ULAB Plume — Methodology Guideline

Version 1.15.2 · October 2026 · Pure Earth · chrisk@pureearth.org

1. Purpose and scope

ULAB Plume is a web-based screening tool for estimating the air, soil, and blood lead level (BLL) impacts of used lead-acid battery (ULAB) recycling operations on surrounding populations. It is designed for use in four contexts:

• Regulatory compliance screening: providing screening-grade impact estimates to support dialogue with environmental ministries
• Retrospective health-burden estimation: characterising the magnitude and footprint of past or ongoing exposure for impact assessment
• Advocacy and public-health communication: quantifying the spatial reach of contamination from a specific facility for public health communication
• Education: an interactive tool for learning how dispersion parameters influence pollution distribution, exposure pathways, and health burden

It is not a regulatory submission tool. Outputs are screening estimates with documented uncertainty bounds. For permit decisions or litigation, site-specific measurements and refined modelling are required. See more on use cases here.

1.5 About AERMOD and the LMIC application gap

What AERMOD is

AERMOD is a steady-state Gaussian plume dispersion model developed by the US Environmental Protection Agency (EPA) and the American Meteorological Society. It simulates how pollutants emitted from industrial sources disperse through the atmosphere, accounting for wind, atmospheric stability, terrain, building wake effects, and boundary-layer dynamics.

The model has been continuously developed since the late 1990s, reviewed extensively in peer-reviewed literature, and refined through real-world application across thousands of regulatory cases.

→ Official AERMOD page: epa.gov/scram/air-quality-dispersion-modeling-preferred-and-recommended-models

Regulatory standing of AERMOD (and what that does and does not mean for this tool)

AERMOD is the US EPA's preferred regulatory dispersion model for near-source air quality assessment (US EPA, 40 CFR Part 51, Appendix W — Guideline on Air Quality Models). This means it is the default tool used in the United States for:

• Permit applications: new or modified industrial facilities must demonstrate compliance with National Ambient Air Quality Standards (NAAQS) using AERMOD modelling
• Risk assessment: determining whether a facility's emissions threaten public health or welfare
• Litigation and enforcement: AERMOD outputs are admissible evidence in environmental enforcement cases
• Policy analysis: Environmental Impact Statements and air quality management plans rely on AERMOD modelling

Equivalent regulatory acceptance exists in the EU (under the EEA Industrial Emissions Directive), Australia, Canada, and many other high-income jurisdictions. AERMOD is not a screening tool — it is the authoritative regulatory model for industrial emissions in data-rich settings.

Important distinction: AERMOD's regulatory acceptance applies only to AERMOD when it is used with the data and procedures that regulators specify — including site-specific hourly meteorology, measured stack parameters, and verified emission factors. ULAB Plume uses AERMOD as one component of a wider screening pipeline that substitutes assumed values for several of these regulatory inputs. The outputs of ULAB Plume should not be treated as carrying the same regulatory weight as a conventional AERMOD modelling exercise. ULAB Plume is not currently validated for regulatory screening in LMIC settings; it is a research and advocacy tool intended to support public health communication and inform further investigation.

Why AERMOD is rarely used for LMIC smelter assessment

Despite its rigour, AERMOD is rarely applied to lead-acid battery recycling (ULAB) facilities in low- and middle-income countries. Three data gaps prevent direct use:

1. Meteorological data

AERMOD requires hourly meteorological observations covering at least one full year, typically including:

• Wind speed and direction
• Temperature and humidity
• Cloud cover and ceiling height
• Solar radiation
• Mixing height and atmospheric stability indicators

In high-income countries, these data are produced continuously by National Weather Service stations and meteorological agencies, freely available, and pre-processed into AERMOD-ready files (SFC and PFL format) using the AERMET preprocessor.

In LMIC contexts, the situation is fundamentally different:

• Meteorological station coverage is sparse, with many countries having fewer than a dozen stations nationwide
• Hourly data, where collected, is often paywalled, intermittent, or not in standard formats
• Upper-air radiosonde data needed for boundary-layer characterisation is collected at very few stations
• The AERMET preprocessing infrastructure is not routinely operated outside high-income jurisdictions

The result: in most LMIC settings, the meteorological inputs AERMOD expects simply do not exist, or exist in forms that require substantial reprocessing before AERMOD can use them.

2. Source emission data for LMIC ULAB recycling

Lead emissions from ULAB recycling in LMIC contexts come from a spectrum of facility types. In sub-Saharan Africa, recent fieldwork (Manhart & Adjei 2026) finds that the recycling itself is dominated by formal industrial plants — registered, licensed, and subject to regulatory oversight. Its model medium-sized plants produce 10,000–20,000 t of lead a year, and it notes that "there are various existing plants with similar processing capacities in the studied African markets" (Manhart & Adjei 2026, §3.1, Tables 2 and 3). Many of these formal plants operate with substandard practices: degraded baghouse maintenance, missing fume hoods, manual battery breaking, and uncontrolled slag handling. In other regions including South and Southeast Asia, smaller informal operations are also common. Both contexts present significant data gaps.

• No measured stack or fugitive lead emission rate exists for any secondary (ULAB) smelter in a low- or middle-income country; the tool's defaults are derived from regulatory coefficients, high-income analogues and judgement (section 3.1)
• Formal plants in many LMIC settings operate with substandard controls, and no LMIC baghouse has been measured; published capture efficiencies describe working filters, not degraded operation
• Stack testing data is rare even for permitted formal facilities
• Informal/artisanal operations operate without permits and emit without oversight
• No particle size distribution of emitted lead has been measured at an LMIC plant or a rotary furnace; the defaults rest on old US blast-furnace data (section 3.3)
• Fugitive emissions (uncaptured releases from furnace charging, kettles, battery breaking, yard activities, raw material handling and slag disposal) have not been measured at any LMIC secondary smelter; in the default scenario they are 79 % of the lead emitted to air
• Operating schedules are often irregular and undocumented

The data needed to parameterise AERMOD for an LMIC ULAB site — emission rate per tonne processed, particle size distribution, stack geometry, temperature, exit velocity — is rarely available without direct site investigation.

3. Receptor and population data

AERMOD outputs concentration fields. Translating these into health risk estimates requires knowing where people live, work, grow food, and spend time. In high-income countries, fine-resolution gridded population data, land-use maps, and residential proximity records are routinely available. In LMIC settings, these are often sparse or outdated, particularly in informal settlements that frequently surround ULAB operations.

For a comprehensive reference of legally binding air quality standards for lead across LMIC contexts in Africa, Asia, and South America, see the Lead Air Legislation in LMICs page.

How this tool addresses the data gaps

ULAB Plume uses transparent, literature-based assumptions to fill each gap, while clearly documenting where assumed values are used and what their uncertainty bounds are.

Each assumption is documented in the relevant methodology section. The tool's outputs are presented with explicit caveats: as screening estimates suitable for advocacy and engagement, not as regulatory submissions or definitive risk assessments.

What this tool's outputs mean — confidence by pathway

The tool generates four kinds of output, and the confidence in each is different. Users should treat them with correspondingly different levels of caution.

Air concentration (highest confidence)

Predicted ambient air lead concentrations come directly from AERMOD's dispersion mathematics, applied to the user's emission inputs and the meteorology (real hourly reanalysis for the built-in locations; synthetic for a user-built wind rose). AERMOD's dispersion physics is well-validated in temperate, mid-latitude settings. The main sources of uncertainty in LMIC contexts are: how well one reanalysis grid cell and one typical year represent the site's meteorology, accuracy of emission rate inputs, and the absence of site-specific terrain. Predictions are best understood as order-of-magnitude correct for typical ULAB facility conditions.

Soil deposition (moderately high confidence)

Predicted deposition rates rely on the same dispersion mathematics plus AERMOD's particle-size-resolved dry deposition algorithm. For the built-in locations, wet deposition is modelled from hourly precipitation, and the headline deposition is total (wet + dry). For a user-built wind rose there is no precipitation, so deposition is dry only and is labelled so. Confidence is somewhat lower than for air concentrations — particularly for fine particles that travel further and where dry deposition velocities have wider uncertainty bounds.

Soil lead increment and BLL (moderate confidence)

Soil-Pb is derived from a linear-accumulation mass balance (deposition × exposure duration / depth × density) — conservative for typical soils on annual-to-decadal timescales, neglecting leaching, erosion, and root uptake. BLL increments are derived from direct US EPA IEUBK biokinetic-model outputs for the inhalation and soil/dust ingestion pathways, interpolated piecewise in log-log space between anchor concentrations. The IEUBK pathway anchors and the soil-mass-balance equation are documented in section 5. ΔIQ is derived from each receptor's modelled ΔBLL via the Crump et al. (2013) concurrent-BPb log dose-response. Both the soil accumulation model and the biokinetic dose-response have validity ceilings. Beyond them the tool suppresses the numeric soil-Pb and ΔBLL predictions rather than display extrapolated values; ΔIQ, by contrast, is capped at the value the dose-response gives at the validity ceiling (ΔBLL = 30 µg/dL) and labelled '≥' — a model floor rather than a hole — see section 5.4 for the validity treatment and threshold rationale.

Crop pathway and crop-attributable BLL (low confidence, screening only)

The crop layers and the crop contribution to ΔBLL apply a foliar-only deposition-uptake screening model with a composite transfer fraction R (Brown 2008 UK FSA / EC JRC TGD 2003 / US EPA OSWER 2002) that bundles interception, weathering, washoff, and (for cereals) threshing into a single literature-calibrated coefficient. Cereal R is empirically tuned against paired-measurement smelter sites in China (Liu 2020 Jiyuan; Liu 2022 Jiaozuo) — model predictions are within a factor of three of observed grain Pb at those sites. Root uptake is not modelled (small at the soil-Pb regime where direct deposition dominates near-field). The crop-to-BLL slope is a single linear coefficient (EFSA 2010 CONTAM Panel) applied to total dietary Pb intake; this is the largest open methodological gap at very high near-field intakes. The crop contribution is opt-in on the BLL surface (via the "Include crops" checkbox on the BLL panel) and is reported as a separate pathway in the response — users can read the crop heat maps (µg Pb / kg fresh-weight edible crop) without folding the crop pathway into ΔBLL or ΔIQ. The pathway is calibrated at screening precision and is only loosely validated against measured crop Pb at LMIC ULAB sites; see section 5.5 for caveats.

2. Atmospheric dispersion

2.1 Model

The tool runs AERMOD v24142, the US EPA's preferred regulatory dispersion model for near-source air quality assessment (US EPA Appendix W). AERMOD is a steady-state Gaussian plume model that includes:

• Atmospheric stability characterised through the Monin-Obukhov length, L (US EPA 2024, EPA-454/B-24-010, §3.2.1, eq. 8). For the built-in locations, L is calculated each hour from MERRA-2's sensible heat flux and friction velocity. The wind-rose builder (custom and multi-site meteorology) takes a Pasquill-Gifford-Turner class and converts it into AERMOD inputs: each class A–F sets a fixed heat flux and fixed mixing heights; u* comes from the neutral log profile, and L follows from u* and the heat flux (class D is neutral).
• Briggs plume rise for buoyant and momentum-driven plumes
• Dry deposition with particle-size-resolved deposition velocities
• Wet deposition from hourly precipitation, with plume depletion by wet removal (built-in locations only) (US EPA 2024, EPA-454/B-24-007, §3.2.2.5, §3.2.2.6, §3.2.2.12 and Appendix C)
• Boundary-layer-aware vertical mixing

The deposition shown as the headline result is total (wet + dry) where precipitation is modelled, otherwise dry only.

Fugitive area sources use AERMOD's non-regulatory FASTAREA option (US EPA 2024, EPA-454/B-24-007, §3.2.2.8). Against AERMOD's default area integration it changes period-average concentration and deposition beyond 500 m by less than about 0.25 %, and fence-line (50 m) soil Pb by up to +5.5 % (higher) for small, low sources (50 m side, 1 m release height); see section 3.5.

AERMOD is configured with AVERTIME 1 PERIOD so that two averaging periods are captured per scenario run:

• PERIOD output — the full-year (8,760-hour) average concentration and total deposition at each receptor. These are the values used for the concentration / deposition / BLL / IQ heat maps; they are the appropriate quantity for an annual-impact screening assessment. PERIOD deposition is treated as annual with a unit conversion only (g/m² → mg/m²/yr); no time-extrapolation factor (annualisation_factor = 1.0).
• 1-HR worst output — the single highest 1-hour concentration at each receptor across the 8,760-hour simulation. This is captured separately on the response as concentration_1hr_worst_ug_m3 for NAAQS-style acute-exposure framing. It is not used on the chronic concentration / BLL / IQ heat maps; map values labelled "concentration" or "annual" are PERIOD-derived unless explicitly labelled "1-HR worst".

The dual capture reflects the tool's design intent: the heat maps screen chronic facility-attributable impact on the surrounding population; the 1-HR worst is available for users who need a separate acute-exposure framing alongside the chronic assessment. ULAB Plume is calibrated for chronic-exposure screening and is not intended as a 1-hour NAAQS-compliance modelling tool — the 1-HR worst is a transparency field, not a regulatory output.

The air concentration heat map includes a dashed reference contour at 0.5 µg/m³ corresponding to the World Health Organization's annual mean Air Quality Guideline for lead (WHO 2021). Receptors above this contour exceed the WHO recommended ambient air lead concentration.

2.2 Meteorological inputs

Real LMIC hourly meteorology from local stations is rarely available. For the four built-in locations the tool uses real hourly meteorology from the NASA MERRA-2 reanalysis (Gelaro et al. 2017), written directly into the surface (SFC) and profile (PFL) files AERMOD reads (US EPA 2024, EPA-454/B-24-007, §2.1.1.2). Because the reanalysis carries precipitation, AERMOD also models wet deposition, and the headline deposition is total (wet + dry). A user-built wind rose still uses synthetic hourly meteorology, with dry deposition only (see the end of this section). Exception: multi-site runs (API only; the interface does not offer them) use the v1 wind-rose meteorology for the built-in locations, with dry deposition only.

Built-in locations: hourly MERRA-2 reanalysis

• Source. MERRA-2 hourly single-level fields (time-averaged, centred on the half hour) for the grid cell containing each site: 10 m wind (U10M, V10M), 2 m temperature (T2M), surface pressure (PS), sensible heat flux (HFLUX), boundary-layer height (PBLH), friction velocity (USTAR) and bias-corrected precipitation (PRECTOTCORR) (Bosilovich et al. 2016; Reichle et al. 2017). Years 2021–2025, the five most recent complete years at the time of the build.
• Greater Accra — Shai Hills cell. The earlier Accra location used the MERRA-2 cell centred 5.5°N, 0.0°E, which is mostly ocean (ocean fraction 0.59, land 0.40) and so represents a marine surface. The built-in Greater Accra location now uses the adjacent cell centred 6.0°N, 0.0°E (land fraction 0.80), which contains the Shai Hills smelter site (5.854°N, 0.047°E; about 17 km from the cell centre).
• Composite typical year. Each location runs one representative year assembled month by month: for each calendar month, the real month from 2021–2025 closest to that month's five-year climatology, measured by an equal-weight sum of standardised deviations over eight feature groups (the 16-sector wind-direction distribution, mean wind speed, stable-hour fraction, precipitation total, number of precipitation hours, mean daytime sensible heat flux, the joint stability × wind-speed frequencies and the joint wind-direction × stability frequencies). 29 February is dropped so the year is exactly 8,760 hours. The month selections are returned with every result. The composite is a climatological "typical year", not a forecast or a specific year. Known bias: the Abuja composite under-represents dry-season convective calm hours (unstable, wind below 1 m/s, mainly March–April), so near-stack (fence-line) concentrations from elevated sources at Abuja are biased low relative to the 2021–2025 average.
• Processing route (a limitation). The SFC and PFL files are written directly from the MERRA-2 fields; they are not produced by AERMET, the meteorological preprocessor for AERMOD (US EPA 2024, EPA-454/B-24-004, §1.0). EPA guidance requires prognostic meteorological data to be processed through AERMET for regulatory applications; for non-regulatory applications, such as this screening tool, the user may choose AERMET or direct AERMOD input, after consulting and seeking the agreement of the agencies or parties involved (US EPA 2024, EPA-454/B-24-005, §1 and §3.2). AERMET's boundary-layer scheme and its low-wind adjustments are therefore not applied. A benchmark against AERMET for Abuja is pending (section 2.5).
• Mapping. Sensible heat flux and friction velocity are taken from MERRA-2 hour by hour; the Monin–Obukhov length is computed from them, L = −ρ c_p T u*³ / (k g H), with air density from pressure and temperature (US EPA 2024, EPA-454/B-24-010, §3.2.1, eq. 8); the convective velocity scale is computed for unstable hours. The convective and mechanical mixing heights are both set to the MERRA-2 boundary-layer height in unstable hours (L < 0); in stable hours only the mechanical height is set. The potential-temperature gradient above the mixed layer is a constant 0.005 K/m. Wind is at 10 m and temperature at 2 m. Cloud cover and relative humidity are not supplied.
• Time. MERRA-2 hours (UTC) are labelled as AERMOD hour-ending local standard time. For Delhi (UTC+5:30) each hour is labelled with a documented 30-minute phase shift; nothing is interpolated.
• Low friction velocity. Night-time friction velocity at the inland sites is often very low. Because the files are not processed by AERMET, its optional low-wind friction-velocity adjustment (ADJ_U*; US EPA 2024, EPA-454/B-24-004, §3.7.7.4; Qian and Venkatram 2011) does not apply, so a friction-velocity floor of 0.05 m/s is applied instead. The floor is a convention, to be checked against the AERMET benchmark.
• Near-neutral hours. A near-zero positive heat flux gives an extremely large negative L that AERMOD would read as its missing-data code; such values are capped at −90,000 m (near-neutral).
• Precipitation rescaling (provisional). Over Africa, MERRA-2's corrected precipitation is adjusted towards the satellite- and gauge-based CMAP product on a 2.5° grid (Reichle et al. 2017). In our assessment that grid is too coarse to resolve local gradients such as the dry coastal strip around Accra. Hourly MERRA-2 precipitation is therefore rescaled month by month to CHIRPS (0.05°; Funk et al. 2015) at each location's representative point (the smelter point for Shai Hills), multiplicatively, so the hourly timing is kept. A month with less than 5 mm of MERRA-2 precipitation is left unscaled, and monthly factors are capped to 0.25–4. The rescaling is provisional, pending a check against national station records (section 2.5).
• Light-rain (drizzle) correction. Reanalyses tend to rain too often and too lightly: over the Sahel, MERRA and ERA-Interim produce too many light-precipitation days (Diaconescu et al. 2015), and over the western Pacific both MERRA-2 and ERA5 rain more often than observations (Serra et al. 2023). In MERRA-2, after the CHIRPS rescaling above, precipitation of at least 0.01 mm/h fell in 49 % of 2022 hours at Abuja and 74 % at the Shai Hills cell; these shares are before the 0.1 mm/h cut described next. Because AERMOD's wet scavenging is sub-linear in rain rate, many light hours scavenge more than the same total in fewer, heavier hours. So, within each month, hours below 0.1 mm/h are set to zero and the remaining rain hours are scaled up so the monthly total is unchanged (a month with only sub-0.1 mm/h rain is left as is). The threshold is a pragmatic correction of the frequency bias, not a calibrated rain-rate distribution; the true number of rain hours at each site is not known, so far-field wet deposition carries this uncertainty. Precipitation is treated as liquid (no composite hour is below 0 °C).
• Integrity and provenance. The composite files are checked against their recorded SHA-256 hashes and for structural validity when they are loaded. Every result carries the builder version and hash, the file hashes, the month-to-year selection, the MERRA-2 cell index and centre, and the precipitation-rescaling provenance.

How results change from the wind-rose meteorology

Measured with the tool's own runs: the lmic_typical scenario's emission assumptions and stack at 8,550 t of lead produced per year (the percentages do not depend on the plant size beyond rounding), Preview, at the four built-in locations, stack only and with the fugitive source on (default 100 m area), each on the wind-rose meteorology used before version 1.15.0 and on the hourly MERRA-2 meteorology. The changes are in the peak values; for deposition they compare the new headline (total, wet + dry) with the old one (dry only). The Greater Accra changes include the move from the marine cell to the Shai Hills cell.

Stack near-field maxima fall: the wind-rose meteorology assigned more strongly convective daytime hours, which brought the elevated plume to the ground near the stack. Low-level (fugitive) concentrations rise: the reanalysis has many more stable night hours, which keep a ground-level plume near the ground.

Wet deposition as a share of total deposition, averaged over each receptor ring (hourly meteorology, same runs):

With the stack alone, wet deposition is 56.6–86.5 % of the total at the 50 m ring, where little of the elevated plume has yet reached the ground, and 8.2–19.1 % at 500 m. With the fugitive source on, the ground-level source dominates near-source deposition, and wet deposition is 0.4–0.9 % at 50 m and 3.6–9.3 % at 500 m. At 5 km it is 7.7–27.5 % in both cases. The 50 km ring is left out: its deposition is close to the 0.01 mg/m²/yr resolution of the reported values, too coarse for a share.

User-built wind roses and multi-site runs

A user-built wind rose is turned into synthetic AERMOD-format meteorology; multi-site runs use the same route with the built-in locations' 2022 wind roses (section 2.3). There is no precipitation, so deposition is dry only. The pipeline:

1. STAR JSON input — Joint frequency table of (wind direction, wind speed bin, Pasquill stability class) for the representative city. 16 directions × 6 speed bins × 6 stability classes = 576 bins, with one frequency value per bin summing to 1.0.
2. Largest-remainder hour allocation — 8,760 annual hours are distributed across bins in proportion to their frequencies, using the largest-remainder algorithm to ensure exact total.
3. Diurnal placement — Each bin is assigned to specific hours of the year respecting class-specific time-of-day constraints:
   • Class A (very unstable): only 12–14h local time
   • Class F (very stable): only nighttime
   • Class D (neutral): any time
   The most-restricted classes are placed first to avoid scheduling conflicts.
4. ±11.25° direction jitter — Within each 22.5° direction wedge, individual hours are jittered uniformly (deterministic hash-based) to produce smooth heat map contours rather than 16-spoke artefacts.
5. Per-hour micrometeorology back-derivation — For each synthetic hour, AERMET-required parameters (friction velocity u*, sensible heat flux H₀, mixing height z_i, Monin-Obukhov length L) are derived from (wind speed, stability class) using AERSCREEN Appendix C lookup tables and Pasquill-Gifford-Turner relationships.
6. AERMET column-format output — SFC and PFL files are written to AERMET specification (CRLF line endings, fixed-width columns) and passed to AERMOD's SURFFILE/PROFFILE input lines.

2.3 Regional archetypes

The tool's built-in locations:

The four built-in locations deliberately span distinct ULAB-relevant climatological regimes so users can screen sensitivity to meteorology. The figures below are for each composite typical year as AERMOD reads it: wind-direction shares of all 8,760 hours (16 sectors; hours with wind below 0.5 m/s have no sector), mean 10 m wind speed, the share of stable hours (positive Monin–Obukhov length), and precipitation after the rescaling and light-rain correction (section 2.2).

• Abuja — inland West African savanna with a bimodal regime: south-westerly monsoon flow April–October and north-easterly Harmattan flow November–March. Leading sectors SSW 13.7 %, SW 13.7 % and NE 11.8 %. Mean wind speed 2.5 m/s; stable hours 63.2 %; 1,269 mm of precipitation, in 26.1 % of hours.
• Greater Accra — Shai Hills — Gulf of Guinea coastal plain, inland of the coast. South-westerly flow dominates in every season: SW 30.9 % and SSW 28.2 % of hours. Mean wind speed 2.7 m/s; stable hours 60.7 %; 1,273 mm of precipitation, in 36.4 % of hours.
• Dhaka — Ganges-Brahmaputra delta with a bimodal Asian monsoon regime: south-south-easterly to southerly flow April–October (SSE 15.7 %, S 14.7 %) and north-north-westerly to north-westerly flow November–March (NNW 8.3 %). Mean wind speed 2.7 m/s; stable hours 64.3 %; 2,107 mm of precipitation, in 36.3 % of hours.
• Delhi — continental Indo-Gangetic Plain. North-westerly flow dominates the year (NW 18.0 %, WNW 15.3 %), with east-south-easterly to south-easterly monsoon flow June–September. Mean wind speed 2.9 m/s; stable hours 61.7 %; 932 mm of precipitation, in 11.8 % of hours. The driest of the four.

The 2022 wind roses (multi-site runs). Multi-site runs still use the built-in locations' 2022 wind roses. They were derived from the same NASA MERRA-2 hourly reanalysis for 2022, sampled at the single grid cell containing each city's centre (for Accra, the marine cell centred 5.5°N, 0.0°E; MERRA-2 native resolution: 0.5° lat × 0.625° lon, ≈50–70 km). Hourly 10 m wind components, surface sensible heat flux, and 2 m temperature were downloaded via the NASA GES DISC OPeNDAP service and binned into a STAR-format joint frequency table:

• 16 wind direction wedges (22.5° each, centred N, NNE, NE, …, NNW).
• 6 wind speed bins per AERMET conventions: <1.54, 1.54–3.09, 3.09–5.14, 5.14–8.23, 8.23–10.80, and ≥10.80 m/s.
• 6 Pasquill-Gifford-Turner stability classes (A–F). Each hour is classified from surface sensible heat flux and 10 m wind speed via the Pasquill-Gifford-Turner lookup (Pasquill 1961; Turner 1969 Workbook Table 2.1; EPA AERMET 2018). Calm hours (WS < 0.5 m/s) are routed to a separate calm-fraction field rather than placed in any A–F bin.

The resulting STAR climatology is fed to wind_rose_processor.py, which generates a synthetic but kinematically consistent 8,760-hour AERMET SFC + PFL pair for AERMOD to consume. Synthesis is deterministic (same JSON → same SFC) so a scenario re-run produces bit-identical met. Surface land-use parameters (roughness length, Bowen ratio, albedo) are sourced from regional literature defaults rather than MERRA-2 — documented per-city in the pipeline source.

Reproducibility: every result on the hourly meteorology carries its provenance — the builder version and hash, the file SHA-256 hashes, the month-to-year selection, the MERRA-2 cell index and centre, and the precipitation-rescaling provenance. For the 2022 wind roses, each archetype's source year and grid cell are recorded in the provenance block of its STAR JSON. The full ingest pipeline lives in backend/tools/merra2_to_star/ with unit tests at backend/tests/test_merra2_to_star.py; raw MERRA-2 fetches are gitignored as regenerable intermediates.

2.4 Receptor grid

Two receptor grid densities are available:

• Preview mode: 5 distance rings (50, 500, 1,500, 5,000, 50,000 m) × 16 directions = 80 receptors. Designed for rapid scenario iteration.
• Final mode: 10 distance rings (50, 100, 200, 500, 1000, 2000, 5000, 10000, 20000, 50000 m) × 36 directions = 360 receptors. Designed for publication-quality output.

Both modes anchor at 0°/N. The bilinear interpolation in the frontend heat map renders smoothly at either density.

The preview grid's coarse near-source spacing (50 m → 500 m → 1,500 m → 5,000 m) under-resolves deposition gradients for coarse PSD bins (≳25 µm), whose deposition flux is concentrated within the first few hundred metres of the source and drops by several orders of magnitude before reaching the next ring. The polar-grid annular integration treats each ring's sampled flux as constant across its annular area, which is a poor approximation when the true flux varies steeply within the annulus. Final mode mitigates this with intermediate rings at 100, 200, 1,000, and 2,000 m. For analytical questions specifically about coarse-bin deposition behaviour (e.g., per-bin airborne-fraction-vs-distance), the tool also exposes a dense per-bin endpoint /api/per_bin_deposition that runs one AERMOD deck per stack PSD bin on its own denser grid (11 rings, 25 m to 50 km, × 16 directions, in either mode) — the Results graphs tab uses it for the stack curve of Graph 2 and the per-bin columns of Graph 3.

Lead still airborne (Results, Graph 2). Graph 2 shows, in kg/yr, how much of the emitted lead has not yet been deposited within each receptor ring: the lead emitted (the site's mass balance, t/yr × 1,000) less the lead deposited within the ring. The deposited lead is each ring's direction-averaged annual deposition times the area of its band, summed outward from the source. A ring's band runs from the geometric midpoint with the ring inside it to the midpoint with the ring outside it; the first band starts at the source and the last, for the 50 km ring, ends at 79 km. Graph 3 uses the same integration. Deposition is total (wet + dry) when the meteorology has precipitation, and dry otherwise. The stack curve comes from the per-bin runs: each bin's deck carries the whole stack emission at that bin's size, and the bins are weighted by their mass fractions. The fugitive curve comes from the Final run's receptors, taking the site total less the stack's contribution at each receptor. The total is the two added together. Hovering shows each value as a share of that source's emission. The curve ends at the 50 km ring, labelled with the lead still airborne beyond it, and is not extrapolated.

Caveat: the fugitive curve is sensitive near the source. In Final mode, a fifth to a quarter of the fugitive lead lands in the first ring band (0–71 m). In the abuja_2022 runs behind this section, that share was 25.5 % for the informal_artisanal default and 20.6 % for lmic_typical with the default 100 m fugitive area, both with dry deposition on the wind-rose meteorology. With total deposition on the hourly meteorology it was 27.1 % and 20.8 %. A single sample at 50 m stands for that whole band, so the curve's shape close to the source rests on it. In Preview mode the first band (0–158 m) holds more lead than the fugitive source emits: 103.1 % for lmic_typical (104.3 % on the hourly meteorology). Graph 2 therefore draws the fugitive curve and the total only for Final runs, and in Preview it asks for a Final run. If a fugitive curve still falls below zero in a Final run, it is shown with a warning and not clamped.

2.5 Validation status

End-to-end pipeline determinism was validated on 2026-05-06 by capturing the AERMOD inputs (synthetic SFC/PFL + input deck) for four scenarios via the /api/debug_dump endpoint, running AERMOD outside the deployed pipeline on those exact inputs, and comparing the locally-parsed receptor values against the deployed API's /api/run response. The four scenarios spanned both execution modes (preview 64 receptors, final 360 receptors), four emission profiles, battery throughputs from 1,000 to 23,500 tonnes/year, and both the pre-recalibration default set (three scenarios under v1.0 — LMIC bad / good / avg) and the post-recalibration set (one scenario under v1.1 — substandard at 15,000 t/y). Across 4,576 individual receptor-field comparisons (concentration and deposition, op-hours and annual averaged), every value matched to the float precision reported by AERMOD's output file. AERMOD v24142.

The pending validations are needed before this tool's outputs should be cited as authoritative. Until then, outputs should be presented as "screening estimates" with explicit uncertainty bounds.

3. Source emissions

3.1 Lead produced

Emissions follow a generation chain (backend 0.4.0 and later). For each site the user enters the lead the plant produces per year, P (t/yr; step 1 of the wizard). The tool applies an uncontrolled emission factor (lead generated per tonne of lead produced, before any capture) separately to the stack and to low-level fugitive sources, subtracts what the controls capture, and converts the emitted lead to AERMOD emission rates in g/s. No measured stack or fugitive lead emission rate exists for any secondary (ULAB) smelter in a low- or middle-income country, so the defaults are derived from regulatory coefficients, high-income analogues and judgement. They were assembled in a Pure Earth literature review (Pure Earth 2026), which covers 372 data points in about 100 sources; each number below is cited to its primary source.

Lead produced. P is entered directly and used as given: the emissions involve neither the lead content of batteries nor the lead recovery. Each quick scenario fills in a default plant size, below. A value the user types (in step 1 or in the Advanced scenario) is kept when the scenario changes, and the scenario's default is offered under the box. The wizard accepts 10–200,000 t/yr.

The source strings are those the tool shows under the box (/api/controls). The g/s are the emission rates at the default, with the scenario's other values, 8 h × 250 days and the fugitive source on. The regulated default is the US sector average: 1,140,000 t of secondary lead in 2010 (Guberman 2016, Table 1) across the 14–15 plants then operating (the 2012 rule's source category "currently includes 16 facilities, including one facility that is not currently operating and one facility that is in the process of being constructed"; US EPA 2012, 77 FR 558), about 80,000 t per plant. The LMIC default is the median annual lead-output capacity of ten Nigerian plants, 1,980–22,000 t (Öko-Institut and partners 2025, Table 2-1, which assumes one month of downtime and 62.5 % lead content, and notes "many operating below capacity"). The informal default takes sites processing 100, 500 and 1,000 t of lead a year receiving 50, 35 and 15 % of the lead processed in the informal sector (Ericson et al. 2016): 375 t of lead processed, × 0.55 recovery = 206 t of lead produced, rounded to 200.

Recovery. The lead recovery r sets only the lead in the feed and the lead lost:

lead in feed = P / r
lead lost    = P / r − P = P × (1 − r) / r

• r = the scenario's lead recovery: 99 % for the EU and US scenarios, 95 % for LMIC industrial plants (high- and low-standard plants in Manhart & Adjei 2026, Table 4) and 55 % for informal recycling (50–60 %, Wilson & Manhart 2021)

Changing r changes the loss and its not-modelled rest, not the emissions or the g/s. The web interface takes r from the scenario and does not offer it for editing; API requests can set recovery_efficiency.

Battery hint. Under the box the wizard shows roughly how many tonnes of batteries that much lead corresponds to, at the scenario's recovery:

batteries ≈ P / (0.60 × r)

• 0.60 = lead content of a battery (batteries are 58–65 % lead alloy; CEC 2013, note 39)

At the defaults that is about 135,000 t of batteries a year for the EU and US scenarios, 17,500 t for LMIC industrial plants and 606 t for informal recycling. The figure is a hint only: nothing in the chain uses it. Whether battery tonnage includes the acid is unresolved, and the hint inherits that: if a plant's battery tonnage includes the acid, lead estimated from it at 60 % could be up to 25 % too high (the International Lead Association slides (2019) imply 0.60 t of refined lead per tonne of drained batteries but 0.46 with the acid; Manhart & Adjei 2026, fn 15, note a 10–12 % weight deduction for electrolyte). Enter lead produced directly where it is known.

Stack generation: 12,000 g Pb per tonne of lead produced (sensitivity 4,800–16,900), for every industrial scenario.

G_stack = PM_gen × f_Pb × 1000
        = 24.129 × 0.50 × 1000
        = 12,065 g/t, rounded to 12,000

• PM_gen = particulate generated by the furnace before any control: 24.129 kg per tonne of crude lead for battery paste in a short rotary kiln (China MEE 2021, coefficient manual 3212, from the Second National Pollution Source Census coefficients). Fifteen of the 17 African plants assessed in 2024 ran rotary furnaces (Manhart & Adjei 2026, §2.3.2).
• f_Pb = lead fraction of that particulate, 0.50 (0.20–0.70): filter dust at African battery-recycling plants is 50–70 % lead (Manhart & Adjei 2026, §2.3.5), dusts from EU secondary lead processes 20–55 % (European Commission 2001, Table 2.2), and EMEP/EEA's factors for traditional secondary plants imply 50 % (EMEP/EEA 2009, Table 3.13). This is the most influential input: across its range the stack factor varies 3.5-fold.

US EPA's AP-42 gives higher uncontrolled factors, 32,000 g/t for reverberatory and 52,000 g/t for blast furnaces, but they date from 1986, rest on lead fractions EPA's background report could not trace, and include no rotary-furnace factor (US EPA AP-42 §12.11 and background report). The stack generation is split into particulate (90 %) and fume (10 %), each with its own capture efficiency; every default scenario sets both to the scenario's stack capture.

Fugitive generation: 2,200 g/t (sensitivity 450–7,200), for every industrial scenario.

G_fug = f_proc × G_stack + G_kettle + G_area
      = 0.10 × 12,065 + 13 + 1,000
      = 2,220 g/t, rounded to 2,200

• f_proc = process fugitives (furnace charging, tapping and slag handling) as a fraction of stack generation, 0.10 (0.05–0.13). AP-42 assumes 0.05 (AP-42 §12.11, Table 12.11-3); two US primary smelters had 0.057 and 0.127 in 1979 (US Bureau of Mines IC 9160). The upper half of the range reflects manual charging at 5 of 17 African plants (Manhart & Adjei 2026, §2.3.2).
• G_kettle = refining kettles and casting without hoods, 13 g/t (AP-42 §12.11, Table 12.11-1). Some African refineries have no kettle hoods (Manhart & Adjei 2026, §2.3.3).
• G_area = battery breaking, storage, materials handling, spilled filter dust and vehicle resuspension, 1,000 g/t (200–5,000). [Judgement]: no LMIC measurement exists. It sits between open fugitives at US primary smelters in 1979, 639–1,163 g/t uncontrolled (US Bureau of Mines IC 9160), and fugitives at Chinese primary smelters after 2010, 1,000–2,000 g/t (Wang & Chen 2017).

Informal and backyard smelting has no stack: all its generated lead, 20,000 g per tonne of lead recovered (5,000–70,000), is fugitive (see section 3.2).

Emitted lead and the mass balance. Emitted lead, g/yr, is the generation (g per t of lead produced) less what is captured, times the lead produced (t/yr):

E_stack = [G_stack × (1 − c_stack)
           + G_fug × c_fug × (1 − c_stack)] × P
E_fug   = G_fug × (1 − c_fug) × P

• c_stack = fraction of the lead reaching the stack's control device that the device removes
• c_fug = fraction of fugitive lead captured by hoods or enclosure. Captured fugitive lead is ducted to the stack's control device, so it leaves through the stack after abatement rather than vanishing (routing). It adds 18 % to stack emissions in the EU and US scenarios and 9 % in the well-run LMIC scenario.

Captured stack lead is assumed to return to the furnace. The lead lost, P × (1 − r) / r, bounds the emitted total: inputs that would emit more than that are refused. The rest of the loss (slag, spills, acid and ground) is reported as not modelled. Slag alone carries an estimated 9,000–36,000 g/t at low-standard plants (Manhart & Adjei 2026, Table 1 and §2.3.7; Öko-Institut and partners 2025).

Emission rate for AERMOD.

Q = E / (3,600 × h × d)

with E in g/yr, calculated separately for the stack and fugitive sources, where h × d is the operating schedule in hours per day × days per year (default 8 h × 250 days = 2,000 h/yr; section 3.4). Users can instead enter the stack and fugitive emission rates directly in g/s.

Worked example: LMIC industrial – typical at its default. P = 10,000 t/yr; 95 % stack capture, no fugitive capture (section 3.2).

E_stack = 12,000 × (1 − 0.95) × 10,000 = 6,000,000 g/yr  (6 t)
E_fug   =  2,200 × (1 − 0)    × 10,000 = 22,000,000 g/yr (22 t)
Q_stack =  6,000,000 / (3,600 × 8 × 250) = 0.833 g/s
Q_fug   = 22,000,000 / (3,600 × 8 × 250) = 3.06 g/s
lead in feed = 10,000 / 0.95 = 10,526 t; lead lost = 526 t,
  of which 498 t (94.7 %) not modelled

For API users. A request sends lead_produced_tpy (t of lead produced per year) or the earlier battery-basis throughput_tpy (t of batteries per year, converted as lead produced = throughput × 0.60 × r), never both: a request with both is refused (422 emission_inputs_conflict). A result whose emission_assumptions carry input_basis: "lead_produced" is on the lead basis; a result without input_basis is on the battery basis. The legacy chain below takes throughput_tpy only.

Legacy chain. API requests that set f_fug or emission_rate_g_s still use the earlier chain until backend 0.5.0, with a deprecation warning: 12 kg of lead per tonne of battery feed (about 21,000 g per tonne of lead produced), split between stack and fugitive by f_fug. The origin of the 12 kg/t factor could not be traced (see the 1.12 entry in section 8). The tool's web interface does not use it.

3.2 Emission controls and default scenarios

Seven quick scenarios share the industrial generation factors (12,000 g/t stack, 2,200 g/t fugitive) and differ in capture, recovery and default plant size (section 3.1); recovery sets only the loss, not the emissions. LMIC industrial – typical is the default, because it matches what assessments of African plants observed: all 17 plants assessed in Ghana, Nigeria and Tanzania had baghouses, but "some off-gas treatment systems were clearly not fully functional and well maintained", with visible particle emissions from stacks at various plants, and furnaces were often unhooded or had weak extraction (Manhart & Adjei 2026, §2.3.5 and §2.3.2). "Stack capture" sets both the particulate and the fume capture. Emitted values are g Pb per tonne of lead produced, computed from the scenario values by the chain in section 3.1.

Basis of each scenario.

• EU BAT. State-of-the-art fabric filter, 99.99 % (EMEP/EEA 2023, Table 3-8). Fugitive capture set so the total lands between the EU-28 average, 1.1 g/t (EMEP/EEA 2023, Tier 2), and Germany’s inventory, 4.5 g/t (UBA 2026).
• High-income regulated (US 2010). Capture values set to reproduce US secondary-smelter emissions around 2010; the scenario emits 9.94 g/t from the stack and 5.5 g/t fugitive, 15.44 g/t in all. See the consistency check below.
• LMIC industrial – well run. Stack: the low end of design fabric-filter efficiency, 98.4 % (AP-42 §12.11, Table 12.11-5), within measured US baghouses, 85–99 % (US EPA 1995, p. 2-29). Fugitive: partial hooding, 51.7 % at Buick in 1979 (US Bureau of Mines IC 9160, Table A-1). Capture sensitivity 95–99 % [judgement].
• LMIC industrial – typical. [Judgement] A poorly maintained baghouse, equal to an outlet of about 175 mg/Nm³; such systems, and visibly emitting stacks, were seen at several of 17 plants in Ghana, Nigeria and Tanzania (Manhart & Adjei 2026, §2.3.5). Fugitive: hoods absent or ineffective (§2.3.2). Capture sensitivity 90–98 %.
• LMIC industrial – degraded baghouse. [Judgement, 70–90 %] The low end of measured US baghouses, 70 % (US EPA 1995), and of dry electrostatic precipitators, 54–95 % (EMEP/EEA 2023, Table 3-8).
• Uncontrolled / bypassed. Generation only: no working stack control, or the baghouse bypassed, and no hoods.
• Informal / backyard. No stack. 20,000 g/t of lead recovered released at ground level: about 2 % losses (International Lead Association slides, 2019; figure given without a source), range 5,000–70,000 g/t; recovery 50–60 % (Wilson & Manhart 2021).

A well-maintained baghouse removes 98–99 % of furnace particulate, 98.4–99.8 % by design (AP-42 §12.11, Table 12.11-5) and 99 % for a bag filter (China MEE 2021, manual 3212), but measured removal at working US plants ranged from 85 % to 99 %, and 70 % at one plant with a very dilute inlet (US EPA 1995, EPA-453/R-95-008b, p. 2-29). Percent capture depends on how dusty the gas entering the filter is: rotary-kiln flue gas carries about 3,500 mg of particulate per Nm³ (China MEE 2021), so a baghouse emitting 175 mg/Nm³ removes 95 % and one emitting 700 mg/Nm³ removes 80 %. No LMIC baghouse has been measured; the 95 % and 80 % values are judgements. The capture sensitivity ranges (with the generation ranges in section 3.1) are served by /api/controls and not yet shown in the interface.

Cross-checks. The typical scenario's stack emission, 600 g/t, falls within EMEP/EEA's 450–1,100 g/t for traditional secondary plants with limited control (EMEP/EEA 2009, Table 3.13). Its total, 2,800 g/t, is far above controlled US and Chinese plants (2–20 g/t stack caps; GB 31574-2015, Henan DB41/2806-2025, Chongqing EIA approval 2024) and similar to controlled US primary smelters in 1979, 1,460–5,550 g/t (US Bureau of Mines IC 9160). Its fugitive share, 79 %, sits between those 1979 smelters (38 % at Glover, 76 % at Buick) and Port Pirie's 66–93 % (Australian National Pollutant Inventory).

High-income regulated: consistency check. The scenario's capture values were set to reproduce US secondary-smelter emissions around 2010. As a consistency check, not the calibration source, the 2012 standards for the industry state: "The final rule will cut lead and arsenic emissions by an estimated 68 percent from current actual emission levels based on the ICR data collected for this rulemaking. The final rule will result in estimated annual lead emissions reductions of 7.2 tpy from process and process fugitive sources and annual lead emissions reductions of 6.4 tpy from fugitive dust sources from 2009 baseline emissions (for a total annual reduction of 13.6 tons per year)." (US EPA 2012, 77 FR 575). These imply about 20 short tons (about 18 t) of lead a year before the 2012 standards. That is about 16.3 g/t of 2009 US secondary lead production, 1,110,000 t (Guberman 2016, USGS 2012 Minerals Yearbook, Table 1, myb1-2012-lead.pdf), close to the scenario's 15.44 g/t (9.94 g/t stack + 5.5 g/t fugitive). The rule's baseline year is 2009, not the scenario's 2010, and its 68 % is stated for lead and arsenic together.

3.3 Particle size distribution (PSD)

AERMOD's default particle deposition method (Method 1) needs a diameter, a mass fraction and a density for each size class (Thurman 2019). Every scenario has a stack PSD and a fugitive PSD over the same six size classes, each represented by its geometric midpoint:

Density 1.0 g/cm³ in every class. All the size data behind the defaults are aerodynamic diameters, and an aerodynamic diameter is the "diameter of a spherical particle with relative density equal to unity that has the same settling velocity in air as the particle in question" (IUPAC 2007). It already includes the particle's density, so entering the densities of the lead phases found at recycling plants (Uzu et al. 2011), from PbSO₄ at 6.3 g/cm³ (Webmineral) to lead metal at 11.3 g/cm³ (Royal Society of Chemistry), as well would count density twice and make particles settle too fast. The AERMOD User's Guide asks for the "mass-mean aerodynamic particle diameter" of each class (US EPA 2024, EPA-454/B-24-007, §3.3.4.1), the guide for AERMOD version 24142, which the tool runs. The 0.5 µm class sits within the fine-mode lead peaks of 0.32–0.55 µm measured near smelters (Csavina et al. 2014).

Each cell is the percentage of emitted lead mass in that size class; the decks take them as fractions, as given. A class with no mass is left out of that source's deck lines.

Stack PSDs (% of mass by size class)

Fugitive PSDs (% of mass by size class)

The Particle tab offers the six stack presets and a custom stack PSD (below); the Fugitive tab shows the scenario's fugitive PSD, with editable fractions (fractions that do not sum to 1 are rescaled, and the tool says so). Informal smelting has no stack; its stack deck runs at 0 g/s with the uncontrolled row. With directly entered emission rates and no scenario, the fugitive source takes fug_open.

Custom stack PSD. Choosing "Custom" in the Particle tab lets the user enter their own stack PSD; the fugitive PSD stays the scenario's.

• Classes. The six classes above are fixed, with diameters of 0.5, 1.6, 5.0, 12.2, 21.2 and 54.8 µm. They are aerodynamic diameters, so the density stays 1.0 g/cm³ in every class. The user enters only the percentage of mass in each class.
• The 100 % rule. Each class takes 0–100 % with at most one decimal, and Run is disabled unless the classes total exactly 100.0 %. The tool shows what is left to assign or over, and rescales nothing. The backend checks the request independently: six fractions, each 0 or between 0.0001 and 1, summing to 1 within 0.000001. Anything else is refused, with a message giving the sum received. AERMOD's own check is looser: it only warns when the fractions sum outside 0.98–1.02, and stops on a fraction outside 0–1.
• The deck. As for the presets, a class at 0 % is left out of the deck lines, and the other classes keep their diameters and density. A custom PSD equal to a preset writes that preset's deck exactly, and gives the same results.
• Scenarios and records. Choosing Custom starts from the stack PSD selected before. A custom PSD is kept when the scenario changes, and the tool offers the scenario's preset instead. Results, exports and the Verification tab record it as "custom (user-entered)" with its six values. Runs with a preset PSD are unchanged.
• Method 1. Method 1 is appropriate for any custom PSD: AERMOD guidance recommends it when the particle size distribution is known, as well as when more than about 10 % of the mass is at 10 µm or larger (US EPA 2024, EPA-454/B-24-007, §3.3.4).

Where the rows come from.

• Uncontrolled stack (stack_raw_furnace): the only measured PSD of uncontrolled secondary-lead furnace gas, blast-furnace charging and tapping gas, with 35 % of mass below 2.5 µm and 60.5 % above 10 µm (AP-42 §12.11 background report, Table 2.3-3, rated D; total particulate, of which lead was 34 %).
• Working baghouse (stack_bat, for EU BAT and high-income regulated, whose capture exceeds the working-filter baseline below): the PSD after a baghouse, 71 % below 2.5 µm and 11 % above 10 µm (same report, Table 2.3-4, rated D). A baghouse removes coarse particles more completely than fine ones, so its output is fine.
• LMIC baghouses (98 %, 95 % and 80 % capture): intact filter fabric passes only 10⁻⁵ to 10⁻⁶ of the dust mass, so capture of 80–98 % implies gas getting past the fabric through torn or missing bags, poor seals or an open bypass (Kurtz et al. 2017). Each is modelled as a working filter (99.26 % capture, from the paired totals in Table 2.3-3) plus an open path. [Assumption] The open path carries equal mass of raw gas (bypass) and raw gas below 5 µm (small leaks, which in pilot tests passed only particles below about 5 µm; Kurtz et al. 2017), because no study shows which dominates. The open path then carries 63 %, 86 % and 97 % of emitted mass, so 23–30 % of LMIC stack mass is above 10 µm, within bounds of under 5 % (small leaks only) and 59 % (bypass only).
• Sensitivity row (stack_fine_raw_gas_sensitivity, not a scenario default): EMEP/EEA's unabated split (EMEP/EEA 2023, Table 3-4), each class divided as in Table 2.3-3. Vaish et al. (1998) and EMEP/EEA put about 60 % of uncontrolled mass below 2.5 µm, against 35 % in the default; if rotary-furnace gas is mostly fume, stacks emit finer lead that deposits less near the plant.
• Coarse tail. Table 2.3-3 puts everything above 15 µm in one open-ended bin. That mass is treated as lognormal, with the 18.9 µm mass median aerodynamic diameter AP-42 gives for kettle particulate (AP-42 §12.11, p. 12.11-9) and a geometric standard deviation of 2.4 measured for coarse dust at a primary smelter (Froines et al. 1986, Table II): 50.5 % of it goes to the 15–30 µm class and 49.5 % to 30–100 µm, including the 4.7 % above 100 µm. The rule applies to stack gas, process fugitives and informal smelting, not to yard and road dust.
• Open plant fugitives (fug_open, for typical, degraded and uncontrolled, which capture no fugitives): process fugitives and kettles (1,213 g/t of the 2,200) take the charging and tapping PSD of Table 2.3-3, consistent with AP-42's 18.9 µm MMAD for uncontrolled kettle particulate. Area fugitives (1,000 g/t) take an equal mix of AP-42's size multipliers for material drop (AP-42 §13.2.4) and unpaved industrial roads at 12 % silt (AP-42 §13.2.2); those multipliers stop at 30 µm and do not split mass below 2.5 µm, which goes in the 1–2.5 µm class. Combined, 21.5 % is below 2.5 µm and 61 % above 10 µm.
• Partial hooding (fug_partial_hood, well run, 50 % fugitive capture): [assumption] hoods and housekeeping remove coarse particles more readily. The open-plant PSD is reduced by the relative efficiencies of good-practice fugitive dust control in China's inventory, 70 %, 50 % and 30 % above 10, 2.5–10 and below 2.5 µm (Lei et al. 2011, Table 5), scaled to 61 %, 43 % and 26 % so that capture totals 50 %.
• Enclosed plant (fug_enclosed, EU BAT and high-income, 99.9 % and 99.75 % capture): [assumption] the residual is paved-road dust, 81 % above 10 µm (AP-42 §13.2.1, Table 13.2.1-1). US rules require total enclosures under negative pressure for process fugitive sources (US EPA 2012, 77 FR 556).
• Informal smelting (fug_informal): no building or hood; the uncontrolled charging and tapping PSD, the nearest measured analogue for open-air melting.

Method 1 is recommended when more than about 10 % of the mass is 10 µm or larger, or when the particle size distribution is known (US EPA 2024, EPA-454/B-24-007, §3.3.4). Every default meets the first condition. US guidance leaves deposition modelling at lead smelters to professional judgement (40 CFR Part 51, Appendix W, §4.2.2.1(e) and §4.2.3.2(b)), so the tool's deposition results are screening estimates, not regulatory ones.

PSD limitations. Every stack PSD rests on undated US blast-furnace tests (Tables 2.3-3 and 2.3-4, both rated D); no PSD was found for rotary or reverberatory furnaces, or measured at an LMIC plant. All the PSDs describe total particulate, and lead was more concentrated in the fine fraction at every smelter site Csavina et al. (2014) sampled except Port Pirie, so lead may be finer than the defaults. What escapes a leaking or bypassed industrial baghouse has not been measured; field signs of gross bypass (open bypass dampers, missing bags) favour the coarser bypass-only case. No study gives hood or building capture by particle size, so the well-run and high-income fugitive PSDs are assumptions. Battery paste and spilled filter dust may be finer than the aggregate and road dust used for area fugitives.

3.4 Operating duty cycle

Real ULAB facilities operate intermittently (typical: 8 hours/day × 250 days/year = 2,000 hours/year). The tool accepts user-specified operating hours and applies a duty-cycle factor:

duty_cycle = (op_hours_per_day × op_days_per_year) / 8760

Two concentration fields are surfaced per receptor: the period-mean concentration during operating hours (no duty-cycle scaling — the concentration a child breathes when the facility is emitting) and the annual time-average across the full year (period-mean × duty cycle — the concentration averaged over all 8,760 hours including non-operating periods). The annual time-average drives the BLL pathway because IEUBK is calibrated for chronic ambient exposure; the during-operating-hours mean is exposed for users assessing what the air is like when the plant is running. Deposition fields are scaled by duty cycle in the same way. The 1-HR worst concentration field (section 2.1) is independent of duty cycle — it is the single worst modelled hour, irrespective of whether that hour falls inside or outside the operating window.

3.5 Fugitive emissions

Fugitive emissions are lead released at low level: furnace charging, tapping and slag handling; refining kettles and casting; battery breaking; storage, handling, spilled filter dust and vehicle resuspension. Published fugitive estimates for uncontrolled or poorly controlled plants span about 100–15,000 g/t, and none comes from an LMIC secondary smelter. Where baghouses work, fugitives dominate near-field concentrations: fugitive dust gave 94–99 % of modelled peak 3-month lead at the 9 of 15 US smelters exceeding the air quality standard (US EPA 2012, 77 FR 562). Direct field measurements distinguishing per-process fugitive sources are unavailable for LMIC ULAB plants, so the tool lumps them into one source.

Amount. Fugitive generation is 2,200 g per tonne of lead produced for every industrial scenario (450–7,200; section 3.1), and 20,000 g/t for informal smelting. The scenario's fugitive capture (hoods and enclosure) removes part of it at source; captured lead is routed through the stack's control device and emitted from the stack after abatement. What is not captured, E_fug = G_fug × (1 − c_fug), is the fugitive source. With directly entered rates, the fugitive source takes the rate entered.

Switching it on. The fugitive source is modelled only when the user turns it on (the Fugitive tab or the wizard's step 03). When it is off, the results state the fugitive emission left out and the share of emitted lead the stack-only run models. Informal smelting, and any inputs that give no stack emission, have nothing else to model, so for them the fugitive source is always on.

Source geometry. Fugitives are modelled in AERMOD as a square AREA source centred on the stack location. Default geometry: 100 m side length, 2.5 m release height, 1.5 m initial vertical dimension (σ_z0), all user-editable.

Particle size. The fugitive PSD is the scenario's six-class fugitive preset (section 3.3): coarse in every scenario, with 48–81 % of mass above 10 µm, and density 1.0 g/cm³ because the diameters are aerodynamic. This replaces the earlier four-bin fugitive PSD with mineral-matrix densities of 2.8–4.0 g/cm³, which the legacy chain still uses. Requests that send four fugitive fractions are refused with a message to reload or send six.

Area-source integration (FASTAREA). Since release 0.3.0 the fugitive AREA source is modelled with AERMOD's FASTAREA model option (US EPA 2024, EPA-454/B-24-007, §3.2.2.8; the tool runs AERMOD version 24142) instead of the default Romberg numerical integration over the source area (same guide, §3.3.2.3); stack (POINT) sources are unaffected. FASTAREA is a non-default AERMOD option: the fugitive run's output header lists it and labels the run NonDFAULT, while the stack run uses neither (neither run sets AERMOD's regulatory DFAULT keyword, so both headers read “Model Allows User-Specified Options”). Deleting FASTAREA from the MODELOPT line of the exported fugitive deck reproduces the default area integration offline. This makes the fugitive run about 2.6 to 14 times faster, so every fugitive run, Preview and Final, is modelled on the mode's full receptor grid (Preview 80 receptors, Final 360) with no interpolation. Compared with the default integration across 29 tested cases, beyond 500 m from the source period-average concentration and deposition change by less than about 0.25 %. The 1-hour worst concentration moves more there: by up to about 7 % in a ring's maximum (fugitive run alone), while every change stays below 0.3 % of its maximum (below 0.1 % except for two 300 m areas with the newer built-in meteorology). Near the source FASTAREA reads somewhat higher for small, low, vertically thin sources: for a 50 m area with a 1 m release height, fence-line (50 m) soil Pb and ΔBLL were 1.7 % to 5.5 % higher, the worst case (+5.5 % soil Pb) at an initial vertical dimension σ_z0 of 0.5 m; in the other tested cases fence-line soil Pb, ΔBLL and ΔIQ changed by less than ±0.8 %. Near-source changes of a few percent, in either direction, can move receptors across a blood-lead threshold: for Abuja, substandard, default fugitive setup, Final, the number of receptors above 5 µg/dL went from 64 to 61, as three 100 m receptors fell from just above 5 µg/dL to just below it (changes of 2.3 % to 4.4 %). Tested range: area side 30–300 m, release height 1–5 m and σ_z0 0.5–3.0 m (the full input ranges); the default and one finer fugitive particle-size distribution; the four built-in locations with synthetic and MERRA-2 meteorology, and one user-built wind rose; one meteorological year; one site, and two sites 2 km apart. Final mode with fugitive emissions runs online for single-site scenarios; multi-site Final with fugitive emissions is not available online (export the AERMOD inputs and run them offline).

Status of fugitive defaults. The 2,200 g/t generation, its 1,000 g/t yard-dust component and the 10 % process-fugitive share are screening-grade, and partly judgements; outdoor manual battery breaking and truck resuspension are unquantified, and no published figure exists for outdoor manual battery breaking (China's coefficient manuals give no air coefficient for breaking; China MEE 2021, manual 42). The fugitive PSDs of the well-run and high-income scenarios are assumptions. Before tool outputs with fugitives enabled are used in advocacy or regulatory contexts, run the generation and capture sensitivity ranges in sections 3.1 and 3.2. As with stack emissions, the most reliable path to higher-confidence fugitive outputs is field measurement at representative LMIC sites: near-field deposition inversion around the Ghana smelters could give the first LMIC fugitive estimate (see Pure Earth's research agenda).

Note on 1-HR worst summing. When both the stack and fugitive AERMOD runs report a 1-HR worst-hour concentration at the same receptor, the combined receptor's 1-HR worst is the arithmetic sum of the two sources' values. This is an upper-bound approximation: the worst hour for the stack source may not coincide temporally with the worst hour for the fugitive source, so the true joint maximum could be lower. Resolving the joint maximum exactly would require hour-by-hour data per source; the conservative sum is accepted as a screening-grade approximation, bounded above by 2× for two independent sources. PERIOD-averaged concentration / deposition / BLL / IQ values are not affected — they sum exactly because the averaging window covers all hours of both sources.

4. Soil accumulation

4.1 Mass-balance model

The tool estimates the increment in topsoil lead concentration caused by atmospheric deposition over a user-specified operational duration, assuming:

• All deposited lead remains in the surface soil layer of specified depth (no leaching, no erosion)
• The soil layer mixes uniformly within itself (no vertical gradient)
• Background soil lead is independent and additive

ΔPb_soil (mg/kg) = (D × T) / (d × ρ × 10)

Where:

• D = annual deposition rate (mg/m²/yr): total (wet + dry) where precipitation is modelled (the built-in locations), dry otherwise; see the wet share by distance in section 2.2
• T = operating years
• d = soil depth (cm)
• ρ = bulk density (g/cm³, default 1.3 — typical mineral surface soil at the Soil tab's exposure framing)

The Soil tab and the BLL pathway both compute soil-Pb via this mass balance over the same operational-years duration (default 20 yr, set by the "Smelter operational years" control on the Soil tab), but apply different mixing depth and bulk density appropriate to their framings: the Soil tab targets agricultural / root-zone questions (15 cm depth, ρ = 1.3 g/cm³); the BLL pathway targets the IEUBK child-residential contact layer (1 cm depth, ρ = 1.5 g/cm³ — see section 5.3). Same accumulating deposition and duration; the depth/density differ because they answer different questions about the same soil column.

Validity cap. Predicted soil-Pb is clamped at 80,000 mg/kg on both pathways. This is the upper bound of soil-Pb observed at the most severe documented secondary-smelter sites (e.g. Doe Run Herculaneum surface soils; ASARCO El Paso). The linear-accumulation model produces non-physical values above this — at very high deposition rates, deposited Pb begins to accumulate in distinct surface phases rather than mixing uniformly within the bulk-density-defined soil column, so the model's uniform-mixing assumption fails. Receptors past the cap are flagged for off-scale display rather than presenting a numeric value (see section 5.4).

4.2 EPA bin thresholds

The soil heat map is anchored to US EPA action levels:

4.3 Limitations

The "no leaching, no erosion" assumption is conservative for typical soils on decadal timescales but overestimates accumulation on:

• Sandy, high-permeability soils with significant rainfall infiltration
• Steeply sloped or actively eroding sites
• Tilled agricultural soils (homogenisation extends mixing depth)

Users should select soil depth appropriate to the exposure pathway:

• Direct ingestion (children playing in dust): top 1 cm
• Vegetable garden uptake: top 15 cm (root zone)
• Cereal/staple crops: top 30 cm (tilled depth)

The linear-accumulation assumption also breaks down at the upper end of the deposition range, where the soil column cannot homogeneously incorporate the deposited mass. The 80,000 mg/kg cap (section 4.1) is the model's explicit validity boundary; receptors past it carry no numeric soil-Pb prediction. The cap is anchored to the highest documented secondary-smelter surface soils, not to a theoretical absorption limit. Above the cap, the receptor's spatial location is still shown on the map (with an off-scale visual marker) so users see where severe contamination occurs even when the absolute concentration cannot be honestly predicted.

5. Blood lead level (BLL) impact estimation

5.1 Conceptual framework

This tool estimates the increment in child blood lead level (ΔBLL) attributable to a specific facility's emissions, decomposed into three pathways:

• Inhalation — direct uptake of airborne lead. Driven by the receptor's annual mean ambient PbA concentration. Anchored to direct IEUBK biokinetic-model outputs (section 5.2).
• Soil / dust — incidental ingestion of accumulated atmospheric deposition. Driven by the cumulative deposition over the smelter's operational years (default 20 — the "Smelter operational years" control on the Soil tab, shared with the soil-accumulation model), apportioned across the IEUBK contact depth (1 cm). Linear in this duration. Anchored to direct IEUBK biokinetic-model outputs (section 5.3).
• Crop (foliar deposition) — ingestion of edible crops grown on receptor land, with deposited Pb retained on foliage and partially translocated to the edible portion. Screening-grade implementation of the food-crop foliar-uptake form with a composite transfer fraction R (Brown 2008 UK FSA / EC JRC TGD 2003 / US EPA OSWER 2002 for leafy; + Liu 2020/2022 paired-measurement empirical tuning for cereal). Converted to ΔBLL via a single linear child-dietary slope (EFSA 2010 CONTAM Panel). Off by default on the ΔBLL surface — see section 5.5.

The inhalation and soil pathways use direct outputs from the EPA IEUBK biokinetic model, anchored at concentration points (8 for inhalation, 6 for soil/dust) and interpolated piecewise in log-log space between anchors. The crop pathway is computed independently and added to the running total in linear ΔBLL space. The response carries two totals — delta_bll_total_excluding_foliar_ug_dl (air + soil) and delta_bll_total_including_foliar_ug_dl (air + soil + crop) — and the corresponding validity flags for each; the BLL surface and the ΔIQ gating switch between them based on the user's "Include crops" checkbox on the BLL panel (section 5.4). Other potential pathways (water, paint, occupational, dietary contamination from non-aerial sources) are not modelled — this tool isolates the contribution attributable to the modelled facility, not the population's total BLL. Implementation lives in backend/app/bll.py + backend/app/crops.py, with JS mirrors at frontend/bll.js + frontend/app.js.

5.2 Inhalation pathway

The IEUBK model was run directly across the relevant outdoor PbA range, assuming indoor air = 30% of outdoor air (US EPA convention for residential exposure). Eight IEUBK anchor points span the range from 0.15 to 150 µg/m³, with denser sampling at the lower end where most modelled receptor concentrations fall:

For an arbitrary PbA between two anchors, ΔBLL is computed by piecewise log-log linear interpolation:

log(ΔBLL) = log(BLL_lo) + t × (log(BLL_hi) − log(BLL_lo))
where t = (log(PbA) − log(PbA_lo)) / (log(PbA_hi) − log(PbA_lo))

Below 0.15 µg/m³ outdoor air, IEUBK predictions are below the resolution at which the model produces meaningful output. The tool linearly extrapolates toward zero in this regime (slope ≈ 0.67 µg/dL per µg/m³); values below 0.1 µg/dL ΔBLL fall outside the heat map's display range. At and above 150 µg/m³ the function returns the saturation cap of 30 µg/dL and raises an air-saturation flag — see Section 5.4.

5.3 Soil pathway

Atmospheric deposition is converted to a topsoil Pb concentration via a simple mass balance:

Pb_soil (mg/kg) = D × t / (ρ × z)

where D is annual deposition (mg/m²/yr; total, wet + dry, where precipitation is modelled, dry otherwise), t is exposure duration (yr), ρ = 1500 kg/m³ is bulk soil density (typical mineral surface soil), and z is mixing depth (m). The resulting soil-Pb concentration is then mapped to ΔBLL via direct IEUBK output anchors:

IEUBK assumes default soil ingestion rate and lead bioavailability for 6–7 yr-old children in a residential setting. Interpolation is piecewise log-log linear, identical in form to the inhalation pathway. Below 50 mg/kg the function extrapolates linearly through the origin (slope 0.002 µg/dL per mg/kg). At and above 1600 mg/kg the function returns the saturation cap of 3.2 µg/dL and raises a soil-saturation flag.

Unified exposure duration. The exposure duration is shared across the soil-accumulation model and the BLL soil pathway: a single control — "Smelter operational years" (t, default 20 yr) on the Soil tab — drives both. The BLL soil pathway's mixing depth is fixed at z = 1 cm (the IEUBK contact-relevant residential contact depth — the surface dust layer for incidental hand-to-mouth ingestion) and is no longer separately user-tunable. The soil-Pb intermediate scales linearly with t up to the 80,000 mg/kg cap, but ΔBLL is not linear in t because the IEUBK soil curve is non-linear and saturates at its anchor table top (1,600 mg/kg → 3.2 µg/dL ΔBLL).

The same operational-years value feeds the Soil tab's own topsoil heat map, which applies a different depth and density (z = 15 cm, ρ = 1.3 g/cm³ — the agricultural root zone) to answer a different physical question about the same soil column. Same accumulating deposition over the smelter's operating life, same duration; the two views differ only in the soil layer each cares about — the child's surface contact layer vs. the agricultural root zone. Unifying t is correct because the smelter operates over years, and that accumulating Pb is seen by both layers over the same period.

Soil-Pb validity cap. The intermediate soil_pb_mg_kg is clamped at 80,000 mg/kg, matching the documented upper bound of surface-soil Pb at the most severely contaminated secondary-smelter sites (Doe Run Herculaneum; ASARCO El Paso). The cap reflects the linear-accumulation model's validity boundary, not a physical maximum. Receptors whose raw soil-Pb would exceed the cap are flagged (soil_pb_capped) for off-scale visual treatment — see section 5.4. The IEUBK soil-pathway BLL contribution from these receptors is unaffected by the cap (the IEUBK table itself saturates at 1,600 mg/kg, well below 80,000); only the reported soil-Pb number is suppressed.

The inhalation pathway is independent of t and z. Moving the Smelter operational years slider re-runs both the Soil-tab accumulation map and the BLL soil pathway (and the combined total + ΔIQ derived from it) in JS, with no AERMOD re-execution.

5.4 Combined ΔBLL and model validity treatment

The tool tracks two facility-attributable ΔBLL totals per receptor, so the user can switch the rendered BLL surface and the downstream ΔIQ gating between an air+soil-only framing and an air+soil+crops framing without re-running AERMOD:

ΔBLL_total_excluding_foliar = ΔBLL_air + ΔBLL_soil
ΔBLL_total_including_foliar = ΔBLL_air + ΔBLL_soil + ΔBLL_foliar

The "Include crops" checkbox on the BLL Impacts panel flips between the two. Both totals are computed on every receptor; the checkbox only changes which one is used for the surface, the summary, the validity gating, and the ΔIQ recompute.

Each receptor is classified into one of four states based on where its inputs and outputs fall relative to the underlying models' validity ranges. The state determines what the map and summary panel show. The validity flag follows the active total: when crops are off the BLL surface, the excluding-crop validity flag (bll_validity_exceeded_excluding_foliar) drives state assignment; when crops are on, the including-crop flag (bll_validity_exceeded_including_foliar) drives it. Both flags are present on every response so consumers can audit either framing.

States 2, 3, and 4 are independent: a single receptor can be in multiple states simultaneously (typically state 3 + state 4 near-source for large informal scenarios). On the map, the off-scale hatched-square marker takes visual priority over the ✕ marker when both would apply.

State 2 — input-table extrapolation. The IEUBK anchor tables for both pathways cover PbA up to 150 µg/m³ (air) and soil-Pb up to 1,600 mg/kg (soil). Above either anchor table top, the pathway's BLL contribution is held at the table's saturation value — physiologically motivated by gastrointestinal absorption saturation (the fraction of ingested Pb absorbed into circulation falls as ingestion load rises) and erythrocyte binding saturation (Pb bound to red blood cells stops increasing linearly with circulating Pb). The total ΔBLL is still a meaningful biokinetic estimate provided it sits below the output validity threshold; the ✕ marker signals "input extrapolated, output still valid".

State 3 — BLL past model validity (≥ 30 µg/dL). The biokinetic model underpinning ΔBLL prediction (IEUBK, which handles both the inhalation and soil/dust pathways) is validated for chronic-exposure BLLs up to approximately 25–30 µg/dL. Above 30 µg/dL the model's steady-state assumptions cease to hold: at these BLLs a child is in the clinical-evaluation regime rather than chronic-exposure dynamics, and the linear/log dose-response curves leave their validity range. ULAB Plume's 30 µg/dL threshold is the standard defensible upper bound for IEUBK output validity; the CDC clinical chelation threshold is 45 µg/dL, placing the validity boundary between chronic-exposure relevance and acute toxicity. For receptors in this regime:

• The numeric ΔBLL is suppressed on the map and in tooltips — displaying it would be false precision past the model's predictive range.
• The receptor is shown with a hatched off-scale marker so users still see where these receptors are located, even when the magnitude cannot be honestly predicted.
• ΔIQ is capped at the IEUBK validity ceiling — computed as if ΔBLL = 30 µg/dL, returned as a normal numeric value on the API response, and shown in tooltips with a '≥' prefix and a 'model validity exceeded' note. Rather than extrapolate Crump 2013 up an unbounded curve (over-stating) or suppress to null (which holed the ΔIQ surface over the worst-exposed receptors), the tool reports the minimum IQ loss consistent with reaching the ceiling.
• The BLL summary panel surfaces a categorical count: "N receptors in saturated zone (≥ 30 µg/dL) — biokinetic model validity exceeded. In the clinical acute-toxicity regime; CDC clinical chelation threshold is 45 µg/dL." The highest ΔBLL among receptors where the model is valid is reported as a separate subnote.

ΔIQ at the validity ceiling (cap, not suppression). Unlike the numeric ΔBLL — suppressed for State-3 receptors because the biokinetic model cannot predict its magnitude — ΔIQ for these receptors is capped: Crump 2013 is evaluated as if ΔBLL were exactly 30 µg/dL, giving a fixed floor of ≈7 IQ points lost at the default background BLL and central CI band; the exact value shifts with background BLL and band. The heat map renders these receptors at the ceiling value and the cursor shows '≥ X.X IQ pts — model validity exceeded'; the hatched off-scale marker still flags them. A child whose facility-attributable ΔBLL has crossed 30 µg/dL has at least the loss the model predicts at 30, so the ceiling is an honest lower bound — communicating 'at least this severe' without the holes that null-suppression punched over the highest-exposure receptors.

State 4 — Soil-Pb past model validity (≥ 80,000 mg/kg). Independently of the BLL output validity, the linear-accumulation soil model itself has a validity boundary at 80,000 mg/kg (see section 4.1). Receptors past this cap carry no numeric soil-Pb prediction but are still shown on the map with a hatched marker on the soil layer. Because the IEUBK soil table already saturates at 1,600 mg/kg (well below the 80,000 cap), the soil-pathway BLL contribution is unaffected by the cap — only the displayed soil-Pb number is suppressed.

The four-state classification is part of the methodology, not just a UI choice. It is the mechanism by which the tool avoids over-claiming: where the underlying physics, biokinetics, or dose-response models cannot honestly predict a magnitude, the tool refuses to fabricate one — suppressing the number (soil-Pb, ΔBLL) or capping it at the model's validity ceiling and labelling it '≥' (ΔIQ), always with an off-scale marker on the receptor. Users see the spatial extent of "off-scale" exposure without the tool fabricating precise values past its validity range.

5.5 Crop pathway (foliar deposition)

The crop pathway estimates the lead concentration in edible crops grown on receptor land, and the contribution of crop ingestion to facility-attributable ΔBLL. The v1.11 pathway is a screening-grade implementation of the food-crop foliar-uptake form with a single composite transfer fraction per crop class — calibrated against the food-crop Pb literature (Brown 2008 UK FSA; EC JRC TGD 2003; US EPA OSWER 2002) and, for cereals, empirically tuned against paired-measurement smelter sites in China (Liu 2020 Jiyuan; Liu 2022 Jiaozuo). It replaces both the v1.0 slope-factor food pathway (retired in v1.2 for poor LMIC validation) and the v1.10 6-parameter IAEA TRS-364 implementation that shipped briefly before this rewrite — the v1.10 form double-counted interception by carrying separate interception (R) and weathering/transfer (W, T) terms whose composite was already baked into the published food-crop literature values.

Scope: foliar only

Only the foliar (above-ground) deposition uptake pathway is modelled: atmospherically deposited Pb intercepted by leaf and stem surfaces, retained against weathering, washoff, and (for cereals) threshing/cleaning losses. Root uptake from contaminated soil — Pb already in the soil being taken up via the roots — is not included. The omission is deliberate: at relevant LMIC ULAB soil-Pb levels (10–10⁴ mg/kg), measured soil-to-plant transfer coefficients for Pb are very low (typically 10⁻³ to 10⁻² for leafy crops, 10⁻⁴ to 10⁻³ for cereals), and the direct foliar interception of fresh deposition dominates the predicted crop Pb at near-field receptors. Beyond ~3 km from the source the foliar contribution falls off and root uptake of accumulated historical soil-Pb may become significant — the tool's framing in that far-field regime is "screening for facility-attributable contamination from current-year deposition", not "total crop Pb measurement". A future iteration may add a root-uptake term once LMIC-specific soil-to-plant transfer factors are calibrated — see FUGITIVE_FOLLOWUPS.md #8.

Equations

For each receptor and each crop class, edible-tissue Pb concentration is:

C_foliar (mg/kg FW) = (D_ann × f_grow × R × (1 − W)) / Y_fw

where
  D_ann     annual atmospheric Pb deposition (mg/m²/yr, AERMOD output;
             total, wet + dry, where precipitation is modelled, dry otherwise)
  f_grow    fraction of the year the crop is exposed = t_grow_days / 365
  R         composite transfer fraction (food-crop literature; bundles
             interception, field weathering, rain washoff, and for cereals
             threshing/cleaning into a single calibrated coefficient)
  W         additional consumer wash removal at the kitchen step (0 by
             default — LMIC-conservative). Applied AFTER R has handled
             field weathering, so W represents only the additional removal
             from vigorous washing of the as-harvested crop.
  Y_fw      edible fresh-weight yield per unit ground area, kg/m²

The equation deliberately drops the separate interception term I that appears in radioecological formulations (IAEA TRS-364, ECOSYS-87, FDMT). The food-crop Pb literature reports R as a composite that is already calibrated against field measurements of crop concentration per unit deposition flux — multiplying by an additional I term would double-count interception and over-predict crop Pb by a factor of 2–5 against measured grain Pb at validation sites.

Combined-class intake and ΔBLL:

Intake_leafy   = C_foliar_leafy   × G_leafy   / 1000        (µg Pb/day; G in g/day)
Intake_cereal  = C_foliar_cereal  × G_cereal  / 1000
ΔBLL_foliar    = (Intake_leafy + Intake_cereal) × β_diet
β_diet         = 0.105 µg/dL per µg Pb/day (EFSA 2010 CONTAM Panel, child)

The implicit deposition-to-edible-Pb constant k = (f_grow × R) / Y_fw at central defaults is:

• Leafy: k = (45/365) × 0.10 / 3.0 = 0.00411 mg/kg FW per mg/m²/yr
• Cereal: k = (120/365) × 0.005 / 0.6 = 0.00274 mg/kg FW per mg/m²/yr

Both k values are asserted in the module's self-check (app/foliar_uptake.py::_self_check) so any drift in the defaults fails the test suite loudly. At the TA Luft annual deposition limit (36.5 mg/m²/yr), combined-class ΔBLL_foliar is 2.36 µg/dL. The 1-µg/dL foliar-only ΔBLL crossover occurs at D = 15.5 mg/m²/yr.

Defaults

Five parameters per crop (t_grow_days, R, Y_fw, W, G) plus the shared dietary BLL slope β_diet are exposed for user override on the Crops tab via an "Override defaults" checkbox. Sliders are disabled by default; ticking the checkbox enables them and sends a foliar_params block on the next /api/run. β_diet default is 0.105 µg/dL per µg/day (EFSA 2010 CONTAM Panel child); slider range 0.02–0.20 covers the adult value (~0.018) through the upper end of older child-cohort literature. Override-at-defaults reproduces no-override results bit-for-bit.

Outputs

The pathway adds four per-receptor fields to the /api/run response — foliar_pb_leafy_ug_kg, foliar_pb_cereal_ug_kg, foliar_intake_leafy_ug_day, foliar_intake_cereal_ug_day — plus the single delta_bll_foliar_ug_dl increment that is folded into delta_bll_total_including_foliar_ug_dl. Three heat map layers expose the pathway:

• Exposure → Crops (leafy) and Crops (cereal) — mg Pb / kg fresh-weight edible crop, with a dashed reference contour at 0.01 mg/kg — "threshold for 1 µg/dL diet". The 0.01 mg/kg anchor is the food concentration at which a full daily diet (~1 kg/day total child food consumption) at this concentration would produce ΔBLL ≈ 1 µg/dL via the EFSA 2010 dietary slope. The EU 2023/915 maximum levels (0.30 mg/kg leafy, 0.20 mg/kg cereal grain) remain visible as palette break boundaries.
• BLL impacts → Crops — the combined leafy + cereal contribution to child ΔBLL (delta_bll_foliar_ug_dl) rendered on the same palette as the other ΔBLL sub-layers. Off-scale validity marker on this layer always uses the including-foliar validity flag (crops only contribute to that total), regardless of the "Include crops" checkbox state.

The "Include crops in ΔBLL surface" checkbox on the BLL panel adds the crop pathway to the combined BLL surface and the ΔIQ gating — when unticked, the combined BLL surface shows air+soil only. The crop sub-layer and per-class µg/kg layers are always available regardless of the checkbox.

Caveats

• Linear BLL slope. The crop-to-BLL slope is a single linear coefficient applied to total dietary Pb intake. At low-to-moderate intakes (a few tens of µg/day) the linear approximation is defensible — the EFSA derivation is itself anchored at this regime. At very high intakes (hundreds of µg/day, which the screening pathway can produce at near-field large substandard scenarios) the linear form materially over-predicts the actual BLL impact because gastrointestinal absorption saturates and erythrocyte binding becomes non-linear. Non-linear absorption is the largest open methodological gap in the crop pathway — see FUGITIVE_FOLLOWUPS.md #10. The BLL validity threshold from section 5.4 (≥ 30 µg/dL) is the operational guardrail: receptors where the linear extrapolation pushes ΔBLL past 30 µg/dL are flagged as off-scale, and ΔIQ is capped at the validity-ceiling value (§5.4) rather than extrapolated.
• Foliar-only, no root uptake. Root uptake is omitted (see "Scope" above). For receptors where atmospheric deposition is low but historical soil-Pb is high, the crop pathway will under-predict actual crop Pb. The tool's framing is "screening for facility-attributable crop contamination from current deposition", not "total crop Pb measurement".
• Deposition basis (provisional). Where precipitation is modelled, the crop pathway uses total (wet + dry) deposition. Whether the retention factor and the paired-measurement validation data are based on bulk (wet + dry) or dry deposition has not been confirmed, so this choice is provisional. The map's 15.5 mg/m²/yr dashed contour (crop ΔBLL +1 µg/dL) is then drawn on total deposition.
• Annual deposition, not monthly. The growing-window deposition is approximated as D_annual × f_grow — i.e. the annual mean rate scaled by the fraction of the year the crop is in the field. This assumes the crop's growing season catches the same mean deposition rate as the annual average. Real LMIC contexts often have strongly seasonal deposition (monsoon vs dry season) and crops grown specifically in the dry season. A future iteration could substitute monthly-resolved deposition — see FUGITIVE_FOLLOWUPS.md #7.
• BLL slope is for children. The EFSA 2010 child dietary slope underlies the conversion. Adults absorb a smaller fraction of dietary Pb (~10% vs ~50% for children) so the same intake produces a smaller ΔBLL in adults. The map currently surfaces child ΔBLL only — see FUGITIVE_FOLLOWUPS.md #9.
• Two crop classes only. Real diets include roots and tubers (potato, cassava, yam), legumes, fruits, animal products, etc. The screening tool collapses this into "leafy" (high composite R, harvested whole) vs "cereal" (low composite R reflecting the small grain-translocation fraction). Users with site-specific diet data can map their crop mix onto these two classes by adjusting R and Y_fw on the Crops tab.
• Cereal R is empirically tuned against two paired-measurement sites only. The reduction of central cereal R from the literature consensus (0.01–0.02) to 0.005 is anchored to Liu 2020 Jiyuan TS and Liu 2022 Jiaozuo. These are heavily contaminated industrial sites in China; transportability to LMIC ULAB-specific contexts (where the dominant Pb species and surface deposition conditions may differ) has not been validated. Cereal predictions carry larger uncertainty than leafy predictions for that reason.

References

Key references for the crop pathway (full citations in Section 7):

• Brown, S.L. (2008). UK Food Standards Agency systematic review of food-crop Pb transfer fractions. Provides the leafy-vegetable composite-R consensus (0.05–0.15).
• European Commission Joint Research Centre (2003) — Technical Guidance Document on Risk Assessment. Source of the food-crop R framework for the EU industrial-emissions screening pipeline.
• US EPA OSWER (2002) — Multimedia, Multipathway, and Multireceptor Risk Assessment Protocol for Hazardous Waste Combustion Facilities. The third independent systematic review converging on the leafy R = 0.10 central value.
• Chamberlain, A.C. (1970) — Interception and retention of radioactive aerosols by vegetation. Atmos Environ 4(1): 57–78. Original empirical basis for the cereal R range of 0.01–0.05 on smooth grain surfaces.
• Liu Y. et al. (2020) — paired soil/crop measurements at Jiyuan TS lead smelter (China). D ≈ 350 mg/m²/yr; grain Pb 0.12–0.48 mg/kg. Primary cereal R tuning reference.
• Liu et al. (2022) — Jiaozuo smelter paired-measurement study. D ≈ 300 mg/m²/yr; grain Pb 0.62 mg/kg. Second cereal R validation point.
• EFSA Panel on Contaminants in the Food Chain (CONTAM) (2010) — Scientific opinion on lead in food. Source of the child dietary BLL slope (β_diet = 0.105 µg/dL per µg/day).
• Carrington, C.D. et al. (2019) — FDA Interim Reference Level (IRL) framework for child dietary Pb. Cross-reference for the BLL slope.
• European Commission Regulation (EU) 2023/915 — maximum levels for certain contaminants in food. Source of the reference contour values (200 µg/kg for cereal grain, 300 µg/kg for leafy vegetables).

5.6 IQ impacts from blood lead level

The ΔIQ map layer translates each receptor's modelled facility-attributable ΔBLL into a predicted IQ-point deficit using a published cohort dose-response. The IQ layer is built on top of the BLL pipeline — it does not require an additional AERMOD run, and switching background BLL or CI band on the BLL Impacts tab triggers an instant client-side recompute without a backend round-trip.

IQ dose-response specification

The tool uses a single, hard-coded dose-response: the Crump et al. (2013) log model on concurrent blood lead level, from Table 5 of the paper:

ΔIQ = β × ln(BPb + 1)
β   = −3.315 IQ points per ln(µg/dL + 1)
95% CI of β: (−4.546, −2.084)

This is the only dose-response the production UI exposes. There is no model selector — both because all six earlier candidate specifications (Lanphear-family and Crump linear variants) are dominated on epidemiological grounds, and because exposing the choice to non-specialist users invited misuse without improving the answer.

Why this specification?

Lanphear et al. (2005) is superseded. The seven-cohort pooled-analysis published by Lanphear and colleagues was the canonical low-dose Pb-IQ dose-response for nearly a decade, but Crump et al. (2013) is a refit of the same cohort data with corrected confounder treatment, a fixed boundary condition at BPb = 0, and full reporting of all five candidate BPb temporal measures under both functional forms. Crump 2013 should be preferred wherever the choice would otherwise have been Lanphear 2005; the underlying epidemiology is identical and the statistical treatment is improved.

Concurrent BPb is the right temporal measure for a screening tool. Crump 2013 reports five candidate BPb measures (concurrent, peak, early childhood, lifetime average, 24-month). For a steady-state screening application that maps a chronic facility-attributable ΔBLL onto IQ outcomes, "concurrent" — BPb measured at the same time as the IQ test — is the temporal measure most closely matched to the modelled exposure. Peak and lifetime-average measures answer different questions (worst-acute-exposure neurotoxicity vs. cumulative-burden trajectories) and are not the right framing for a permanent facility footprint.

The log form covers the full observed BPb range. Crump fits each of the five BPb measures under two functional forms: the log specification above and a linear specification (ΔIQ = β × BPb) that is adequate only below ~10 µg/dL. Near-source ULAB receptors routinely sit above that cutoff once facility ΔBLL is added on top of background; the log form is the only specification that stays valid across the full observed range and that satisfies the physically required ΔIQ = 0 at BPb = 0 boundary condition.

Alternative dose-responses considered but not used

The backend's compute_iq_loss() function retains the full set of nine Crump 2013 alternative specifications (4 alternative BPb temporal measures × 2 functional forms, plus the linear form on concurrent BPb), pinned to Crump Table 5 in the unit-test suite. They are kept on the public API surface — but not exposed in the UI — for three reasons:

• Regression armour. Pinning every β against the published table in unit tests means accidental edits to any coefficient fail loudly. Removing the specifications would remove that guardrail.
• Reproducibility. A researcher rerunning a historical scenario, or comparing this tool's output against a published Crump-table application, can pass model=... programmatically to reproduce a prior dose-response choice.
• Future sensitivity tooling. An advanced internal analysis mode (not user-facing) may later sweep the alternatives to bound dose-response uncertainty. The coefficients are zero-cost to keep available.

The Lanphear et al. (2005) coefficients themselves are not implemented. Anyone needing a Lanphear comparison should use Crump 2013 with the same cohorts and refer to the Lanphear paper directly for the unadjusted figures.

The 95% confidence interval band represents the statistical uncertainty in the cohort meta-analysis β coefficient — it does not bracket dispersion uncertainty, PSD uncertainty, IEUBK biokinetic uncertainty, individual susceptibility variation, or LMIC transportability. Total predictive uncertainty including those terms is substantially wider than the displayed CI band. The CI band sub-toggle on the BLL Impacts tab lets users render the lower (more loss) and upper (less loss) bounds separately for sensitivity assessment.

Why ln(BPb + 1) and not ln(BPb)

The +1 inside the logarithm is load-bearing. ln(BPb) diverges to −∞ as BPb → 0, which would predict infinite IQ loss at the boundary of zero exposure — a nonsense result. ln(BPb + 1) gives ΔIQ = 0 at BPb = 0, satisfying the physically required boundary condition. Crump 2013 fits the data with this form precisely for this reason; the tool's implementation enforces it in both the Python backend and the JavaScript client mirror.

Facility-attributable computation

Crump's β was estimated against total blood lead in the cohort studies. A child living near a ULAB facility carries a non-zero background BPb from food, soil, paint, water, and other non-facility sources — a substantial fraction of the global child population has baseline BPb in the 3–7 µg/dL range (IHME, GBD 2023). The tool models the IQ deficit attributable to the facility's contribution on top of that background as the marginal effect on Crump's dose-response:

ΔIQ_attributable = f(BPb_background + ΔBLL_facility) - f(BPb_background)

For the log specification this expands to:

ΔIQ_attributable = β × [ ln(bg + ΔBLL + 1) - ln(bg + 1) ]
                  = β × ln( (bg + ΔBLL + 1) / (bg + 1) )

For the linear specification the background term cancels (β × (bg + ΔBLL) − β × bg = β × ΔBLL), so the linear-form prediction is independent of background.

The default BPb_background is 3.0 µg/dL — a defensible central estimate for global LMIC childhood populations per IHME GBD 2023. The BLL Impacts tab exposes a slider so users can raise the background to ~5–7 µg/dL for higher-burden settings (sub-Saharan Africa, South Asia industrial regions) or drop it to 0 to reproduce Crump's raw dose-response. Crump's Table 5 values are reproduced exactly when bg = 0.

Per receptor the tool stores:

delta_iq_central(receptor) = β × [ln(bg + ΔBLL_total + 1) - ln(bg + 1)]
delta_iq_lower(receptor)   = β_ci_lower × [...]
delta_iq_upper(receptor)   = β_ci_upper × [...]

where ΔBLL_total(receptor) is the facility-attributable BLL increment from Section 5.4. Crump returns negative values for IQ loss; the heat map displays the magnitude as positive "IQ points lost" (<0.5 trace, 0.5–1 minimal, 1–3 low, 3–5 moderate, 5–10 high, >10 severe). Hover readouts show the central estimate with the CI range inline, rounded to one decimal place — Crump's ±60% CI on β does not support tighter display precision.

Why the marginal-effect framing matters

An earlier implementation of this layer applied Crump's dose-response by treating ΔBLL_facility as if it were the child's total BPb — effectively assuming BPb_background = 0. Because the log dose-response is concave (steeper at low BPb, flatter at high BPb), plugging a 10 µg/dL facility increment in at BPb = 0 predicts a larger IQ deficit than the same increment added on top of a 3 µg/dL background. For a typical informal LMIC scenario, the corrected marginal-effect math gives roughly half the IQ loss of the bare-substitution math at near-source receptors. The marginal formulation is the correct attribution framing because Crump's β is itself a curve, not a per-µg/dL constant.

Caveats

• Concurrent BPb is a temporal measure, not a spatial one. Crump's "concurrent" refers to the BPb measured at the same time as the IQ test. Applying this dose-response to a spatial receptor field implicitly assumes steady-state exposure: interpret ΔIQ at a receptor as the IQ deficit a child would experience if their chronic BPb equalled bg + ΔBLL_total(receptor). Alternative temporal measures (peak, lifetime average) live in the backend's coefficient table for sensitivity analysis but are not exposed in the production UI.
• Below total BPb = 5 µg/dL the prediction is extrapolative. Crump 2013 notes that the dose-response's statistical significance becomes marginal at low BPb; the underlying cohort data are sparse below ~5 µg/dL. Predictions at low-BPb receptors (often the far-field) carry larger relative uncertainty than the central estimate suggests. With the default background of 3 µg/dL, low-ΔBLL receptors still sit within the cohort's calibration range as long as ΔBLL pushes total BPb above ~5.
• Source cohorts are US, European, Mexican, and Australian children. Applying the dose-response to LMIC contexts (West Africa, South Asia) assumes the biological dose-response between BPb and IQ is transportable across populations. This is a defensible default — the biokinetic and neurological mechanisms are universal — but it has not been empirically validated in LMIC contexts. Population differences in nutritional status, age distribution, and concomitant exposures may shift the dose-response in ways the CI band does not capture.
• Background BLL is a single scalar. Real backgrounds vary spatially, demographically, and temporally — children in a peri-urban subsistence settlement may have very different baseline BPb from children in a regional capital. The current implementation uses one user-supplied scalar across the entire receptor grid. A future revision could plumb a per-receptor background field from a population-BPb raster.
• No population weighting. The map shows ΔIQ per receptor, not summed over the people who live there. Two facilities with the same per-receptor ΔIQ field can produce very different total community IQ losses depending on local population density. Population-weighted summary statistics are planned but not yet wired.
• Concavity at high ΔBLL. The log dose-response is concave: in the upper part of its validity range, doubling exposure adds progressively fewer IQ points. The heat map has limited resolving power for distinguishing severities at the upper end of the valid BLL range, even before the validity threshold is crossed.
• ΔIQ suppressed past biokinetic model validity. Crump 2013 is not evaluated on receptors with ΔBLL_total ≥ 30 µg/dL — these are in state 3 of the four-state classification (section 5.4). The receptor's delta_iq_* fields are returned as null and the IQ map renders them with an off-scale hatched marker rather than a numeric value. This is intentional: the dose-response cannot honestly be applied to a BLL the underlying biokinetic model cannot predict.

Reference

Crump KS, Van Landingham C, Bowers TS, Cahoy D, Chandalia JK (2013). A statistical reevaluation of the data used in the Lanphear et al. (2005) pooled-analysis that related low levels of blood lead to intellectual deficits in children. Critical Reviews in Toxicology 43(9): 785–799. DOI: 10.3109/10408444.2013.832726.

Background BLL default: Institute for Health Metrics and Evaluation (2024). Global Burden of Disease Study 2023 — Lead Exposure Estimates. https://ghdx.healthdata.org/record/ihme-data/gbd-2023-lead-exposure-estimates-1990-2023.

5.7 LMIC validation gap

The IEUBK anchor values used in this tool are derived from US and European child populations. Validation in LMIC populations near ULAB recycling facilities is limited. Several factors may shift the underlying biokinetic relationships in these contexts:

• Differences in baseline nutritional status (calcium, iron, vitamin D status affect Pb absorption).
• Concomitant exposures (paint, cosmetics, foodware) that may shift the BLL baseline and saturation regime.
• Bioavailability of speciated lead in ULAB-affected soils, which has not been characterised in detail.

Pure Earth's research agenda includes population BLL surveillance in ULAB-affected communities to refine these anchor values. Until that work is published, BLL predictions in this tool should be interpreted as screening estimates indicating the relative magnitude of facility-attributable exposure, not as quantitative individual risk assessments.

5.8 References

Key methodology references for this section (full citations in Section 7):

• US EPA IEUBK Guidance Manual (1994) — biokinetic model used to derive both pathway anchor tables.
• Richmond-Bryant et al. (2014) — re-evaluation of EPA IEUBK uptake parameters.
• Lanphear et al. (2005) — pooled analysis of low-level lead exposure cohort studies.
• Crump et al. (2013) — Lanphear reanalysis with refined confounder treatment; supplies the ΔIQ dose-response used in Section 5.5. Crit Rev Toxicol 43(9): 785–799.
• CDC Blood Lead Reference Value Update (2021) — current 3.5 µg/dL reference value.

6. Scope: what ULAB Plume predicts, and what it does not

This section is the consolidated statement of ULAB Plume's scope. Every individual prediction the tool surfaces falls within the "Predicts" column; questions in the "Does not predict" column require methodologies or data the tool does not currently incorporate.

Additional limitations of the predicted quantities:

• Meteorology: For the built-in locations, hourly MERRA-2 reanalysis (section 2.2), with these limitations: (1) the surface and profile files are written directly, not through AERMET, and the benchmark against AERMET is pending; (2) the 0.05 m/s friction-velocity floor is a convention; (3) one reanalysis grid cell (0.5° × 0.625°, about 55 × 60–70 km at these latitudes) represents the site's meteorology, so local terrain and land-use effects within the cell are not resolved, and surface roughness, Bowen ratio and albedo are fixed per location; (4) the precipitation amounts (CHIRPS rescaling) are provisional pending a check against national station records, and the rain-hour frequency is corrected only by a threshold rule; (5) one composite typical year is run rather than several years, and the Abuja composite biases near-stack concentrations from elevated sources low; (6) one representative point per location, so a site elsewhere in the region inherits that cell's meteorology. A user-built wind rose uses synthetic meteorology with dry deposition only. Exception: multi-site runs (API only; the interface does not offer them) use the v1 wind-rose meteorology for the built-in locations (their 2022 wind roses), with dry deposition only.
• Validation: Determinism, mass-balance, and integration tests cover the dispersion-to-receptor pipeline (see tests/test_parser_integration.py and the cross-comparison documented in section 2.5). External validation against ambient measurement campaigns at documented smelter sites is a planned future addition.
• Spatial resolution: 50 m minimum receptor distance; sub-50 m exposures cannot be characterised. Preview-mode's coarse near-source spacing under-resolves coarse-bin deposition (section 2.4).
• Source resolution: Single-stack + co-located fugitive AREA source only. Multi-stack facilities require running each stack as a separate scenario and superposing.
• Terrain and surface: Flat terrain assumed (no AERMAP integration); open-country surface roughness (no urban-canyon effects). Both are reasonable defaults for typical LMIC peri-urban / industrial siting but should be flagged in dense urban or hilly contexts.
• BLL pathways: Inhalation + soil/dust ingestion + foliar-crop deposition. The crop pathway is opt-in on the BLL surface via the "Include crops" checkbox; the response always carries both excluding-crop and including-crop totals plus their validity flags. Drinking water, lead paint, occupational exposures, and dietary uptake from non-aerial sources (including root uptake of historical soil Pb) are excluded — these are aggregated implicitly into the user-supplied background BLL.
• Demographic variation: A single population-average dose-response (IEUBK + Crump 2013) is used. Susceptibility variation by nutritional status, age, and concomitant exposures is not modelled at the receptor level.
• Food-chain pathways: A foliar-deposition crop pathway (leafy + cereal classes) is modelled at screening precision and surfaced as two heat map layers plus an optional contribution to ΔBLL — see section 5.5. Root uptake (soil-to-plant transfer of historical soil Pb via the roots) is not modelled. The crop-to-BLL conversion is a single linear slope (EFSA 2010), which over-predicts ΔBLL at very high dietary intakes — the validity guardrail is the 30 µg/dL BLL ceiling from section 5.4. Dietary uptake from non-aerial sources (animal products, fish, water) is implicitly aggregated into the user-supplied background BLL.

Foundational-model validity. ULAB Plume composes three published models: AERMOD (dispersion), IEUBK (biokinetic), and Crump et al. 2013 (ΔBLL → ΔIQ dose-response). The validity ranges and limitations within each model's predicted range are accepted as foundational and are not separately enumerated here. Users assessing whether a specific ULAB Plume output should be cited as authoritative for a given context should consult the source models' validation documentation: US EPA AERMOD Implementation Guide (2024), US EPA IEUBK Guidance Manual (1994), and Crump et al. 2013 (Critical Reviews in Toxicology 43(9): 785–799).

7. References

• US EPA, AERMOD Implementation Guide (2024). EPA-454/B-19-035.
• US EPA, AERSCREEN User's Guide (2024). EPA-454/B-21-005.
• US EPA, Guidance Manual for the Integrated Exposure Uptake Biokinetic Model for Lead in Children (1994). EPA/540/R-93/081.
• US EPA, Integrated Exposure Uptake Biokinetic Model for Lead in Children (IEUBK) (2007).
• US EPA, Adult Lead Methodology — Recommendations of the Technical Review Workgroup for Lead (2003). OSWER 9285.7-54.
• Brunekreef, B. (1984). "The relationship between air lead and blood lead in children: a critical review." Science of the Total Environment 38: 79–123.
• Hilts, S.R. (2003). "Effect of smelter emission reductions on children's blood lead levels." Science of the Total Environment 303(1–2): 51–58.
• Richmond-Bryant, J., Meng, Q., Davis, A., et al. (2014). "Influence of exposure measurement errors on results from epidemiologic studies of different designs." Journal of Exposure Science & Environmental Epidemiology 24: 624–634.
• Lanphear, B.P., Hornung, R., Khoury, J., et al. (2005). "Low-level environmental lead exposure and children's intellectual function: an international pooled analysis." Environmental Health Perspectives 113(7): 894–899.
• WHO (2021). WHO global air quality guidelines: particulate matter (PM2.5 and PM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide. World Health Organization. https://www.who.int/publications/i/item/9789240034228
• CDC, Blood Lead Reference Value Update (2021).
• Manhart, A. & Adjei, F. (2026). Lead-Acid Battery Recycling in Selected African Settings: Status Quo and Considerations for Sector Policies. CGD Policy Paper 387, Center for Global Development. https://www.cgdev.org/publication/lead-acid-battery-recycling-selected-african-settings-statusquo-and-considerations
• Stanaway, J.D. et al., "Global, regional, and national disability-adjusted life-years for 359 diseases and injuries…" JAMA, 2026.
• Brown, S.L. (2008). UK Food Standards Agency systematic review of food-crop Pb transfer fractions; foliar pathway composite-R framework.
• European Commission Joint Research Centre (2003). Technical Guidance Document on Risk Assessment. EUR 20418 EN; Part II covers food-crop deposition transfer.
• US EPA Office of Solid Waste and Emergency Response (2002). Multimedia, Multipathway, and Multireceptor Risk Assessment (3MRA) Protocol for Hazardous Waste Combustion Facilities. Source of the leafy-crop composite R = 0.10 central anchor.
• Chamberlain, A.C. (1970). "Interception and retention of radioactive aerosols by vegetation." Atmospheric Environment 4(1): 57–78. Original empirical basis for cereal R = 0.01–0.05 on smooth grain surfaces.
• Liu Y. et al. (2020). Paired soil/crop Pb measurements at the Jiyuan TS lead smelter (Henan, China). Primary cereal R empirical tuning reference: D ≈ 350 mg/m²/yr, observed grain Pb 0.12–0.48 mg/kg.
• Liu et al. (2022). Jiaozuo smelter paired-measurement study. Second cereal R validation point: D ≈ 300 mg/m²/yr, grain Pb 0.62 mg/kg.
• EFSA Panel on Contaminants in the Food Chain (CONTAM) (2010). Scientific opinion on lead in food. EFSA Journal 8(4): 1570. Source of the child dietary BLL slope (β_diet = 0.105 µg/dL per µg/day). DOI: 10.2903/j.efsa.2010.1570
• Carrington, C.D. et al. (2019). FDA Interim Reference Level (IRL) framework for child dietary lead exposure. Cross-reference for the BLL slope applied to the 7-year, 22.6 kg default receptor.
• FAO (2022). FAOSTAT — global crop yield statistics. Source of cereal edible-grain yield defaults (wheat 3.4 t/ha global; rice 4.8 t/ha Asia, 6.5 t/ha Americas).
• European Commission (2023). Commission Regulation (EU) 2023/915 of 25 April 2023 on maximum levels for certain contaminants in food. Official Journal of the European Union L 119/103.

Source emissions and particle size (section 3). The primary sources, from the reference list of the review that assembled the defaults (Pure Earth 2026).

• Pure Earth (2026). Lead emission factors for LMIC smelters: literature review and ULAB Plume default scenarios, including particle size distributions. Unpublished internal review; available on request.
• Australian National Pollutant Inventory. Nyrstar Port Pirie (facility SA0018), lead and compounds, 2010/11 and 2018/19 reports. https://www.npi.gov.au/npidata/action/load/emission-by-individual-facility-result/criteria/state/SA/year/2019/jurisdiction-facility/SA0018
• Chongqing Ecology and Environment Bureau (2024). EIA approval 渝（市）环准〔2024〕61号, Chongqing Deneng Renewable Resources. https://sthjj.cq.gov.cn/stj_qbxt/qbxt_ffsl/202410/W020241025810026607620.pdf
• Commission for Environmental Cooperation (2013). Hazardous Trade? An Examination of US-generated Spent Lead-Acid Battery Exports and Secondary Lead Recycling in Canada, Mexico, and the United States. https://www.cec.org/files/documents/publications/11220-hazardous-trade-examination-us-generated-spent-lead-acid-battery-exports-and-en.pdf
• Csavina J, Taylor MP, Félix O, Rine KP, Sáez AE, Betterton EA (2014). Size-resolved dust and aerosol contaminants associated with copper and lead smelting emissions: implications for emissions management and human health. Science of the Total Environment 493:750–756. https://pmc.ncbi.nlm.nih.gov/articles/PMC4137906
• EMEP/EEA (2009). Air Pollutant Emission Inventory Guidebook, 2.C.5.b Lead production. https://www.eea.europa.eu/en/analysis/publications/emep-eea-emission-inventory-guidebook-2009/part-b-sectoral-guidance-chapters/2-industrial-processes/2-c-metal-industry/2-c-5-b-lead-production.pdf/@@download/file
• EMEP/EEA (2023). Air Pollutant Emission Inventory Guidebook, 2.C.5 Lead production. https://www.eea.europa.eu/en/analysis/publications/emep-eea-guidebook-2023/part-b-sectoral-guidance-chapters/2-industrial-processes-and-product-use/2-c-metal-production/2-c-5-lead-production-2023/@@download/file
• Ericson B, Landrigan P, Taylor MP, Frostad J, Caravanos J, Keith J, Fuller R (2016). The Global Burden of Lead Toxicity Attributable to Informal Used Lead-Acid Battery Sites. Annals of Global Health 82(5):686–699. doi:10.1016/j.aogh.2016.10.015. https://doi.org/10.1016/j.aogh.2016.10.015
• European Commission (2001). Reference Document on Best Available Techniques in the Non-Ferrous Metals Industries. https://ga.prtr-es.miteco.gob.es/Data/images//BREF Metalurgia no férrea II-40A118A62FFCB979.pdf
• Froines JR, Liu WCV, Hinds WC, Wegman DH (1986). Effect of aerosol size on the blood lead distribution of industrial workers. American Journal of Industrial Medicine 9:227–237. https://stacks.cdc.gov/view/cdc/224273/cdc_224273_DS1.pdf
• Guberman DE (2016). Lead. In: 2012 Minerals Yearbook. US Geological Survey. Table 1, “Secondary lead, lead content”: 1,140,000 t in 2010. https://d9-wret.s3.us-west-2.amazonaws.com/assets/palladium/production/mineral-pubs/lead/myb1-2012-lead.pdf
• Henan Province (2025). DB41/2806-2025 emission standard for the lead and zinc industry. https://xcoss.henan.gov.cn/typtfile/20250616/c9547ff0da25459eb71d6520c58a229d.pdf
• International Lead Association, Wilson B. ULAB recycling slides (UNEP repository, 2019). https://wedocs.unep.org/bitstreams/2119ef89-3dbe-452f-b8a7-c70ae23e0305/download
• IUPAC (2007). Aerodynamic diameter, from Glossary of terms used in toxicology, 2nd edition. Pure and Applied Chemistry 79:1153. https://goldbook.iupac.org/terms/view/15449
• Kurtz O, Meyer J, Kasper G (2017). The contribution of small leaks in a baghouse filter to dust emission in the PM2.5 range: a system approach. Particuology 30:40–52. https://www.mvm.kit.edu/english/5160_5188.php
• Lei Y, Zhang Q, He KB, Streets DG (2011). Primary anthropogenic aerosol emission trends for China, 1990–2005. Atmospheric Chemistry and Physics 11:931–954. https://acp.copernicus.org/articles/11/931/2011/acp-11-931-2011.pdf
• Ministry of Ecology and Environment of China (2021). Coefficient manual 3212, lead and zinc smelting and manual 42, waste resource utilisation; basis of the manuals (FAQ). manual 3212 · manual 42 · FAQ
• Ministry of Environmental Protection of China (2015). GB 31574-2015, emission standards for the secondary copper, aluminium, lead and zinc industry. https://epb.zibo.gov.cn/attach/0/dd6af6b78f4641528b0f07589904032e.pdf
• Öko-Institut and partners (2025). ProBaMet summary report. https://giz.de/sites/default/files/media/pkb-document/2025-09/giz2025-en-probamet.pdf
• Royal Society of Chemistry. Periodic Table: lead. https://periodic-table.rsc.org/element/82
• Thurman J (2019). Deposition in AERMOD: overview. US EPA, 12th Conference on Air Quality Modeling. https://gaftp.epa.gov/Air/aqmg/SCRAM/conferences/2019_12th_Conference_On_Air_Quality_Modeling/Presentations/2-6_12thMC-Deposition.pdf
• UBA (2026). German Informative Inventory Report, 2.C.5 Lead production. https://iir.umweltbundesamt.de/2026/sector/ippu/metal_production/lead_production/start
• US Bureau of Mines (c.1986). Information Circular 9160: Lead Reduction in Ambient Air: Technical Feasibility and Cost Analysis at Domestic Primary Lead Smelters and Refineries. https://stacks.cdc.gov/view/cdc/206304/cdc_206304_DS1.pdf
• US Code of Federal Regulations. 40 CFR Part 51, Appendix W: Guideline on Air Quality Models. https://www.law.cornell.edu/cfr/text/40/appendix-W_to_part_51
• US EPA (1986, reformatted 1995). AP-42 §12.11, Secondary Lead Processing; background report. §12.11 · background report
• US EPA (1995). Secondary lead smelters NESHAP: background information for promulgated standards, EPA-453/R-95-008b. http://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=P1004YD7.txt
• US EPA (2006). AP-42 §13.2.2, Unpaved Roads and §13.2.4, Aggregate Handling and Storage Piles. §13.2.2 · §13.2.4
• US EPA (2011). AP-42 §13.2.1, Paved Roads. https://www.epa.gov/sites/default/files/2020-10/documents/13.2.1_paved_roads.pdf
• US EPA (2012). NESHAP for secondary lead smelting, final rule, 77 FR 556. https://www.govinfo.gov/link/fr/77/591
• US EPA (2024). User's Guide for the AMS/EPA Regulatory Model (AERMOD), EPA-454/B-24-007, November 2024 (the edition for AERMOD 24142). archived copy (web.archive.org, 18 June 2025)
• Uzu G, Sobanska S, Sarret G, Sauvain JJ, Pradere P, Dumat C (2011). Characterization of lead-recycling facility emissions at various workplaces: major insights for sanitary risks assessment. Journal of Hazardous Materials 186:1018–1027. https://www.documentation.ird.fr/hor/fdi:010053412
• Vaish AK, Singh SD, Ghosh SB, Singh O, Bhattacharyya AP, Maharaj L, Ramachandrarao P (1998). A comprehensive pollution control system for the mitigation of fine particulates of lead and SOx gases from secondary lead smelters. In Bandopadhyay A, Goswami NG, Rao PR (eds), Environmental and Waste Management, NML Jamshedpur, pp. 225–229. https://eprints.nmlindia.org/1551/1/225-229.PDF
• Wang C, Chen Y (2017). Status and development trends of China's lead and zinc metallurgy technology: energy-saving potential. Nonferrous Metals Science and Engineering 8(3):1–6. https://html.rhhz.net/YSJSYKXGC/html/2017030001.htm
• Webmineral. Anglesite mineral data. https://webmineral.com/data/Anglesite.shtml
• Wilson B, Manhart A (2021). Standard Operating Procedures for Environmentally Sound Management of Used Lead-acid Batteries. Sustainable Recycling Industries programme and Öko-Institut. https://www.oeko.de/fileadmin/oekodoc/ULAB_recycling_SOPs.pdf

Meteorology and deposition (section 2). The AERMOD User's Guide (US EPA 2024, EPA-454/B-24-007) is listed with section 3 above. The three EPA documents below are the editions for AERMOD and AERMET 24142, which SCRAM no longer serves; the links are archived copies (web.archive.org).

• Bosilovich MG, Lucchesi R, Suarez M (2016). MERRA-2: File Specification. GMAO Office Note No. 9 (Version 1.1). NASA Global Modeling and Assimilation Office. https://gmao.gsfc.nasa.gov/pubs/docs/Bosilovich785.pdf
• Diaconescu EP, Gachon P, Scinocca J, et al. (2015). Evaluation of daily precipitation statistics and monsoon onset/retreat over western Sahel in multiple data sets. Climate Dynamics 45:1325–1354. https://doi.org/10.1007/s00382-014-2383-2
• Funk C, Peterson P, Landsfeld M, et al. (2015). The climate hazards infrared precipitation with stations — a new environmental record for monitoring extremes. Scientific Data 2:150066. https://doi.org/10.1038/sdata.2015.66
• Gelaro R, McCarty W, Suárez MJ, et al. (2017). The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). Journal of Climate 30(13):5419–5454. https://doi.org/10.1175/JCLI-D-16-0758.1
• Qian W, Venkatram A (2011). Performance of steady-state dispersion models under low wind-speed conditions. Boundary-Layer Meteorology 138:475–491. https://doi.org/10.1007/s10546-010-9565-1
• Reichle R, Liu Q, Koster R, Draper C, Mahanama S, Partyka G (2017). Land Surface Precipitation in MERRA-2. Journal of Climate 30(5):1643–1664. https://doi.org/10.1175/JCLI-D-16-0570.1
• Serra YL, Rutledge SA, Chudler K, Zhang C (2023). Rainfall and Convection in ERA5 and MERRA-2 over the Northern Equatorial Western Pacific during PISTON. Journal of Climate 36:845–863. https://doi.org/10.1175/JCLI-D-22-0203.1
• US EPA (2024). AERMOD Model Formulation, EPA-454/B-24-010, November 2024. archived copy (web.archive.org, 18 June 2025)
• US EPA (2024). User's Guide for AERMOD Meteorological Preprocessor (AERMET), EPA-454/B-24-004, November 2024 (AERMET 24142). archived copy (web.archive.org, 19 July 2025)
• US EPA (2024). Guidance on the Use of the Mesoscale Model Interface Program (MMIF) for AERMOD Applications, EPA-454/B-24-005, November 2024. https://gaftp.epa.gov/aqmg/SCRAM/models/related/mmif/MMIF_Guidance.pdf

8. Version history

Calibration of default values

The tool's default emission factors, PSD distributions, and stack-geometry hints are intended to track the best published evidence available for LMIC ULAB recycling. Updates to those defaults — like the 1.1 and 1.12 recalibrations above — change the tool's starting point for screening but do not change the validation status of the underlying physics: AERMOD execution, met synthesis, parsing, and bilinear interpolation are unchanged. The pending validations documented in Section 2.5 (comparison vs ambient measurements, vs reanalysis-driven AERMOD runs, and vs alternate dispersion models) remain pending and apply equally to the new defaults.

Users supplying site-specific values via the wizard (custom lead produced, control efficiency, stack geometry) bypass the defaults entirely; the calibration above governs only the quick-scenario presets and any field the user leaves at its default.

9. Contact

For methodology questions, validation collaboration, or to report discrepancies in tool output:

Chris Kinally, Pure Earth — chrisk@pureearth.org

This document is a living methodology and will be updated as the tool evolves. The latest version is always available at https://ulabplume.org/methodology.html.