Production Costs

Overview

The model incorporates production costs for both crop and livestock systems to represent the economic considerations of agricultural production. Costs are applied as marginal costs on production links in the PyPSA network, ensuring that the optimization accounts for both physical and economic constraints.

This page provides an overview of how production costs are sourced, processed, and applied throughout the model.

Cost Categories

The model distinguishes between three main categories of production costs:

Crop Production Costs

Costs associated with growing crops, including labor, machinery, energy, and other inputs (excluding fertilizer, which is modeled endogenously).

Livestock Production Costs

Costs associated with raising animals for meat, milk, and eggs, including labor, veterinary services, housing, and energy (excluding feed and land, which are modeled endogenously).

Grazing Costs

Costs specifically associated with pasture-based livestock production, representing the management and maintenance of grassland feed systems.

Land Conversion Costs

Investment costs for expanding agriculture onto new land, covering physical clearing and soil preparation. Differentiated by forest vs. non-forest cover type, annualized using a capital recovery factor over a configurable investment horizon. Applied as marginal costs on land_conversion and new_to_pasture links. See Land Use & Resource Classes for details.

Marketing Costs (Farm-to-Wholesale)

The markup between the farm gate and the wholesale (inter-regional trade) market. Covers drying, on-farm and commercial storage, first-mile transport, elevator / aggregator handling, slaughter and packing for meat, processing for foods and feeds, insurance, and the trader margin. Applied as a per-tonne markup on the relevant production link (crop_production, food_processing, feed_conversion, animal_production). See Marketing Costs (Farm-to-Wholesale) below for the per-class table.

What Costs Include and Exclude

Included Costs

Production costs in the model capture the following expense categories:

  • Labor: Both hired labor and the opportunity cost of unpaid/family labor

  • Veterinary services: Animal health care and preventive treatments (livestock only)

  • Energy: Electricity and fuel for farm operations

  • Machinery and equipment: Depreciation and maintenance

  • Housing and facilities: Depreciation of buildings and infrastructure (livestock only)

  • Interest on operating capital: Financial costs of production

  • Other variable inputs: Seeds, pesticides, and other operational expenses (crops only)

Excluded Costs (Modeled Endogenously)

The following cost categories are excluded from the production cost data because they are represented explicitly in the optimization model:

  • Feed costs: Crop and residue feed inputs are modeled as network flows with their own costs

  • Fertilizer costs: Synthetic fertilizer is a constrained resource in the model

  • Land costs and rent: Land opportunity cost is implicit in the land allocation decisions

  • Grazing feed costs (for livestock production costs): Grassland feed is modeled separately with its own grazing costs

This separation ensures that costs are not double-counted while maintaining accurate economic representation.

Data Sources

Crop Costs: FAOSTAT Producer Prices

Crop production costs are derived from FAOSTAT data, using producer prices as a revenue proxy scaled by a configurable cost share.

Coverage:

  • Spatial: Per-(crop, country) for all modeled countries

  • Temporal: Configurable averaging period (default 2015-2022)

  • Crops: All crops with FAOSTAT price and yield data; unmapped crops use proxy mappings from data/curated/faostat_cost_proxies.yaml

Data characteristics:

  • Prices from the FAOSTAT Prices (PP) domain (element 5532: Producer Price USD/tonne)

  • Yields from the FAOSTAT Production (QCL) domain (element 5412: Yield kg/ha)

  • CPI-deflated to the configured base year before averaging

Source:

Workflow script: workflow/scripts/prepare_faostat_crop_costs.py

Livestock Costs: USDA and FADN

Livestock production costs are sourced from two agricultural accounting systems.

USDA Economic Research Service (United States):

  • Coverage: Dairy (milk), beef cattle (cow-calf), hogs (pork)

  • Time period: 2015-2024 (averaged across years)

  • Units: USD per acre or USD per head

  • Workflow script: workflow/scripts/retrieve_usda_animal_costs.py

FADN - Farm Accountancy Data Network (European Union):

  • Coverage: All major animal production systems

  • Time period: 2004-2020 (averaged across years)

  • Units: EUR per farm (allocated to livestock categories)

  • Workflow script: workflow/scripts/retrieve_fadn_animal_costs.py

Cost Processing Methodology

The cost data undergoes several processing steps to ensure consistency and accuracy across different sources and production systems.

Crop Costs

Crop production costs are derived from FAOSTAT producer prices and yields, providing per-(crop, country) cost estimates.

Script: workflow/scripts/prepare_faostat_crop_costs.py

The cost model uses revenue per hectare as a proxy for total production cost, scaled by a configurable share:

\[C_{\mathrm{ha}} = P \times Y \times f_{\mathrm{non\text{-}endog}}\]

where \(P\) is the producer price (USD/t), \(Y\) is yield (t/ha), and \(f_{\mathrm{non\text{-}endog}}\) is the non-endogenous cost share (default 0.7).

Processing steps:

  1. Load FAOSTAT bulk data: Producer prices from the PP domain (element 5532, USD/tonne) and yields from the QCL domain (element 5412, kg/ha, converted to t/ha)

  2. Map crops to FAOSTAT items: Using data/curated/faostat_crop_item_map.csv; crops without a direct mapping use proxy crops defined in data/curated/faostat_cost_proxies.yaml (e.g., alfalfa uses soybean prices, biomass-sorghum uses sorghum)

  3. CPI deflation: Prices are deflated to the configured base year using US CPI-U data

  4. Merge price and yield: For each (crop, country, year), compute revenue per hectare = price × yield

  5. Temporal averaging: Average revenue per hectare across the configured period (default 2015-2022)

  6. Apply cost share: Multiply by non_endogenous_cost_share (default 0.7) to obtain the cost estimate

  7. Winsorize per-crop outliers: For each crop, clip non-fallback per-tonne cost (cost_per_ha / yield_per_ha) above the configured outlier_cap_quantile (default p90 of the non-fallback distribution) to that quantile, then recompute per-hectare cost as capped_per_tonne * actual_yield. Capping per-tonne rather than per-hectare preserves elevated implicit prices in high-yield greenhouse producers while bounding them at realistic wholesale levels. See Outlier Cap (Greenhouse / Cold-Climate Producers) below. Capped rows are flagged via the is_capped audit column. Set outlier_cap_quantile: null to disable the cap.

  8. Fill gaps: Missing (crop, country) pairs receive the global median cost for that crop (computed from the post-cap distribution) as fallback.

  9. Output: processing/{name}/faostat_crop_costs.csv with columns:

    • crop: Crop name

    • country: ISO3 country code

    • cost_usd_{base_year}_per_ha: Production cost estimate (USD/ha)

    • n_years: Number of years with valid price-yield data

    • is_fallback: Whether the value is a global median fallback

    • is_capped: Whether the value was winsorized in step 7

Outlier Cap (Greenhouse / Cold-Climate Producers)

FAOSTAT producer prices and yields are reported on a country basis and make no distinction between field and protected (greenhouse) cultivation. For a small group of high-value crops (tomato, carrot, mango, tea, cabbage, apple, banana, watermelon, sweet-potato) in cold-climate or high-income producers (ISL, NOR, DNK, FIN, CHE, IRL, SWE, DEU, GBR, NLD, BEL, AUT, JPN, GUY, BRB, etc.), virtually all reported production comes from greenhouses. The combined producer-price * yield product is 1-3 orders of magnitude above the field-cultivation norm: pre-cap tomato cost reaches USD 1.4 million per hectare in Iceland, USD 980k in Norway, USD 200k+ for carrot in Iceland, USD 150k for mango in Japan.

The model treats this country-aggregate value as field cost applied to the model’s notional cropland area, so these outliers feed directly into crop_production link costs. They are not field cost in any useful sense and were forcing the cost calibration to absorb the gap.

The winsorization step is applied to per-tonne cost (cost_per_ha / yield_per_ha, equivalent up to the non_endogenous_cost_share factor to the FAOSTAT producer price): for each crop, non-fallback country values above the configured quantile (default 0.90) of the per-tonne distribution are clipped to that quantile, and per-hectare cost is recomputed as capped_per_tonne * actual_yield_per_ha. The cap is applied before the global median is computed for missing (crop, country) fallback rows, so the post-cap distribution drives both. Fallback rows are themselves never marked as capped.

Capping per-tonne rather than per-hectare matters because the underlying outlier pattern is both high producer prices and high greenhouse yields. Capping per-hectare cost while leaving yields untouched would collapse the implicit per-tonne cost to artificially low values (e.g. tomato in Belgium would drop to ~$700/t against a wholesale-realistic ~$2-3 k/t), which then feeds large positive corrections in the cost calibration and inflates the production-stability L1 penalty. Capping per-tonne preserves the elevated implicit price but bounds it at realistic wholesale levels; per-hectare cost scales with the actual yield.

The resulting cost estimates vary substantially across crops and countries, reflecting differences in local prices, yields, and agricultural productivity. The map below shows the median cost per country (across all crops), while the distribution plot reveals the cross-country spread for each crop individually.

World map of median crop production costs per country

Median crop production cost across all crops per country (USD/ha, log scale). Higher costs in Europe and North America reflect higher input prices and labour costs; lower costs in Sub-Saharan Africa and South Asia reflect lower price levels.

Distribution of crop production costs across countries for each crop

Cross-country cost distributions per crop (USD/ha, log scale). Boxen (letter-value) plots show the full distributional shape; crops are grouped by category. Vegetables and fruits have the highest and most variable costs, while cereals and legumes cluster at lower levels.

Calibration Correction

An optional additive calibration correction adjusts production costs for crops, grassland, and animals based on shadow prices from a two-step paired solve:

  • Step 1 pins consumption to the baseline diet, enables hard production-stability bounds at +/-20 %, and applies the file-level validation.slack_marginal_cost: 5.0 override (5 000 USD/t) to cap the duals of the small set of foods whose FAOSTAT-vs-FBS mismatch still exceeds the +/-20 % band (buckwheat, plantain, coffee, tea, olive-oil). The food-bus duals feed the piecewise consumer-utility blocks used by step 2.

  • Step 2 activates the step-1 piecewise utility and tightens production stability to +/-1 %. The duals on these tight constraints become the per-group additive cost corrections.

The corrections are additive, clipped to zero (no negative costs), and applied at build time whenever cost_calibration.enabled is true (the default).

Regenerate with tools/calibrate cost. See Calibration for the full dependency graph and algorithm, including the upstream food_demand step that closes residual per-food gaps before the cost solve.

Marketing Costs (Farm-to-Wholesale)

Farm-gate production costs alone substantially understate the price a commodity carries when it reaches an inter-regional wholesale market. The intervening marketing margin – drying, storage, first-mile transport, elevator / aggregator handling, slaughter and packing for meat, dairy and oil processing, insurance, and the trader margin – typically accounts for 10-25 % of bulk-grain wholesale price, 15-40 % of fresh-produce wholesale price, and 8-15 % of the meat wholesale value paid to slaughter and packing.

The model captures this as a single per-tonne marketing_cost_per_t parameter per commodity class, configured under the unified commodities: block (see Configuration). The markup is applied as a one-shot per-tonne cost on each commodity’s production link:

  • On crop_production links: marketing_cost_per_t * yield, added to the link’s marginal cost (USD/ha basis).

  • On food_processing links: sum over outputs of marketing_cost_per_t * efficiency, summed across the pathway’s output foods (multi-output Links).

  • On feed_conversion links: marketing_cost_per_t * share for the destination feed category.

  • On animal_production links: marketing_cost_per_t * efficiency for the produced animal product.

The defaults below come from the public agricultural-marketing literature – principally USDA ERS and FAO. They are order of magnitude anchors that the cost-calibration step adjusts locally.

Default marketing-cost parameters (USD_2024 per tonne)

Class

Default marketing_cost_per_t

Composition and source

Coverage

crops.bulk_dry_goods

30

Farm-to-grain-elevator markup for storable bulk commodities: drying (~$3-7/t), shrink (~1.4 % of grain value), commercial storage 3-6 months (~$7-13/t), elevator handling in+out (~$3-11/t), first-mile truck (~$3-6/t). Bottom end of the $9-65/t range reported by Texas A&M Transportation Institute (TTI 2019) for US soybeans farm-to-Gulf, picking the no-delay base case.

cereals, oilseeds, pulses, stimulant crops, cotton fibre

crops.bulky_fresh

60

Twice the bulk-dry-goods markup. Bulky low-value perishables (roots, tubers, sugar crops, biomass) carry a larger handling margin relative to wholesale price because their mass per value is high and shelf life is short. FAO (FAO 1997, chapter 12) places handling in this group near 15-25 % of wholesale price.

roots and tubers (potato, cassava, yam, plantain), sugarbeet, sugarcane, oil-palm, fodder/biomass crops

crops.perishable_high_value

200

Cooling, cold-chain transport, packing, sorting and the wider trader margin pull this class to the 25-40 % share of wholesale price reported by FAO (FAO 1997) and the USDA Food Dollar Series (USDA ERS Food Dollar) farm-share data for fresh produce.

fresh vegetables, fruits, olives

foods.processed_dry_goods

80

Mill / dry-processing margin plus dry-goods wholesale: USDA ERS reports the wholesaling and retailing share of food-at-home spending in the 13-16 % range (USDA ERS Food Dollar), which for a $400-600/t milled-grain or processed-pulse wholesale price puts the marketing layer around $80/t.

milled cereals, dehulled grains, food-grade pulses, whole seeds, sugar, dried stimulants

foods.processed_oils

120

Oil extraction, refining and wholesale margin. Oil mill margins reported by AOCS / industry surveys (USDA WASDE oilseed chapter) are typically 5-10 % of refined-oil wholesale price (~$1100-1500/t), giving ~$60-150/t.

vegetable oils

foods.fresh_produce

200

Same magnitude and same logic as crops.perishable_high_value; most items in this class are simply the crop sold as food.

fresh fruits, vegetables, roots delivered as food

foods.chilled_meat

800

USDA ERS Meat Price Spreads (USDA ERS Meat Price Spreads) gives a 2023 farm-to-wholesale spread for Choice beef of 36.8 cents per pound retail-equivalent (wholesale 452.9 minus gross farm 416.1), i.e. ~$810/t retail-equivalent. Pork and broiler farm-to-wholesale margins fall in the same magnitude once expressed per tonne retail equivalent.

meat-cattle, meat-pig, meat-chicken, meat-sheep

foods.dairy_and_eggs

300

Dairy processing, packaging and cold-chain wholesale margin. USDA ERS reports a 2022 farm share of US dairy retail near 28 % on a ~$5.3/kg basket, which on a per-tonne wholesale basis implies a farm-to-wholesale margin of $200-400/t (USDA ERS Dairy).

dairy, dairy-buffalo, eggs

foods.feed_byproduct

30

Bulk low-value co-products diverted to feed (brans, meals, hulls, gluten products, distillers grains, molasses) carry the same farm-to-wholesale handling cost as bulk grains.

milling and oil-extraction byproducts, sugar molasses

foods.industrial_byproduct

40

Fibre, biofuel and industrial co-products. Cotton lint at the gin-to-warehouse stage adds roughly $30-50/t in handling (USDA AMS cotton classing fees); fuel ethanol and starch carry similar bulk-storage handling margins.

cotton-lint, ethanol products, maize-starch, rendered-fat

feeds.grain_protein

30

Compounded feed mill margin (grinding, mixing, bagging) on grain and protein concentrates. Mirrors the bulk-dry-goods crop markup since these feeds are largely those crops in a different form.

ruminant_grain, ruminant_protein, monogastric_grain, monogastric_protein

feeds.forage

25

Hay-and-forage handling: baling, on-farm or commercial barn storage, short-haul truck. Lower than grain marketing because most forage moves short distances on or near the farm of origin.

ruminant_forage

feeds.bulky_low_quality

35

Bulk roughage and low-quality co-products with higher per-tonne handling cost than baled forage.

ruminant_roughage, monogastric_low_quality

The configuration is the single source of truth – raising any value here moves that marketing layer directly into the optimisation. Every modelled commodity must appear in exactly one class; missing assignments are rejected by workflow/validation/commodities.py before any solve.

Livestock Costs

Livestock production costs follow a similar two-source approach with additional complexity due to the need to convert per-head costs to per-tonne costs.

USDA Livestock Cost Processing

Script: workflow/scripts/retrieve_usda_animal_costs.py

  1. Download Data: Fetch cost and returns data from USDA ERS for each animal product

  2. Filter and Aggregate Costs:

    • Included: Operating costs, allocated overhead, labor (including opportunity cost)

    • Excluded: Feed costs (endogenous), land rent

    • Grazing costs: Separately extracted using grazing_cost_items parameter (e.g., “Grazed feed” line item)

  3. Per-Head Calculation: Sum relevant cost line items to get total cost per animal per year

  4. Physical Yield Data: Use USDA production statistics to get output per head:

    • Milk: Pounds per cow per year → tonnes per head

    • Meat: Live weight → carcass weight → retail meat weight (using USDA conversion factors)

  5. Convert to Per-Tonne Costs:

    \[\text{Cost per tonne product} = \frac{\text{Cost per head (USD/year)}}{\text{Yield (tonnes product/head/year)}}\]

  6. Separate Grazing Costs: Maintain separate column for grazing-specific costs

  7. Inflation Adjustment: Convert to base year USD using CPI

  8. Output: processing/{name}/usda_animal_costs.csv

FADN Livestock Cost Processing

Script: workflow/scripts/retrieve_fadn_animal_costs.py

FADN livestock costs use a sophisticated allocation methodology:

  1. Farm-Level Cost Extraction:

    • Read FADN farm accounting data for livestock-specialized farms

    • Extract total livestock costs (labor, veterinary, energy, depreciation, etc.)

    • Excluded: Feed costs, land rent

    • Grazing costs: Separately identified using grazing_cost_items (SE codes for pasture/grazing)

  2. Allocation to Livestock Categories:

    • Specific costs (veterinary, animal-specific inputs): Allocated by livestock output value share

    • Shared overhead (buildings, energy, general labor): Allocated by livestock sector’s share of total farm output to avoid over-allocation in mixed farms

  3. Normalize to Livestock Units (LU):

    • Convert farm-level costs to per-LU using standard coefficients:

      • 1 Dairy Cow = 1.0 LU

      • 1 Beef Cow = 0.8 LU

      • 1 Pig = 0.3 LU

      • 1 Sheep = 0.1 LU

    • This produces cost per head for each animal category

  4. Physical Yield Calculation:

    • Use FAOSTAT country-level data: Total Production ÷ Total Stocks = Yield per head

    • This captures regional differences in:

      • Slaughter weights and cycles per year (meat)

      • Dairy productivity (milk yield per cow)

      • Herd structure (breeding vs. production animals)

  5. Convert to Per-Tonne Costs (same formula as USDA)

  6. Currency and Inflation: Inflate to base year EUR (HICP), convert to international USD (PPP)

  7. Separate Grazing Costs: Maintain separate column for pasture/grazing-related costs

  8. Output: processing/{name}/fadn_animal_costs.csv

Merging Livestock Costs

Script: workflow/scripts/merge_animal_costs.py

  1. Load Multiple Sources: Combine USDA and FADN livestock cost estimates.

  2. Average Across Sources: For products with multiple data sources, compute mean.

  3. Resolve Fallbacks: For products without direct source data, walk the fallback chain configured under animal_costs in config/default.yaml (see Fallbacks for Products Without Source Data below for the values and sources).

    1. fallback_aliases: copy another product’s per-tonne cost verbatim.

    2. fallback_values_usd_per_t: literature-based defaults with separate non-grazing production and grazing components.

    3. Otherwise: zero cost, with a warning.

  4. Maintain Separate Grazing Costs: Keep grazing cost column distinct from general production costs.

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

    • product: Animal product name

    • n_sources: Number of source datasets averaged (0 when filled by a fallback)

    • source: How the row was resolved – data, alias:<proxy>, literature, or zero

    • cost_per_t_usd_{base_year}: Non-grazing production cost (USD/tonne product)

    • grazing_cost_per_t_usd_{base_year}: Grazing-feed cost (USD/tonne product)

Fallbacks for Products Without Source Data

USDA Commodity Costs and Returns covers cattle, hogs, dairy, and broiler breakouts (live weight), but does not publish ongoing cost-of-production series for sheep or buffalo dairy, and the broiler series alone is US-centric. FADN covers cattle, dairy, and sheep in the EU but is biased toward high-cost systems. To avoid leaving large global products with a zero per-tonne base cost (forcing the calibration to manufacture the entire cost signal), the merge step applies the following defaults.

Buffalo dairy – alias to cow dairy

dairy-buffalo is resolved by copying the cow dairy per-tonne cost verbatim. Buffalo milk’s per-tonne production-cost structure is dominated by feed and labour at scales similar to cow dairy. ICAR-NDRI cost-of-production studies (India) report buffalo milk at roughly INR 25-30/L (about USD 300-360/t) versus cow milk at INR 27-32/L, so cow dairy is a defensible (slightly conservative) upper-bound proxy until product-specific source data are added.

References:

  • ICAR-National Dairy Research Institute, Cost of milk production studies, https://ndri.res.in/

  • Birthal, P.S. et al. (2017), Buffalo Production in India: Performance, Trends and Drivers, NIAP Policy Paper.

Broiler chicken – USD 1 300/t carcass (non-grazing only)

USDA ERS Commodity Costs and Returns: Broilers place broiler operating + ownership cost at roughly USD 0.55-0.65 per pound live weight (about USD 1 200-1 430/t live, USD 1 700-2 000/t carcass at ~70 % dressing) over 2018-2023. Brazil and China together account for more than 40 % of global broiler meat output and produce at lower cost; the OECD-FAO Agricultural Outlook 2024-2033 implied production cost is around USD 1 100-1 500/t carcass. A production-weighted global anchor of USD 1 300/t carcass is used here. Chickens do not graze in the systems represented in the model, so the grazing component is zero.

References:

Sheepmeat (lamb) – USD 3 500/t carcass total

NZ Beef + Lamb Economic Service Sheep and Beef On-farm Inventory weighted-average cost of production runs about NZD 5-6/kg carcass (USD 3 000-3 700/t). MLA Cost of Production – Lamb (Australia) is about AUD 4-5/kg live weight (USD 2 700-3 500/t carcass). UK DEFRA Farm Business Survey lamb is higher (USD 5 500-6 800/t carcass) but the UK is a much smaller share of global production. NZ, AU, CHN and IND dominate global sheepmeat output, so a production-weighted anchor of USD 3 500/t carcass total is used here. Extensive pasture systems are the global norm; the AGRI Benchmark Beef and Sheep Network reports roughly 60-75 % of variable cost as pasture/forage on sheep farms, so the total is split as USD 1 200/t non-grazing operating cost + USD 2 300/t grazed-forage cost (~65 % grazing share).

References:

Grazing Costs

Grazing costs are extracted as a separate component during livestock cost processing and then converted to feed-basis costs for application in the model.

Extraction from Source Data

During USDA and FADN livestock cost processing, grazing costs are identified using configured line items:

USDA: Line items labeled “Grazed feed” or similar in the cost and returns spreadsheets

FADN: SE codes corresponding to pasture management, grassland maintenance

These costs are:

  • Allocated to livestock products by output value share (same methodology as other costs)

  • Stored in separate grazing_cost_per_t_usd_{base_year} column

  • Expressed per tonne of animal product (not per tonne of feed)

Conversion to Feed-Basis Costs

Script: workflow/scripts/build_model/grassland.py (function calculate_grazing_cost_per_tonne_dm)

Since grazing costs in the source data are per tonne of animal product, they must be converted to per tonne of dry matter (DM) feed for application to grassland feed links:

  1. Load Data:

    • Grazing costs per tonne product from animal_costs.csv

    • Feed conversion efficiencies from feed_to_products.csv (tonnes product per tonne feed DM)

  2. Calculate Implied Feed Cost:

    For each ruminant product, the grazing cost per tonne of feed is:

    \[\text{Feed Cost (USD/tonne DM)} = \text{Product Cost (USD/tonne)} \times \text{Efficiency (tonne product/tonne DM)}\]

  3. Global Averaging: Average the implied feed costs across all ruminant products to get a single grazing cost rate

  4. Result: A single global grazing cost in USD per tonne dry matter feed

This approach ensures that grazing costs are:

  • Properly allocated across different ruminant products

  • Consistent with the feed conversion efficiencies used in the model

  • Applied at the correct point in the production chain (grassland feed production)

Application in the Optimization Model

Production costs are applied as marginal costs on PyPSA network links, affecting the objective function during optimization.

Crop Production Costs

Implementation: workflow/scripts/build_model/crops.py

Crop costs are applied to production links that convert land into crop output:

Link structure:

  • Input (bus0): Land pool (Mha) by region, resource class, water supply

  • Output (bus1): Crop commodity bus (Mt) by country

  • Efficiency: Crop yield (Mt/Mha)

Cost calculation:

For single-season crops:

# Look up per-(crop, country) cost from FAOSTAT-derived data
cost_per_ha = crop_costs.get((crop, country), global_median_cost.get(crop, 0.0))

# Convert USD/ha to bnUSD/Mha (PyPSA units)
marginal_cost = cost_per_ha * 1e6 * USD_TO_BNUSD

# Optional: add calibration correction (bnUSD/Mha, additive)
marginal_cost += cost_calibration.get((crop, country), 0.0)

For multi-cropping systems (multiple crops per year on the same land):

# Sum per-(crop, country) costs across all crops in the combination
total_cost = sum(crop_costs.get((c, country), median) for c in crops_in_cycle)

# Convert to bnUSD/Mha
marginal_cost = total_cost * 1e6 * USD_TO_BNUSD
Interpretation:

  • The marginal cost represents the economic cost of using one Mha of land for crop production

  • Costs vary by country, reflecting local price and yield conditions

  • The optimization balances production costs against other objectives (nutrition, emissions, etc.)

Livestock Production Costs

Implementation: workflow/scripts/build_model/animals.py (lines 230-243)

Livestock costs are applied to links that convert feed into animal products:

Link structure:

  • Input (bus0): Feed bus (Mt DM) by feed category and country

  • Output (bus1): Animal product bus (Mt fresh weight) by country

  • Efficiency: Feed conversion efficiency (Mt product / Mt feed DM)

Cost calculation:

# Cost from animal_costs.csv (USD per tonne of product)
cost_per_t_product = animal_costs.loc[product]

# Efficiency (t product per t feed DM = Mt product per Mt feed DM)
efficiency = feed_requirements.loc[product, feed_category, 'efficiency']

# bus0 dispatch is Mt feed: scale tonne -> Mt and weight by product output
marginal_cost = (
    cost_per_t_product
    * efficiency
    * MEGATONNE_TO_TONNE
    * USD_TO_BNUSD
)  # bnUSD per Mt feed
Interpretation:

  • The marginal cost penalises feed dispatch in proportion to the resulting product output, so more productive feed allocations incur larger costs in absolute terms.

  • Higher feed-conversion efficiency raises the per-Mt-feed cost coefficient but produces proportionally more product, so the per-tonne-product cost is unchanged.

Grazing Costs

Implementation: workflow/scripts/build_model/grassland.py (lines 235-236)

Grazing costs are applied to links that produce grassland feed from land:

Link structure:

  • Input (bus0): Land pool (Mha) by region and resource class (rainfed only)

  • Output (bus1): Ruminant grassland feed bus (Mt DM) by country

  • Efficiency: Grassland yield (Mt DM / Mha)

Cost calculation:

# Grazing cost (USD per tonne DM)
grazing_cost_per_tonne_dm = calculate_grazing_cost_per_tonne_dm(...)

# Grassland yield (Mt DM per Mha) — already effective feed yield
efficiency = grassland_yield

# Convert to cost per Mha (bnUSD/Mha)
marginal_cost = (grazing_cost_per_tonne_dm * efficiency *
                 MEGATONNE_TO_TONNE * USD_TO_BNUSD)
Interpretation:

  • The marginal cost represents the economic cost of producing grassland feed from one Mha of pasture

  • Higher-yielding grassland has higher costs per Mha (but the cost per tonne DM is constant)

  • Grassland yields are already corrected for utilization in the merge step (see Livestock & Grazing)

Model Units and Conversions

Mass Units

  • Input data: Often in tonnes (t) or kilograms (kg)

  • Model buses: Megatonnes (Mt) for all commodity flows

  • Conversion: 1 Mt = 1,000,000 t = 1e6 t

Area Units

  • Input data: Usually hectares (ha) or acres

  • Model land buses: Mega-hectares (Mha)

  • Conversions:

    • 1 Mha = 1,000,000 ha = 1e6 ha

    • 1 acre = 0.404686 ha

    • 1 ha = 2.47105 acres

Currency Units

  • Input data: USD or EUR (various base years)

  • Model objective: Billion USD (bnUSD) in configured base year

  • Conversion: 1 bnUSD = 1,000,000,000 USD = 1e9 USD

  • Constant: USD_TO_BNUSD = 1e-9

Configuration

Cost-related configuration parameters are specified in config/default.yaml:

Crop costs:

crop_costs:
  non_endogenous_cost_share: 0.7  # Fraction of revenue attributed to non-endogenous costs
  outlier_cap_quantile: 0.90      # Crop-specific upper winsorization; null disables
  faostat:
    price_element_code: 5532      # Producer Price (USD/tonne)
    yield_element_code: 5412      # Yield (kg/ha)
cost_calibration:
  enabled: false       # Apply calibration corrections
  generate: false      # Generate calibration from solved model
  scenario: "calibration"
  crop_correction_csv: "data/curated/calibration/crop_cost.csv"
  grassland_correction_csv: "data/curated/calibration/grassland_cost.csv"
  animal_correction_csv: "data/curated/calibration/animal_cost.csv"

Animal cost fallbacks (see Fallbacks for Products Without Source Data):

animal_costs:
  fallback_aliases:
    dairy-buffalo: dairy
  fallback_values_usd_per_t:
    meat-chicken:
      production: 1300
      grazing: 0
    meat-sheep:
      production: 1200
      grazing: 2300

Grazing cost items (USDA):

grazing_cost_items:
  - "Grazed feed"

Grazing cost items (FADN, SE codes):

fadn_grazing_cost_items:
  SE105: "Forage crops"
  SE110: "Pasture"

See Also