Current Diets

Overview

The model uses a hybrid approach to represent current consumption patterns, combining empirical dietary intake data from the Global Dietary Database (GDD) [GDD2024] [Miller2021] with food supply data from FAOSTAT Food Balance Sheets (FBS). This baseline data serves multiple purposes:

• Health impact assessment: Calculating disease burden attributable to current dietary patterns

• Baseline reference: Comparing optimized diets against current consumption

• Model constraints: Optionally constrain the optimization to remain near current diets

Population-weighted mean food group consumption (g/person/day) by UN M49 macro-region, showing how dietary patterns vary across world regions.

Global population-weighted mean consumption (g/person/day) broken down by individual foods within each food group.

Data Sources

Global Dietary Database (GDD)

• Provider: Tufts University Friedman School of Nutrition Science and Policy

• Coverage: 185 countries, individual-level dietary surveys (1990-2018)

• Variables: 54 dietary factors including foods, beverages, and nutrients

• Download: Requires free registration at https://globaldietarydatabase.org/data-download

• Citation: [GDD2024]

FAOSTAT Food Balance Sheets (FBS)

• Provider: FAO Statistics Division

• Coverage: Global, annual estimates of food supply

• Variables: Food supply quantity (kg/capita/year)

• Usage: Supplements GDD for food groups where intake survey data is sparse or inconsistent (Dairy, Poultry, Vegetable Oils)

Weight Conventions

GDD reports all dietary intake values in grams per day using “as consumed” weights [Miller2021]. This means:

• Fresh vegetables and fruits: Reported in fresh weight (e.g., a fresh banana, fresh tomato)

• Grains: Reported in cooked weight (e.g., cooked rice, prepared bread)

• Dairy: Reported as total milk equivalents, which includes milk, yogurt, cheese and other dairy products converted to their milk equivalent weight

• Meats: Reported in cooked/prepared weight

The model preserves these conventions in the processed output files. Units in the output CSV distinguish between general fresh weight (g/day (fresh wt)) and dairy milk equivalents (g/day (milk equiv)).

GDD to Food Group Mapping

The model maps GDD dietary variables to the food groups defined in config/food_groups. This mapping is implemented in workflow/scripts/prepare_gdd_dietary_intake.py.

Food Groups with GDD Data

The following food groups are populated from GDD variables:

Food Group	GDD Code	Description
fruits	v01	Total fruits (whole fruits only, excluding juices)
vegetables	v02	Non-starchy vegetables
starchy_vegetable	v03, v04	Potatoes + other starchy vegetables (aggregated)
legumes	v05	Beans and legumes
nuts_seeds	v06	Nuts and seeds
grain	v07	Refined grains (white flour, white rice)
whole_grains	v08	Whole grains
red_meat	v10	Unprocessed red meats (cattle, pig)
prc_meat	v09	Total processed meats
eggs	v12	Eggs
sugar	v15, v35	Sugar-sweetened beverages and added sugars
stimulants	v17, v18	Coffee and tea. GDD reports these in cups/day (brewed beverage); the script converts to dry commodity weight using configured factors (default: 14.4 g-dry/cup for coffee, 2.4 g-dry/cup for tea). Cocoa is not covered by GDD and enters only via FAOSTAT within-group shares.

Notes:

• Multiple GDD variables can map to a single food group (e.g., starchy_vegetable = v03 potatoes + v04 other starchy veg)

• When aggregating, values are summed within each food group

• The fruits food group uses only v01 (whole fruits), excluding v16 (fruit juices), to align with the GBD fruit risk factor definition used in health impact modeling

• GDD also tracks fish/seafood (v11), but fish is not currently modelled as a food group

Food Groups Sourced from FAOSTAT

The following food groups are populated from FAOSTAT Food Balance Sheets (FBS) because intake survey data (GDD) is often sparse, inconsistent, or structurally missing for these commodities:

Food Group	Description & Source Items
dairy	Total Milk Equivalent. Aggregated from FAOSTAT items: Milk - Excluding Butter (2848), Butter/Ghee (2740), and Cream (2743). Butter and cream are converted to milk equivalents using FAO dairy commodity tree extraction rates (≈21.3× for butter/ghee, ≈6.7× for cream); milk-excl.-butter is taken as-is.
poultry	Poultry Meat (2734).
oil	Vegetable Oils (2586).

Methodology for FAOSTAT Data: FAOSTAT reports “Food Supply” (retail weight), which typically includes household waste. The model converts this to “Dietary Intake” (consumed weight) by applying country-specific waste fractions derived from the UNSD Food Waste Index (see Food Processing & Trade).

Data Processing

The dietary data processing pipeline involves three stages:

1. Prepare GDD Data (workflow/scripts/prepare_gdd_dietary_intake.py): Processes GDD survey data for most food groups.

2. Prepare FAOSTAT Data (workflow/scripts/prepare_faostat_gdd_supplements.py): Fetches FAOSTAT supply data for dairy, poultry, and oil; converts supply to intake by subtracting waste; fills missing countries using proxies.

3. Merge Sources (workflow/scripts/merge_dietary_sources.py): Combines the datasets into a unified dietary_intake.csv.

The GDD processing step (Step 1) performs the following:

1. Load GDD files: Read country-level CSV files (v*_cnty.csv) for each dietary variable

2. Filter to reference year: Extract data for config.health.reference_year (default: 2018)

3. Map age groups: Convert GDD age midpoints to GBD-compatible age buckets (0-1, 1-2, 2-5, 6-10, 11-74, 75+ years)

4. Aggregate strata: Use GDD’s pre-computed population-weighted national aggregate rows (female=999, urban=999, edu=999) rather than a simple mean across all demographic strata, which would ignore stratum sizes. Falls back to simple mean only when aggregate rows are absent.

5. Map to food groups: Apply the GDD-to-food-group mapping defined in the script

6. Aggregate variables: Sum multiple GDD variables that map to the same food group (preserving age stratification)

7. Handle missing countries: Apply proxies for territories without separate GDD data

8. Validate completeness: Ensure all required countries and food groups are present

9. Output: Write processing/{name}/gdd_dietary_intake.csv with age-stratified data

Output Format

The processed dietary intake file has the following structure:

unit,item,country,age,year,value
g/day (milk equiv),dairy,USA,0-1 years,2018,252.3
g/day (milk equiv),dairy,USA,1-2 years,2018,258.3
g/day (milk equiv),dairy,USA,11-74 years,2018,174.6
g/day (milk equiv),dairy,USA,All ages,2018,187.1
g/day (fresh wt),fruits,USA,11-74 years,2018,145.2
...

Where:

• unit: Weight convention specific to the food group

  • g/day (fresh wt): Fresh/cooked “as consumed” weight for most foods

  • g/day (milk equiv): Total milk equivalents for dairy

  • g/day (refined sugar eq): Refined sugar equivalent for the sugar food group

• item: Food group name

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

• age: Age group using GBD-compatible naming

  • 0-1 years: Infants under 1 year

  • 1-2 years: Toddlers 1-2 years

  • 2-5 years: Early childhood 2-5 years

  • 6-10 years: Middle childhood 6-10 years

  • 11-74 years: Adults 11-74 years

  • 75+ years: Elderly 75+ years

  • All ages: Population-weighted average across all age groups

• year: Reference year

• value: Mean daily intake in grams per person for the specified age group

Country Coverage

The GDD dataset covers 185 countries. For a small number of territories without separate dietary surveys, the model uses proxy data from similar countries:

• American Samoa (ASM): Uses Samoa (WSM) data

• French Guiana (GUF): Uses France (FRA) data

• Puerto Rico (PRI): Uses USA data

• Somalia (SOM): Uses Ethiopia (ETH) data

These proxies are defined in the COUNTRY_PROXIES dictionary in prepare_gdd_dietary_intake.py.

GBD Dietary Risk Exposure Data

In addition to the GDD survey data and FAOSTAT supplements described above, the model also incorporates dietary exposure estimates from the Global Burden of Disease (GBD) Study 2019 [Brauer2024]. These estimates cover adults aged 25 and older and are derived from the GBD’s dietary risk factor framework (see the “GBD Dietary Risk Factors” section in Health Impacts for the full risk factor definitions).

The GBD data provides country-level intake estimates (g/day) for the following food groups:

GBD Risk Factor	Model Food Group	Notes
FRUIT	fruits	Whole fruits, excluding juices
VEG	vegetables	Non-starchy vegetables
WHOLEGRAINS	whole_grains	Whole grains
LEGUMES	legumes	Beans and pulses
NUTS	nuts_seeds	Nuts, seeds, and peanuts
REDMEAT	red_meat	Unprocessed red meats
MILK	(cross-validation only)	Logged for comparison against FAOSTAT dairy

These six food groups (excluding milk) overlap with GDD estimates, enabling both cross-validation and averaging to produce more robust group totals. The GBD data is processed by workflow/scripts/prepare_gbd_dietary_risk_exposure.py, which filters the raw IHME CSV files for the configured reference year, maps GBD location names to ISO3 country codes, and outputs processing/{name}/gbd_dietary_risk_exposure.csv.

Baseline Diet Estimation

The dietary intake pipeline described above produces food-group-level totals (e.g., “fruits: 145 g/day in the USA”). The model, however, operates at the level of individual foods (e.g., banana, citrus). The baseline diet estimation algorithm bridges this gap by combining food-group totals with FAOSTAT item-level supply data to produce per-food, per-country consumption estimates.

This algorithm is implemented in workflow/scripts/estimate_baseline_diet.py and proceeds in four steps.

Step 1: Food Group Totals

For food groups where both GDD and GBD estimates are available, the two sources are averaged to produce a more robust estimate:

\[T_g = \frac{T_g^{\mathrm{GDD}} + T_g^{\mathrm{GBD}}}{2}\]

This averaging applies to six food groups: fruits, vegetables, whole_grains, legumes, nuts_seeds, and red_meat. If GBD data is missing for a particular country, the GDD value is used alone.

For all other food groups (dairy, poultry, oil, grain, starchy_vegetable, prc_meat, eggs, sugar, stimulants), the GDD or FAOSTAT value from dietary_intake.csv is used as-is.

The script also logs cross-validation metrics between GDD and GBD for the overlapping groups, reporting the median and range of the GDD/GBD ratio across countries. GBD milk intake is logged separately for comparison against the FAOSTAT-derived dairy estimate.

The age group used for baseline totals is configured via config.diet.baseline_age (default: "11-74 years").

Step 2: Within-Group Food Shares

Once food-group totals are established, the algorithm determines how to distribute each group’s total across its constituent foods. This uses FAOSTAT FBS item-level supply data to compute relative proportions.

Food-to-FBS-item mapping

The mapping between model foods and FBS items is defined in data/curated/faostat_food_item_map.csv. Key characteristics:

• A food can map to multiple FBS items (supplies are summed). For example, citrus maps to four FBS items: Oranges/Mandarines (2611), Lemons/Limes (2612), Grapefruit (2613), and Citrus Other (2614).

• Multiple foods can share a single FBS item (requiring disambiguation). For example, cowpea, chickpea, gram, phaseolus-bean, and pigeon-pea all map to FBS item “Beans” (2546).

Three resolution scenarios arise when computing within-group shares from FBS supply:

1. Unique mapping: The food is the sole claimant of its FBS item and receives 100% of that item’s supply.

2. Shared FBS item with QCL resolution: Multiple foods share an FBS item. Country-level production data from FAOSTAT QCL (production) statistics is used to split the shared supply proportionally (see below).

3. No QCL mapping: If no production data is available, the shared supply is split equally among the unresolved foods.

QCL-based disambiguation

When multiple model foods map to the same FBS item, their relative consumption shares are estimated from country-level production data. The file data/curated/faostat_food_qcl_resolution.csv maps each such food to a specific QCL (production statistics) item code.

Example

cowpea, chickpea, gram, phaseolus-bean, and pigeon-pea all map to FBS item “Beans” (2546). The QCL resolution file maps each to a distinct production item:

• cowpea → QCL 195 (Cow peas, dry)

• phaseolus-bean → QCL 176 (Beans, dry)

• pigeon-pea → QCL 197 (Pigeon peas, dry)

• chickpea → QCL 191 (Chick peas, dry)

• gram → QCL 191 (Chick peas, dry)

In a country where cow pea production is 2 Mt and chickpea production is 1 Mt (with no other pulses), cowpea would receive 2/3 of the “Beans” supply while chickpea and gram together receive 1/3 (split equally between them, since they share QCL code 191).

The same approach is used for dairy: dairy (cattle milk) and dairy-buffalo (buffalo milk) share FBS item 2848 and are resolved via QCL items 882 and 951 respectively.

Millet species split

Pearl millet (pearl-millet) and foxtail millet (foxtail-millet) both map to FBS item 2517 (“Millet and products”) and to the same QCL aggregate “Millet”, which does not distinguish species. Since FAO lacks species-level millet production data, a fixed 80/20 split is applied globally: 80% pearl millet, 20% foxtail millet, based on literature estimates of global production shares.

Vegetable residual projection

FAOSTAT FBS item 2605 (“Vegetables, other”) is a residual catch-all covering all vegetables not individually itemized. Tomato has its own explicit FBS item (2601) and is unaffected by the residual.

The residual supply is distributed across the model’s “other vegetable group” (OVG) crops — onion, cabbage, and carrot — using blended production shares:

\[s_{c,f} = 0.7 \times s_{c,f}^{\mathrm{country}} + 0.3 \times s_f^{\mathrm{global}}\]

where \(s_{c,f}^{\mathrm{country}}\) is the country-specific production share of crop \(f\) among the three OVG crops, and \(s_f^{\mathrm{global}}\) is the global production share. The 70/30 blending avoids over-reliance on country-level data, which can be noisy for small producers.

Note

onion also has an explicit FBS item (2602, “Onions”), so its total supply is the sum of its explicit item supply and its projected share of the residual. cabbage and carrot rely entirely on the residual projection.

Starchy vegetable residual projection

An analogous residual exists for starchy vegetables: FBS item 2534 (“Roots, Other”) covers root and tuber crops not individually itemized. The four modeled starchy foods — potato, sweet-potato, yam, and cassava — each have their own explicit FBS items (2531, 2533, 2535, and 2532 respectively), but the residual supply from item 2534 is additionally distributed across all four using the same blended production share approach (70% country-specific, 30% global).

Note

Because the crop production data uses the name white-potato while the model food is called potato, the projection applies a name mapping before computing production shares.

Nuts/seeds residual projection

FBS item 2551 (“Nuts and products”) is a residual catch-all for tree nuts and seeds not individually itemized. The residual supply is distributed across groundnut, sesame-seed, coconut, and sunflower-seed using the same 70/30 blended production share approach. Other nuts/seeds foods with their own explicit FBS items are unaffected.

Step 3: Per-Food Consumption

The final per-food consumption estimate is the product of the food-group total and the within-group share:

\[c_{i,f} = T_{i,g(f)} \times s_{i,f}\]

where \(c_{i,f}\) is the estimated consumption (g/day) of food \(f\) in country \(i\), \(T_{i,g(f)}\) is the group total for the food group containing \(f\), and \(s_{i,f}\) is the within-group share.

As a validation check, within-group sums are verified to match group totals within a tolerance of 0.1 g/day. Any discrepancies are logged as warnings.

Step 4: FBS Supply Overrides

For specific foods where GDD survey-based group totals substantially underestimate actual consumption, the per-food estimate from Step 3 is replaced with waste-corrected FAOSTAT Food Balance Sheet supply data. The list of overridden foods is configured via config.diet.fbs_override_foods (default: ["yam"]).

For each overridden food and country, the replacement consumption is:

\[c_{i,f} = \frac{S_{i,f} \times 1000}{365} \times (1 - w_{i,g(f)})\]

where \(S_{i,f}\) is the FAOSTAT food supply (kg/capita/year) summed across the food’s FBS item codes, and \(w_{i,g(f)}\) is the country- and group-level consumer waste fraction from the food loss and waste dataset.

Why yam needs an override

GDD starchy vegetable intake for sub-Saharan Africa is 7–33× below FAOSTAT food supply (e.g., Nigeria: GDD ≈ 72 g/day vs. FBS ≈ 700 g/day for starchy vegetables). Because yam production is almost entirely concentrated in West Africa, the GDD underestimate translates directly to a ~10× underestimate of yam demand. The within-group shares are correct — the problem is entirely in the GDD group total for starchy vegetables in these countries. Overriding yam consumption with FBS supply ensures that the model’s demand matches observed food availability.

Baseline Diet Output

The output file processing/{name}/baseline_diet.csv contains one row per (country, food) combination with the following columns:

Column	Description
country	ISO 3166-1 alpha-3 country code
food	Model food name (e.g., banana, rice-white, cowpea)
food_group	Food group to which the food belongs
consumption_g_per_day	Estimated daily consumption in grams per person

Rows are sorted by (country, food_group, food).

Downstream Uses

The baseline diet feeds into several parts of the model:

• Baseline diet enforcement: When config.validation.enforce_baseline_diet is enabled, the solver adds per-food, per-country equality constraints on food consumption links, forcing the solution to replicate observed intake.

• Within-group ratio fixing: When config.food_groups.fix_within_group_ratios is enabled, the solver constrains foods within each group to maintain their baseline proportions while allowing group totals to vary.

• Piecewise consumer utility calibration: In the consumer-values workflow, baseline per-food consumption and baseline food-equality duals are combined to calibrate results/{name}/consumer_values/utility_blocks.csv. These blocks are then used in the solve objective when config.food_utility_piecewise.enabled is true.

  The current calibration anchors utility at baseline quantity: the block containing baseline consumption uses the extracted dual value, while blocks below baseline are more valuable and blocks above baseline are less valuable.

• Health impact assessment: Baseline consumption is used when computing the population-attributable fraction of diet-related disease burden (see Health Impacts).

Workflow Integration

Snakemake rules:

• prepare_gdd_dietary_intake

• prepare_faostat_gdd_supplements

• merge_dietary_sources

• prepare_gbd_dietary_risk_exposure

• estimate_baseline_diet

Input data:

• data/manually_downloaded/GDD-dietary-intake/Country-level estimates/*.csv (GDD)

• data/manually_downloaded/IHME_GBD_2019_DIET_RISK_1990_2019_DATA/*.csv (GBD)

• FAOSTAT API (live fetch for FBS, QCL, and animal production data)

Curated data files:

File	Purpose
data/curated/faostat_food_item_map.csv	Maps model foods to FAOSTAT FBS item codes for within-group share calculation
data/curated/faostat_food_qcl_resolution.csv	Maps foods sharing an FBS item to individual QCL production codes for disambiguation
data/curated/food_groups.csv	Defines the food-to-food-group mapping

Configuration parameters:

• config.countries: List of countries to process

• config.food_groups.included: Food groups to filter and aggregate

• config.health.reference_year: Year for GDD dietary intake data

• config.diet.baseline_reference_year: Year for GBD exposure data

• config.diet.baseline_age: Age group for baseline totals (default: "11-74 years")

• config.diet.fbs_override_foods: Foods whose consumption is overridden with waste-corrected FBS supply (default: ["yam"])

• config.byproducts: Foods to exclude from share calculation (e.g., wheat-bran)

Output:

• processing/{name}/dietary_intake.csv — Merged food-group-level intake

• processing/{name}/gbd_dietary_risk_exposure.csv — GBD risk exposure estimates

• processing/{name}/baseline_diet.csv — Per-food, per-country consumption

Scripts:

• workflow/scripts/prepare_gdd_dietary_intake.py

• workflow/scripts/prepare_faostat_gdd_supplements.py

• workflow/scripts/merge_dietary_sources.py

• workflow/scripts/prepare_gbd_dietary_risk_exposure.py

• workflow/scripts/estimate_baseline_diet.py

References

[GDD2024] (1,2)

Global Dietary Database. Dietary intake data by country, 2018. Tufts University Friedman School of Nutrition Science and Policy. https://www.globaldietarydatabase.org/ (accessed 2025)

[Miller2021] (1,2)

Miller V, Singh GM, Onopa J, et al. Global Dietary Database 2017: Data Availability and Gaps on 54 Major Foods, Beverages and Nutrients among 5.6 Million Children and Adults from 1220 Surveys Worldwide. BMJ Global Health, 2021;6(2):e003585. https://doi.org/10.1136/bmjgh-2020-003585