Validation

Model validation fixes food production and demand to observed 2020 values in order to uncover potential data inconsistencies and verify that emissions calculations produce plausible results. By constraining both supply and demand to match reality, any remaining imbalances — surfaced via slack variables — reveal where the model’s input datasets, processing assumptions, or structural simplifications fail to reproduce observed commodity flows.

This page presents results from the validation configuration and explains how to interpret the diagnostic figures. For configuration reference, see Validation Configuration; for details on the slack mechanism, see Slack Mechanism.

Validation Configuration

The validation configuration (config/validation.yaml) pins the model to the 2020 baseline by enabling several flags:

planning_horizon: 2020

validation:
  use_actual_yields: true          # Use observed yields instead of potential
  use_actual_production: true      # Fix harvested areas to observed values
  enforce_baseline_diet: true      # Fix consumption to baseline diet
  enforce_baseline_feed: true      # Fix animal feed use to GLEAM baseline
  disable_spared_cropland: true    # No cropland retirement
  disable_spared_grassland: true   # No grassland retirement
  slack_marginal_cost: 10          # bn USD per Mt/Mha slack penalty
  grassland_yield_multiplier: 1.2  # Calibration knob for grassland yields

These settings collectively remove the optimizer’s degrees of freedom:

• Supply side: use_actual_yields swaps GAEZ potential yield rasters for observed yields (see Crop Production), while use_actual_production fixes harvested areas to present-day values. Grassland production is similarly pinned (see Livestock & Grazing).

• Demand side: enforce_baseline_diet adds per-food, per-country equality constraints that force consumption to match the processed GDD baseline diet (see Current Diets). enforce_baseline_feed pins animal feed use to GLEAM-derived baseline levels (see Baseline Feed Intake).

• Land use: Sparing of existing cropland and grassland is disabled so the model matches the historical land footprint (see Land Use & Resource Classes).

• Calibration multiplier: grassland_yield_multiplier applies a small adjustment to grassland feed yields to compensate for known data gaps.

Additional settings select present-day water availability (water.supply_scenario: current_use) and disable health impacts, since the goal is physical mass balance rather than optimization.

Slack Mechanism

When production and demand are fully fixed, the model may be unable to balance commodity flows exactly. Slack generators allow small violations of constraints at a configurable penalty cost (slack_marginal_cost), so the solver always finds a feasible solution. The magnitude of slack in each category reveals where data inconsistencies exist:

• Food slack: Difference between baseline demand and what the supply chain can deliver. Large food slack indicates missing processing pathways, incorrect yields, or data coverage gaps.

• Feed slack: Imbalance between the feed requirements of livestock and the available feed supply from crops, residues, and grassland.

• Land slack: Cases where fixed harvested areas exceed the available land endowment in a region.

• Water slack: Irrigation demand exceeding available water supply.

See Configuration for full details on the validation configuration keys.

Crop Production

The crop production map shows the dominant crop group and land-use intensity when production is fixed to observed 2020 values. Pasture is excluded from this map to prevent it from dominating the visualization; it is shown separately below.

Dominant crop group and cropland utilization intensity under the validation configuration, with production fixed to observed 2020 values. Colour indicates the crop group with the largest area in each grid cell; alpha encodes utilization of potential cropland. Pasture/grassland is excluded and shown separately in the pasture map below.

Pasture

Grassland and pasture production typically accounts for the largest share of agricultural land globally. The map below shows pasture utilization intensity in isolation.

Pasture utilization intensity under the validation configuration. Green intensity encodes the fraction of available pasture land that is used for grazing. High-intensity regions correspond to areas with dense livestock production (see Livestock & Grazing).

Food Group Slack

The food group slack plot shows the difference between baseline consumption targets and what the model’s supply chain can actually deliver when production is fixed. Large deviations indicate data inconsistencies — for example, missing food processing pathways, incorrect yield data, or gaps in trade network coverage.

The figure has two panels:

• Top panel: Absolute slack in megatonnes. Bars above zero indicate excess (the model produces more than the baseline demands); bars below zero indicate shortage (the model cannot fully supply baseline demand).

• Bottom panel: Relative deviation as a percentage of baseline demand. This normalizes for the vastly different scales of food groups, making it easier to spot proportionally large imbalances.

Food groups are sorted by relative deviation (largest first) in both panels.

Food group slack under the validation configuration. Top: absolute slack (Mt). Bottom: relative deviation (% of baseline demand). Groups are sorted by relative deviation. See Nutrition for food group definitions.

Slack Overview

The slack overview aggregates all slack categories — food, feed, land, and water — into a single chart showing the total penalty cost incurred in each category. This provides a high-level view of the overall model balance: categories with large slack costs are the primary areas where data or structural assumptions need refinement.

Slack penalty by category under the validation configuration. Bar length shows cost (bn USD); annotations show the physical quantity and unit. Categories with zero slack are omitted. See Land Use & Resource Classes for land slack and Livestock & Grazing for feed slack details.

Feed Breakdown

The feed breakdown shows the composition of dry-matter feed consumed by each animal type. This is useful for validating that the model’s livestock sector produces plausible feed mixes — for example, that ruminants receive predominantly grass and roughage while monogastrics rely on grains and protein meals.

Dry-matter feed use (Mt) by animal type and feed category under the validation configuration. Animals are sorted by total feed intake. See Livestock & Grazing for the feed conversion model and category definitions.

Feed slack from the validation solve drives the calibration pipeline described in Calibration.

Running the Validation

To run the validation configuration locally:

# Full validation run (build + solve + analysis)
tools/smk -j4 --configfile config/validation.yaml

# Generate validation plots
tools/smk -j4 --configfile config/validation.yaml -- \
    results/validation/plots/scen-default/crop_production_map.pdf \
    results/validation/plots/scen-default/food_group_slack.pdf \
    results/validation/plots/scen-default/slack_overview.pdf \
    results/validation/plots/scen-default/feed_breakdown.pdf