Tiered LCI estimates¶

A database "for all industrial processes" needs datasets for processes that have no flowsheet yet. cheme assess lci-estimate builds a screening LCI dataset (kind: lci_dataset) from the reaction stoichiometry, a yield and a set of sourced defaults, and labels it with its tier. The established screening approaches are the reference: Hischier et al. (2005, Int J LCA 10:59) and the ecoinvent chemicals report (Althaus et al. 2007, ecoinvent report No. 8); Parvatker and Eckelman (2019, ACS Sustain. Chem. Eng. 7:350) compare such methods.

tier	what the numbers are	how	caps
0	stoichiometric	reactions, a yield (95 % by default), losses and energy from defaults	calculation maturity `sourced`, pedigree floor `[4, 4, 3, 3, 4]` (DQR bound for floored exchanges: >= 3.33; the typical Tier 0 default scores about 4.67)
1	sourced conceptual	yield, electricity and heat (and any further exchange) from open literature, patents or BREFs, each with its source	calculation maturity `sourced`, pedigree floor `[3, 4, 3, 3, 3]` (DQR bound for floored exchanges: >= 3.0)
2+	the flowsheet	`cheme assess lca` from `process/flowsheet.yaml` and its simulation	none from the tier

A dataset without generation is a flowsheet-derived (Tier 2+) or hand-written dataset.

The estimate¶

lca/estimate.yaml (cheme schema lci_estimate) names the reactions of process/reactions.yaml and the reference product, a component of process/materials.yaml whose formulas give the molar masses:

schema: cheme/lci_estimate/v1
kind: lci_estimate
id: ethylene-oxide-t0
tier: 0
product: EO
reactions: [EPOX, BURN]          # several reactions: key and selectivity fix the extents
key: C2H4
selectivity: {EO: 0.8, CO2: 0.4}  # kmol per kmol of C2H4 converted
byproducts:                       # every product of the reactions besides the reference one
  CO2: {fate: air}
  H2O: {fate: wastewater}
location: GLO
year: 2026
# Tier 1 (or a Tier 0 that knows better): a value with its source replaces the default
# yield: {value: 0.97, source: "<publication, table>"}
# electricity: {value: 0.15, source: "..."}     # kWh per kg of product
# heat: {value: 1.1, source: "..."}             # kWh per kg of product
# losses: {fate: {C2H4: air}, reason: "a gas: vented, not drained"}
# exchanges: [{key: input:catalyst, type: input, amount: 1e-5, unit: kg, source: "..."}]
# documentation: {version: 1.0.0, review: unreviewed, rights: CC-BY-4.0, ...}

cheme assess lci-estimate                  # the dataset as YAML
cheme assess lci-estimate --write          # lca/datasets/<id>.yaml (or output:)
cheme assess lci-estimate --check          # exit 1 when the written dataset is not current
cheme assess lci-estimate --compare        # against the repository's flowsheet, per kg
cheme assess lci-estimate --json           # dataset, balance, findings, source revision

Python: chemeascode.domains.assessment.lca.estimate.lci_estimate(root, path) returns the dataset, its element balance, the findings and the source revision.

How the dataset is built¶

Per kg of the reference product (basis: 1 kg of <product>):

Reactants. The reactions' net consumption per kg of product, divided by the efficiency over the whole scheme (yield, default 0.95: "an efficiency of 95% is assumed" over the entire stoichiometric equation, Hischier et al. 2005, section 2.3). With several reactions, selectivity_extents solves the extents from the selectivities. The selectivities need a source (selectivity_source: {source: ..., source_id: ...}): Tier 1 with several reactions requires it. Without one (Tier 0) every amount that depends on them (the reactants, the by-products) is assumed with the default poor pedigree, not calculated.
By-products. Every product of the scheme other than the reference product needs a fate: coproduct (a product exchange, with properties for allocation), air or water (an emission with its elementary flow identity from the packaged crosswalk), wastewater or waste (a technosphere output to treatment). No default fate: the estimate names them.
Unconverted input. For each reactant, an air share of 0.2 % of its input mass (Hischier et al. 2005) and the rest to water. By default the rest goes to wastewater treatment (an output), since the library's cut-off system model keeps treatment without recovery in the producer's system; losses.water: water makes it a direct emission, as Hischier et al. calculate it. Solid waste: 0 (omitted, as there). losses.fate sends a component's whole loss elsewhere (a gas that is purged), with its reason. Where the default split routes a component to wastewater without an explicit fate (losses.fate, or losses.water set), the estimate warns with cheme.lci-loss-split: for a component that is a gas at ambient conditions (normal_boiling_point below 298.15 K in process/materials.yaml) and for one whose boiling point is not given; a known liquid or solid is not flagged. The hint names losses.fate.
Energy. Electricity and heat per kg: the estimate's sourced values (not negative), else the site averages of the defaults. The heat is a heat flow in kWh (carrier heat), the heat that fossil fuel generates for preheating and distillation (Hischier et al. 2005, p. 61), not a fuel input; the exchange says so in its title and carrier source.
Documentation. A uuid stable across versions (uuid5 of estimate/<id>), system model cut-off, attributional, the system boundary (infrastructure, transport and catalysts excluded unless given), the time period and valid_until from the year (reference year + 5; a Tier 0 estimate on the energy defaults says in its description that these date from 2000, beyond the 10-year limit for secondary data, and that valid_until does not endorse them); the estimate's documentation: overrides and completes it.
Data quality. Each default exchange takes the pedigree its default gives; the others the rule cheme default pedigree v1; none is better than the tier's floor (see below).
Element balance. Mandatory. The dataset records the formula of each material exchange's component (generation.formulas) and the balance of every element (generation.balance). A dataset whose elements do not close is not written (cheme.lci-unbalanced): an unbalanced reaction or a by-product without its formula is caught here. A mass exchange of a Tier 1 estimate without a component (a catalyst) stays outside the balance (generation.balance.outside), so a closed balance says nothing about it: the warning cheme.lci-mass-outside-balance names it.

generation records it:

generation:
  tier: 0
  method: stoichiometric screening (Hischier et al. 2005; Althaus et al. 2007)
  defaults: cheme tier-defaults 1.0.0
  defaults_used: [yield, loss-air, loss-water, loss-waste, electricity, heat]
  spec: lca/estimate.yaml
  maturity_cap: sourced
  pedigree_floor: [4, 4, 3, 3, 4]
  dqr_floor: 3.3333
  components: {feed:N2: N2, ...}
  formulas: {feed:N2: N2, ...}
  balance: {closed: true, tolerance: 1.0e-06, elements: {H: [0.1869, 0.1869], N: [...]}}

Every exchange that takes a default is provenance: assumed, origin design-rule, and its source cites the rule (cheme tier-defaults 1.0.0: <name>, with [ASSUMED] where the number is not confirmed) and the publication. A reactant amount computed with the default yield counts as a default exchange.

The defaults¶

src/chemeascode/domains/assessment/lca/data/tier-defaults.yaml, versioned (version): a change of a value bumps it, and every dataset records the version it was built with. Each value states its source and status: confirmed when the number was read in the cited publication (checked says where and when), ASSUMED when it is the approach's commonly quoted number, attributed to the source but not read in it (no open copy prints it).

default	value	status	source
`yield`	0.95 (efficiency over the scheme)	confirmed	Hischier et al. 2005, section 2.3
`loss-air`	0.002 of the input mass	confirmed	Hischier et al. 2005, section 2.3
`loss-water`	the rest of the unconverted input	confirmed	Hischier et al. 2005, section 2.3 (routed to wastewater treatment by default)
`loss-waste`	0	confirmed	Hischier et al. 2005, section 2.3 (solid wastes omitted)
`electricity`	0.3333 kWh (1.2 MJ) per kg	ASSUMED	attributed to the Gendorf (2000) site average that the ecoinvent v2-era proxies used; not read in Hischier et al. 2005 or Althaus et al. 2007
`heat`	0.5556 kWh (2.0 MJ) per kg, from fossil fuel	ASSUMED	as electricity
`infrastructure`	4e-10 units of a generic organics plant per kg	confirmed, not applied	Hischier et al. 2005, section 2.3 (p. 61)
`transport-lorry`, `transport-train`	100 km lorry, 600 km train per input, within Europe	confirmed, not applied	Hischier et al. 2005, section 2.1 (p. 60), after Frischknecht et al. 2004 (HCl: 200 km by train only)

The open version of Hischier et al. (2005, p. 61) confirms that the energy and water defaults are averages of the Gendorf site's environmental report (Werk Gendorf 2000) but does not print the numbers, and no open copy of ecoinvent report No. 8 was found: the 1.2 MJ electricity and 2 MJ heat per kg are ASSUMED until someone checks them against the report. They are the 2000 site averages that the ecoinvent v2-era chemicals proxies used; newer ecoinvent proxies rest on later Gendorf reports, so these values may be low for a present-day process, and the temporal pedigree stays poor. A site average says little about one process: a Tier 1 value from a source is always better. The yield, air share, water difference, omitted solid waste and infrastructure are on p. 61; the 95 % efficiency is also applied in Althaus et al. 2007 (ecoinvent report No. 8), which was not read directly.

What is not estimated. The approach also gives standard transport distances (p. 60: 100 km lorry and 600 km train within Europe for basic chemicals, after Frischknecht et al. 2004) and a water consumption from the Gendorf report. The distances are confirmed and recorded in the defaults, but not applied: the generator builds no transport exchanges, and the system boundary states transport as excluded. Plant infrastructure is excluded likewise (the library includes it only when a design provides it). Cooling water and process water are not estimated, which biases the inventory low. As in ecoinvent (stated on p. 61 for the APME data as the overall methodology), air, N2 and O2 are not inventoried as resources: the generator does not add any such resource exchange; a reactant among them is an input feed:<id> because the element balance needs it, and its unconverted share is an emission to air or a loss as any other component.

The pedigrees in the file are cheme's convention (ASSUMED): defaults score reliability 4 (a qualified estimate), completeness 5 (representativeness unknown), temporal 5 (data of 2000 or a rule from 2005), geographical 4 or 5 and technological 4 or 5 (a site average or a generic rule, not the technology).

The caps¶

Calculation and design maturity. generation.maturity_cap (calculation maturity) is sourced for Tiers 0 and 1: the level balanced needs a balanced conceptual PFD, which is the flowsheet path. The schema refuses a Tier 0 or 1 dataset claiming more, and a library manifest (process.yaml) whose maturity (design maturity) is above concept next to a Tier 0 or 1 dataset fails cheme.lci-tier-maturity, both in cheme assess lci-estimate and when a study reads the dataset. This applies to a repository without process/flowsheet.yaml, whose manifest describes the process by its datasets. With a flowsheet the manifest's design maturity rests on the design (cheme maturity), and a screening dataset that the study reads as a background does not cap it.
Data quality. pedigree_floor, the best scores any exchange may have: Tier 0 [4, 4, 3, 3, 4], Tier 1 [3, 4, 3, 3, 3] (cheme's convention, ASSUMED). The default pedigree rule applies it to a generated dataset's exchanges, so the TfS DQR is never better than dqr_floor, the mean of the floor's temporal, geographical and technological scores. A pedigree set by hand below the floor fails cheme.lci-tier-quality.

tier_findings checks a generated dataset whenever a study reads it as a background: the balance (cheme.lci-unbalanced), a default exchange not assumed or not listed (cheme.lci-default-not-assumed, cheme.lci-default-unlisted), the floor and the design maturity, plus, for a Tier 0 or 1 dataset, an info cheme.lci-tier that the study rests on a screening estimate. lci_estimate itself adds the warning cheme.lci-loss-split (see step 3).

Validation against a full design¶

cheme assess lci-estimate --compare (compare_with_design) compares an estimate with the repository's flowsheet, per kg of the product: each reactant fed, the unconverted input that leaves (summed over its fates against the design's other product terminals), and electricity and heat where the design has energy terminals. Only input terminals naming electricity or power, or heat, steam, fired heat or fuel, count as a supply; cooling and heat rejection, exported energy and terminals of another name are listed as not compared (design_energy, tested on a synthetic set of terminals).

core has one full design that makes a pure product by reaction, examples/process-ammonia (the synthesis loop; the other reacting example, process-shift, sends a mixed gas onward). A Tier 0 estimate of NH3_SYN, with the purged gases sent to air (tests/test_lci_estimate.py):

quantity per kg NH3	Tier 0	design	error
N2 fed	0.8657 kg	0.9130 kg	-5.2 %
H2 fed	0.1869 kg	0.1971 kg	-5.2 %
N2 lost (purge)	0.0433 kg	0.0906 kg	-52 %
H2 lost (purge)	0.0093 kg	0.0196 kg	-52 %

The default efficiency (95 %) is close to the loop's (90 %): the reactants are 5 % low and the losses half the design's. The design has no energy terminals, so the energy defaults are not compared (the energy defaults are not validated against any design), and argon (an inert of the makeup gas) is not estimated. No impact category is compared yet: the examples carry no background, and the ranges per impact need the library's designs (ammonia, methanol) with a reference background.

Where it sits¶

domains/process/screening_balance.py: the stoichiometric balance per kg of product and the element masses (a balance belongs to process).
domains/assessment/lca/balance.py: the element balance of a dataset, the minimal check behind an interface (element_balance(dataset) -> ElementBalance) that the balance gates replace.
domains/assessment/lca/estimate.py: the estimate, the defaults, the tier checks and the comparison; cheme assess lci-estimate is its CLI.