcass (MJ per kg whole milk & carcass as-is)

NE_l NE_m NE_g

8.2 2.3 0.46

8.2 1.3 0.45

11 1.9 0.30

12 2.0 0.32

5.3 1.1 0.50

11 2.5 0.32

23 5.4 0.70

5.6 1.3 0.52

Dairy bulls & heifers carcass (MJ per kg carcass as-is)

NE_m NE_g

187 22

47 14

143 24

130 21

53 16

344 19

211 20

41 16

Beef carcass (MJ per kg carcass as-is)

NE_m NE_g

288 25

141 19

236 28

262 23

109 23

479 20

352 21

103 23

Pig carcass (MJ per kg carcass-side as-is)

ME

86

84

131

86

65

115

123

64

Eggs & hen carcass (MJ per kg whole egg & carcass as-is)

ME

43

42

39

43

32

53

56

30

Meat-type chicken carcass (MJ per kg eviscerated carcass as-is)

ME

60

56

51

61

42

72

77

38
Energy types:

  • NE_l: Net energy for lactation (dairy production)

  • NE_m: Net energy for maintenance (basic metabolism)

  • NE_g: Net energy for growth (body mass gain)

  • ME: Metabolizable energy (for monogastrics)

Notes:

  • Values calculated from productivity estimates in Wirsenius (2000) Table 3.8

  • Regional variation reflects differences in production systems, breed genetics, climate, and management practices

  • Sub-Saharan Africa shows significantly higher requirements due to less intensive production systems

  • North America and Western Europe have lowest requirements, reflecting highly optimized industrial systems

Baseline Feed Intake

To ground the livestock module in observed feed flows, the model constructs a country-level baseline from GLEAM 2.0 (Mottet et al. 2017 [4]) that describes how much dry-matter feed of each category each animal product consumed in the reference year. In validation mode this baseline can pin the model to observed feed mixes; in optimisation runs it serves as a reference point for comparison with solved results.

The script workflow/scripts/prepare_feed_baseline.py produces processing/{name}/feed_baseline_uncalibrated.csv.

Data Sources

  • GLEAM 2.0 SI Table 2 (data/curated/gleam_tables/gleam_2_0_si2_global_livestock_feed_intake.csv): Global dry-matter feed intake by species (Cattle & buffaloes, Small Ruminants, Poultry, Pigs), production system (Grazing, Mixed, Feedlots, Layers, Broilers, Backyard, Intermediate, Industrial), and feed type (Roughages, Cereal grains, Brans, Soybean cakes, Oil seed cakes, Other edible, Other non-edible, Swill). Totals are reported separately for OECD and Non-OECD country groups.

  • GLEAM 2.0 SI Tables 4–5 (gleam_2_0_si4_dairy_cattle_composition.csv, gleam_2_0_si5_beef_cattle_composition.csv): Percentage breakdown of ruminant roughage into components (fresh grass, hay, legumes & silage, crop residues, etc.) by GLEAM region, used to decompose the aggregate “Roughages” entry into model-specific feed pools.

  • FAOSTAT QCL: National animal product output for 2010 (the GLEAM reference year), the model’s configured reference year (validation.production_year), and the calibration year (validation.gleam_calibration_year). Used to disaggregate GLEAM totals to individual countries, scale the baseline forward in time, and calibrate the efficiency correction.

  • Wirsenius (2000) [1]_ feed energy requirements: Regional metabolizable energy demand per unit of product output, used to split feed between co-products in multi-product production systems.

Disaggregation Pipeline

Global GLEAM totals are converted to a country × product × feed-category matrix through seven sequential steps.

Step 1 — Country disaggregation

GLEAM Table 2 reports totals for OECD and Non-OECD groups. Each country is assigned to one group, and its share of the group’s total species-level output in 2010 (from FAOSTAT) determines how much of the group’s feed is allocated to it:

\[\begin{split}\begin{aligned} \text{share}_{c,s} &= \frac{\text{production}_{c,s,2010}} {\sum_{c' \in \text{group}(c)} \text{production}_{c',s,2010}} \\[6pt] \text{intake}_{c,s,f} &= \text{intake}_{\text{GLEAM},\,\text{group}(c),s,f} \times \text{share}_{c,s} \end{aligned}\end{split}\]

Step 2 — Product split for multi-product systems

Several GLEAM systems serve more than one model product simultaneously: cattle Grazing and Mixed systems supply dairy, dairy-buffalo, and meat-cattle; poultry Backyard systems supply both eggs and meat-chicken. Feed is allocated between co-products in proportion to their energy demand, using ME requirements from Wirsenius (2000):

\[\text{product_share}_{p} = \frac{\text{production}_{c,p} \times \text{FCR}_{c,p}} {\sum_{p'} \text{production}_{c,p'} \times \text{FCR}_{c,p'}}\]

Countries with no reported production for a co-product fall back to equal sharing.

Step 3 — Roughage decomposition

GLEAM Table 2 lumps all ruminant roughage into a single “Roughages” entry. SI Tables 4 and 5 supply regional percentage breakdowns of this roughage into specific components. These percentages are applied to country-level roughage totals and the components are mapped to model feed categories:

Roughage component

Model feed category

Fresh grass, Hay

ruminant_forage

Legumes and silage

ruminant_forage

Crop residues, Sugarcane tops

ruminant_roughage

Leaves (tree leaves/browse)

ruminant_roughage (tracked as exogenous)

Dairy and dairy-buffalo products use Table 4 (dairy cattle composition); meat-cattle and meat-sheep use Table 5 (beef cattle composition).

Tree leaves and forest browse are mapped to ruminant_roughage but tracked separately via the exogenous_mt_dm column, since the model has no endogenous production route for these feeds.

Step 4 — Mapping remaining GLEAM feed types

Swill (food waste recycled as animal feed) is mapped to monogastric_low_quality for monogastrics and ruminant_grain for ruminants, with the full amount marked as exogenous since swill is not produced endogenously by the model.

Feed types other than “Roughages” and “Swill” (handled above) are mapped to model feed categories via a two-step chain: each GLEAM SI2 feed type is first mapped to a representative model feed item, and that item’s category is then looked up from the authoritative ruminant_feed_mapping.csv / monogastric_feed_mapping.csv produced by the categorize_feeds rule. This avoids a second hardcoded source of truth for feed categorisation.

GLEAM feed type

Rum. item

Rum. category

Mono. item

Mono. category

Cereal grains

maize

ruminant_grain

maize

monogastric_grain

2nd grade grain

—

—

maize

monogastric_grain

Brans, spent brewer & biofuel grains

wheat-bran

ruminant_grain

wheat-bran

monogastric_low_quality

Soybean cakes

oilseed-meal

ruminant_protein

oilseed-meal

monogastric_protein

Other oil seed cakes

rapeseed-meal

ruminant_protein

rapeseed-meal

monogastric_protein

Other edible

sugarbeet

ruminant_grain

cassava

monogastric_grain

Other non-edible

barley

ruminant_grain

wheat-bran

monogastric_low_quality

Step 5 — Scaling to the reference year and normalization

The GLEAM baseline reflects 2010 conditions. Country- and product-level feed intakes are scaled to the configured reference year using FAO production trends:

\[\text{feed}_{\text{ref}} = \text{feed}_{2010} \times \frac{\text{production}_{\text{ref}}} {\text{production}_{2010}}\]

After scaling, totals within each OECD/Non-OECD group are normalized to match the scaled GLEAM group total, correcting for countries with incomplete FAOSTAT coverage.

Step 6 — Efficiency calibration

The production-based scaling in Step 5 assumes constant feed conversion efficiency, but efficiencies improved between 2010 and the reference year. GLEAM 3.0 (FAO 2023 [5]) reports a global feed total of approximately 6.2 Gt DM for its 2015 baseline, whereas naively scaling the 6.0 Gt DM GLEAM 2.0 total (2010) by production growth predicts a substantially higher figure—roughly 6.7 Gt for 2015.

The pipeline calibrates against this known data point. First, a constant-efficiency prediction for the calibration year is computed using species-level production growth from FAOSTAT:

\[\hat{T}_{\text{cal}} = \sum_s T_{s,2010} \times \frac{\text{production}_{s,\text{cal}}} {\text{production}_{s,2010}}\]

The ratio of the known GLEAM 3.0 total to this naive prediction yields the cumulative efficiency improvement at the calibration year. Assuming a constant annual rate, the correction for the reference year is:

\[\begin{split}\begin{aligned} r &= \left( \frac{T_{\text{known}}}{\hat{T}_{\text{cal}}} \right)^{1/(\text{cal_year} - 2010)} \\[6pt] \text{correction} &= r^{\,\text{ref_year} - 2010} \end{aligned}\end{split}\]

All feed values are multiplied by this correction factor. The two configuration keys validation.gleam_calibration_year (default: 2015) and validation.gleam_calibration_total_gt_dm (default: 6.2) control the calibration data point.

Step 7 — Production-based feed scaling

After scaling to the reference year and applying the efficiency calibration, the pipeline rescales feed quantities per (country, product) so that the implied animal output matches FAOSTAT production data. This preserves GLEAM’s feed composition (the relative split across feed categories) while correcting absolute feed levels to observed production:

\[\begin{split}\begin{aligned} \text{implied}_{c,p} &= \sum_f \text{feed}_{c,p,f} \times \text{efficiency}_{c,p,f} \\[6pt] \text{scale}_{c,p} &= \frac{\text{FAOSTAT}_{c,p}} {\text{implied}_{c,p}} \\[6pt] \text{feed_scaled}_{c,p,f} &= \text{feed}_{c,p,f} \times \text{scale}_{c,p} \end{aligned}\end{split}\]

The efficiencies used here are the uncalibrated values from feed_to_animal_products_uncalibrated.csv. Scale factors outside the range [0.3, 3.0] are logged as potential data inconsistencies in the GLEAM disaggregation (e.g., a mismatch between GLEAM’s regional feed totals and FAOSTAT’s country-level production data). If FAOSTAT reports zero production for a (country, product) pair, all feed is set to zero; if the implied production is zero but FAOSTAT is positive, the scale factor defaults to 1.0 and a warning is logged.

Output

processing/{name}/feed_baseline_uncalibrated.csv contains one row per (country, product, feed category) combination:

  • country: ISO 3166-1 alpha-3 country code

  • product: Animal product (dairy, dairy-buffalo, meat-cattle, meat-sheep, eggs, meat-chicken, meat-pig)

  • feed_category: Feed pool (e.g., ruminant_forage, ruminant_grain, monogastric_protein)

  • feed_use_mt_dm: Dry-matter feed consumption (Mt DM) in the reference year

  • exogenous_mt_dm: Portion of feed demand that must be supplied exogenously (Mt DM) — tree leaves/browse for ruminants and swill for monogastrics

All (country, product, feed category) combinations are always present, including zeros, so every animal production link in the model has an explicit baseline entry.

When feed efficiency calibration is enabled (see Calibration), the calibrated baseline is written to processing/{name}/feed_baseline.csv and the calibrated efficiencies to processing/{name}/feed_to_animal_products.csv.

Model Integration

The baseline participates in the model in two distinct ways:

  1. Validation mode: When validation.enforce_baseline_feed: true, the baseline feed quantities are imposed as equality constraints on the model, fixing the feed mix to GLEAM-derived estimates. This removes a degree of freedom and is useful for diagnosing supply-side inconsistencies—any imbalance shows up as feed slack (see Feed Breakdown).

  2. Optimisation mode: The model is free to choose any feed mix within the bounds set by available supply and feed conversion efficiencies. The baseline is not enforced but is available for post-hoc comparison with solved solutions.

Exogenous Feed Supply

Some GLEAM feed types have no endogenous supply route in the model:

  • Tree leaves/browse (~100–150 Mt DM globally): Forest browse consumed by ruminants, mapped to ruminant_roughage.

  • Swill (~75 Mt DM globally): Food waste recycled as pig/poultry feed, mapped to monogastric_low_quality (or ruminant_grain for ruminants).

These are tracked in the exogenous_mt_dm column of the baseline and supplied via exogenous_feed generators on the corresponding feed buses (named supply:exogenous_{category}:{country}).

In validation mode, these generators are fixed at the baseline amount (forced dispatch). In optimisation mode, they are extendable up to the baseline amount at zero marginal cost, allowing the solver to use them if beneficial but not requiring it.

Calibration

The uncalibrated baseline and feed efficiencies inevitably produce supply–demand gaps when tested in the full model: missing feed sources, regional mismatches between grassland output and ruminant forage demand, and approximate feed conversion efficiencies all contribute. The calibration pipeline uses a validation solve to diagnose these gaps via slack variables and compute corrections. Two independent calibration systems address different gap sources.

Feed Efficiency Calibration

Feed efficiency calibration adjusts feed conversion efficiencies upward in (country, feed-category) combinations where the validation solve reveals positive feed slack — i.e. where the model cannot supply enough feed through its endogenous routes.

Algorithm (implemented in workflow/scripts/compute_feed_efficiency_calibration.py):

  1. Solve the model with uncalibrated inputs and validation.enforce_baseline_feed: true. The solved model contains slack_positive_feed generators on each feed bus.

  2. Extract positive feed slack per (country, feed_category) from the solved network.

  3. Compute the calibration multiplier:

    \[\begin{split}\begin{aligned} \text{supply}_{c,f} &= \text{baseline}_{c,f} - \text{slack}_{c,f} \\[6pt] \text{multiplier}_{c,f} &= \min\!\left( \frac{\text{baseline}_{c,f}}{\text{supply}_{c,f}},\; \text{max_multiplier} \right) \end{aligned}\end{split}\]

    where \(\text{max_multiplier}\) (default 2.0) caps extreme adjustments.

  4. Apply to efficiencies: \(\text{efficiency}_\text{cal} = \text{efficiency}_\text{uncal} \times \text{multiplier}\)

  5. Adjust baseline feed quantities to preserve implied production:

    \[\text{adj}_{c,p} = \frac{\sum_f \text{feed}_{c,p,f} \times \text{eff}_\text{uncal}} {\sum_f \text{feed}_{c,p,f} \times \text{eff}_\text{cal}}\]

    This ensures that the calibrated baseline still implies the same animal output as the uncalibrated version.

The calibration step is carried out by workflow/scripts/apply_feed_calibration.py, which writes the calibrated files feed_baseline.csv and feed_to_animal_products.csv.

Configuration:

animal_products:
  feed_efficiency_calibration:
    enabled: true          # Apply pre-computed calibration
    generate: false        # Set true to trigger generation from a solve
    source: "data/curated/feed_efficiency_calibration.csv"
    max_multiplier: 2.0    # Cap on per-(country, category) multiplier
    scenario: "default"    # Scenario whose solved model is used

The figure below should be read as a diagnostic view of where feed efficiency assumptions were too strict in the uncalibrated system. Each dot is a (country, feed-category) pair that needed an upward adjustment (multiplier != 1), with marker size proportional to that country’s animal production volume in the category. The right-side x/total labels show how many countries in each category were adjusted.

Strip plot of feed efficiency calibration multipliers by feed category

Feed efficiency calibration multipliers by category and country. The reference line at 1.0 marks “no adjustment”; values to the right indicate upward efficiency corrections required to remove positive feed slack in the validation solve.

Grassland Forage Calibration

Grassland forage calibration addresses mismatches between grassland output and ruminant forage demand that persist after feed efficiency calibration. In some countries grassland yields exceed forage demand (surplus), while in others demand exceeds what grasslands can supply (deficit).

Algorithm (implemented in workflow/scripts/compute_grassland_calibration.py):

  1. Solve the model with calibrated feed efficiencies but uncalibrated grassland parameters.

  2. Per country, extract forage slack from feed:ruminant_forage:{country} buses:

    • Negative slack (surplus): grassland output exceeds forage demand. Reduce effective yield:

      \[\text{yield_correction}_c = \frac{\max(0,\; \text{output}_c - \text{surplus}_c)}{\text{output}_c}\]

    • Positive slack (deficit): forage demand exceeds grassland output. Supply the gap via an exogenous forage source:

      \[\text{exogenous_forage}_c = \text{deficit}_c\]

  3. yield_correction (in [0, 1]) scales grassland yields downward; exogenous_forage_mt_dm adds an external forage supply for countries where grassland alone cannot meet ruminant forage requirements.

Configuration:

grazing:
  grassland_forage_calibration:
    enabled: true          # Apply pre-computed calibration
    generate: false        # Set true to trigger generation from a solve
    source: "data/curated/grassland_forage_calibration.csv"
    scenario: "default"    # Scenario whose solved model is used

This second figure complements the feed-efficiency plot by separating grassland-specific calibration effects. yield_correction is one-sided: values run from 0 to 1 and only reduce effective grassland yield (never increase it). Countries with forage deficits are additionally flagged via hatching, indicating reliance on exogenous_forage_mt_dm after endogenous grassland supply is exhausted.

Map of grassland forage yield correction by country

Grassland forage calibration by country. Colour indicates the yield_correction factor applied to grassland yields (1.0 = no adjustment; lower values = stronger downward correction). Countries receiving exogenous forage (exogenous_forage_mt_dm > 0) are marked with hatching.

Taken together, the two figures show that calibration corrects two distinct issues: feed conversion mismatches across feed categories (first figure) and country-level forage balance mismatches in grassland systems (second figure).

Calibration Pipeline

The two calibration systems run sequentially within the Snakemake workflow. The pipeline avoids circular dependencies by defining dedicated scenarios whose effective configuration disables the calibration being computed:

  1. Phase 1 — Feed efficiency calibration: The uncalibrated scenario sets feed_efficiency_calibration.enabled: false and grassland_forage_calibration.enabled: false. Building and solving this scenario produces the uncalibrated model from which feed slack is extracted. compute_feed_efficiency_calibration writes the multipliers to the configured source path (by default data/curated/feed_efficiency_calibration.csv).

  2. Phase 2 — Grassland forage calibration: The feed_calibrated scenario applies feed efficiency calibration but sets grassland_forage_calibration.enabled: false. Building and solving this scenario produces a model with calibrated efficiencies from which forage slack is extracted. compute_grassland_calibration writes yield corrections and exogenous forage amounts to the configured source path (by default data/curated/grassland_forage_calibration.csv).

  3. Phase 3 — Calibrated model: All other scenarios (including default) apply both calibrations. apply_feed_calibration reads the pre-computed multipliers and writes the calibrated feed_baseline.csv and feed_to_animal_products.csv.

Pre-computed calibration files are stored under data/curated/ so they can be reused across configurations without re-running the validation solves. Set generate: true in the relevant configuration block to re-generate them (requires a full validation solve).

Model Implementation

In workflow/scripts/build_model.py, livestock production is represented as multi-bus links:

Crop Residue Feed Supply

Crop residues (e.g., straw, stover, pulse haulms) are now generated explicitly using the new Snakemake rule build_crop_residue_yields:

  • Configuration: Select residue crops via animal_products.residue_crops in config/default.yaml. Only crops present in config.crops are processed.

  • Data sources: - GLEAM Supplement S1 Table S.3.1 (slope/intercept) and Tables 3.3 / 3.6 (FUE factors) - GLEAM feed codes → model mapping in data/curated/gleam_feed_mapping.csv

  • Outputs: Per-crop CSVs at processing/{name}/crop_residue_yields/{crop}.csv with net dry-matter residue yields (t/ha) by region, resource class, and water supply.

  • Integration: build_model reads all residue CSVs, adds residue_{feed_item}_{country} buses, and attaches them as additional outputs on crop production links. Residues flow through the same feed supply logic as crops/foods and enter the appropriate feed pools or soil incorporation.

Residue Removal Limits for Feed

To maintain soil health and prevent land degradation, the model constrains the fraction of crop residues that can be removed for animal feed. The majority of residues must be left on the field and incorporated into the soil to maintain organic matter and nutrient cycling.

Constraint formulation:

  • Maximum feed removal: 30% of generated residues (configurable via residues.max_feed_fraction; override per ISO3 country or M49 region/sub-region via residues.max_feed_fraction_by_region with country > sub-region > region)

  • Minimum soil incorporation: 70% of generated residues

The optimization model implements this as a constraint between residue feed use and soil incorporation for each residue type and country:

\[\text{feed use} \leq \frac{\text{max feed fraction}}{1 - \text{max feed fraction}} \times \text{incorporation}\]

With the default 30% limit:

\[\text{feed use} \leq \frac{3}{7} \times \text{incorporation}\]

This ensures that for every 3 units of residue used as feed, at least 7 units are incorporated into the soil. The constraint is applied during model solving (in solve_model.py) after the network structure is built.

Environmental implications: Residues incorporated into soil generate direct N₂O emissions according to the IPCC EF1 emission factor applied to their nitrogen content (see Environmental Impacts). The model therefore balances:

  • Feed benefits: Residues reduce demand for dedicated feed crops (reducing land use and associated emissions)

  • Soil incorporation costs: Incorporated residues produce N₂O emissions but maintain soil health

Emissions from Livestock

Livestock production generates significant greenhouse gas emissions from two primary sources:

  • Enteric fermentation (CH₄): Ruminants produce methane through digestive fermentation

  • Manure management (CH₄, N₂O): All livestock produce emissions from manure storage and handling

For detailed methodology, data sources, and IPCC calculations, see Environmental Impacts (sections on Enteric Fermentation (CH₄) and Manure Management (CH₄)).

Enteric Fermentation (CH₄)

Ruminants (cattle, sheep) produce methane through digestive fermentation. The model uses IPCC Tier 2 methodology based on methane yields (MY) per unit dry matter intake (DMI).

Summary

  • Enteric fermentation produces CH₄ in ruminants during digestion

  • Methane yield (MY) varies by feed quality (roughage > forage > grain > protein)

  • Model uses IPCC Tier 2 methodology with feed-specific emission factors

  • See Enteric Fermentation (CH₄) for full details

Data Sources

  • data/curated/ipcc_enteric_methane_yields.csv: IPCC methane yields by feed category

  • processing/{name}/ruminant_feed_categories.csv: Feed categories with MY values (generated from GLEAM 3.0 data)

Manure Management (CH₄, N₂O)

All livestock produce emissions from manure storage, handling, and application:

  • CH₄: From anaerobic decomposition (especially liquid systems like lagoons)

  • N₂O: From nitrogen in manure (direct and indirect emissions)

Manure CH₄ emissions are calculated for all animal products (ruminants and monogastrics) and combined with enteric emissions in the model. See Manure Management (CH₄) for full methodology.

Production Costs

The model incorporates livestock production costs to represent the economic considerations of animal farming beyond feed and land costs. Costs include labor, veterinary services, energy, housing, and other operational expenses, while excluding feed (modeled endogenously) and land rent (implicit opportunity cost).

Livestock costs are applied as marginal costs on feed-to-product conversion links in the optimization model. The costs are sourced from USDA (United States) and FADN (European Union) agricultural accounting data, processed to per-tonne product costs, then converted to per-tonne feed costs using feed conversion efficiencies.

Grazing costs are handled separately from general livestock production costs. These costs represent the economic expenses specific to pasture-based feed production and are applied directly to grassland feed links rather than animal production links.

For comprehensive details on production cost data sources, processing methodology, and model application, see:

  • Production Costs - Complete documentation of all production costs (crops, livestock, and grazing)

The livestock-specific sections include:

  • Data sources: USDA and FADN livestock cost data

  • Processing methodology: Allocation by output value, yield calculations, and unit conversions

  • Grazing costs: Separation, processing, and application to grassland feed

  • Model application: How costs are applied as marginal costs on production links

Quick reference for livestock cost workflow:

  • retrieve_usda_animal_costs: Processes USDA livestock cost data (US)

  • retrieve_fadn_animal_costs: Processes FADN livestock cost data (EU)

  • merge_animal_costs: Combines sources and applies fallback mappings

  • Output: processing/{name}/animal_costs.csv with columns:

    • product: Animal product name

    • cost_per_mt_usd_{base_year}: Production cost excluding grazing (USD/tonne)

    • grazing_cost_per_mt_usd_{base_year}: Grazing-specific cost (USD/tonne)

Configuration Parameters

animal_products:
  include:
  - meat-cattle
  - meat-pig
  - meat-chicken
  - dairy
  - eggs
  - dairy-buffalo
  - meat-sheep
  # Feed conversion efficiency mode (how much feed is required per unit product)
  # Source: Wirsenius (2000) regional feed energy requirements
  # Options:
  #   - List of regions: average efficiencies across those regions (all countries use same values)
  #   - null: use country-specific regional efficiencies based on geographic mapping
  # Available regions: East Asia, East Europe, Latin America & Caribbean,
  #   North Africa & West Asia, North America & Oceania, South & Central Asia,
  #   Sub-Saharan Africa, West Europe
  feed_efficiency_regions:
  - North America & Oceania
  - West Europe
  # Ruminant net-to-metabolizable energy conversion efficiency factors
  # Used to convert net energy (NE) requirements to metabolizable energy (ME) requirements
  # Based on NRC (2000) typical values for mixed diets
  # ME_required = NE_m/k_m + NE_g/k_g (+ NE_l/k_l for dairy)
  # TODO: Should check the reference for this.
  net_to_metabolizable_energy_conversion:
    k_m: 0.60  # Maintenance efficiency
    k_g: 0.40  # Growth efficiency
    k_l: 0.60  # Lactation efficiency (dairy)
  # Carcass-to-retail meat conversion factors
  carcass_to_retail_meat:
    meat-cattle: 0.67  # kg boneless retail beef per kg carcass (OECD-FAO 2023)
    meat-pig: 0.73     # kg boneless retail pork per kg carcass (OECD-FAO 2023)
    meat-chicken: 0.60 # kg boneless retail chicken per kg carcass (OECD-FAO 2023)
    eggs: 1.00         # No conversion needed (whole egg = retail product)
    dairy: 1.00        # No conversion needed (whole milk = retail product)
    meat-sheep: 0.63   # kg boneless retail lamb per kg carcass (slightly lower than beef)
    dairy-buffalo: 1.00 # No conversion needed (whole milk = retail product)
  # FAOSTAT QCL item names to aggregate for each model product.
  # First item is the primary product; additional items are proxied species
  # whose production is lumped into the model product.
  faostat_items:
    dairy:
      - "Raw milk of cattle"
      - "Raw milk of goats"       # proxy: goat milk → dairy
      - "Raw milk of sheep"       # proxy: sheep milk → dairy
      - "Raw milk of camel"       # proxy: camel milk → dairy
    meat-cattle:
      - "Meat of cattle with the bone, fresh or chilled"
      - "Meat of buffalo, fresh or chilled"   # proxy: buffalo → cattle
    meat-pig:
      - "Meat of pig with the bone, fresh or chilled"
    meat-chicken:
      - "Meat of chickens, fresh or chilled"
      - "Meat of ducks, fresh or chilled"            # proxy: duck → chicken
      - "Meat of turkeys, fresh or chilled"           # proxy: turkey → chicken
      - "Meat of pigeons and other birds n.e.c., fresh, chilled or frozen"
    eggs:
      - "Hen eggs in shell, fresh"
    dairy-buffalo:
      - "Raw milk of buffalo"
    meat-sheep:
      - "Meat of sheep, fresh or chilled"
      - "Meat of goat, fresh or chilled"              # proxy: goat → sheep
  feed_proxy_map:
    dairy-buffalo: dairy
    meat-sheep: meat-cattle
  feed_efficiency_calibration:
    enabled: true
    generate: false
    source: "data/curated/feed_efficiency_calibration.csv"
    max_multiplier: 2.0
    scenario: "default" # Which scenario's solved model to generate calibration from
  residue_crops:
  - banana
  - barley
  - chickpea
  - cowpea
  - dry-pea
  - dryland-rice
  - foxtail-millet
  - gram
  - maize
  - oat
  - pearl-millet
  - phaseolus-bean
  - pigeonpea
  - rye
  - sorghum
  - sugarcane
  - wetland-rice
  - wheat

grazing:
  enabled: true
  isimip_utilization_rate: 0.50 # Applied to ISIMIP yields in merge step
  forage_overlap_subtraction_alpha: 0.0 # Subtract alpha * modeled forage-crop supply from LUIcube grass supply by country
  forage_overlap_crops:
  - alfalfa
  - silage-maize
  - biomass-sorghum
  grassland_forage_calibration:
    enabled: true
    generate: false
    source: "data/curated/grassland_forage_calibration.csv"
    scenario: "default"

Disabling grazing (enabled: false) forces all animal products to come from feed-based systems or imports, useful for exploring intensification scenarios.

Workflow Rules

build_grassland_yields

  • Input: ISIMIP grassland yield NetCDF, resource classes, regions

  • Output: processing/{name}/grassland_yields.csv

  • Script: workflow/scripts/build_grassland_yields.py

prepare_feed_baseline

  • Input: GLEAM SI tables, FAOSTAT QCL, feed mappings, uncalibrated efficiencies

  • Output: processing/{name}/feed_baseline_uncalibrated.csv

  • Script: workflow/scripts/prepare_feed_baseline.py

apply_feed_calibration

  • Input: Uncalibrated baseline and efficiencies, calibration multipliers

  • Output: processing/{name}/feed_baseline.csv, processing/{name}/feed_to_animal_products.csv

  • Script: workflow/scripts/apply_feed_calibration.py

compute_feed_efficiency_calibration

  • Input: Solved network (uncalibrated scenario), uncalibrated baseline and efficiencies

  • Output: data/curated/feed_efficiency_calibration.csv

  • Script: workflow/scripts/compute_feed_efficiency_calibration.py

compute_grassland_calibration

  • Input: Solved network (feed-calibrated scenario)

  • Output: data/curated/grassland_forage_calibration.csv

  • Script: workflow/scripts/compute_grassland_calibration.py

Livestock production is then integrated into the build_model rule using the grassland yields and feed conversion CSVs.

References