Consequential modelling

Premise can import and transform the consequential system model of ecoinvent versions 3.8 to 3.12. It uses production trajectories from an integrated assessment model (IAM) to identify the suppliers that are likely to respond to a marginal change in demand. The method is described by Maes et al. (2023).

If you use this module, please cite:

Maes, B., Sacchi, R., Steubing, B., Pizzol, M., Audenaert, A., Craeye, B., and Buyle, M. (2023). Prospective consequential life cycle assessment: Identifying the future marginal suppliers using integrated assessment models. Renewable and Sustainable Energy Reviews, 188, 113830. https://doi.org/10.1016/j.rser.2023.113830

What the consequential parameters describe

The system_args parameters do not describe when environmental impacts occur. They determine the IAM time interval Premise uses to identify the background suppliers that respond to a marginal change in demand.

year is both the scenario year of the generated database and the year in which the modelled change in demand begins. duration describes how long that change persists forward from year. range time defines an observation window for a short-lived change.

It is useful to distinguish three clocks:

Clock	Meaning
Database year	The year of the prospective background database and the initial year of the marginal demand change.
Demand-change period	The period for which the additional or reduced demand persists.
Supplier-observation interval	The IAM years over which production trends are measured to identify the marginal suppliers.

Important

duration is not automatically the lifetime of the assessed product, and it does not integrate emissions or impacts over several years. Lead time does not move the foreground activity or its emissions to another year. Premise produces a static database containing one marginal supplier mix for each transformed market.

Premise builds marginal mixes for commodities for which the selected IAM scenario provides production volumes. A technology’s contribution is derived from its production trend over the selected supplier-observation interval.

Choosing the demand-change period

First define the marginal demand change represented by the study:

For a single occurrence or a change lasting less than three years, use duration=0 and a non-zero range time. The default range time=2 measures the trend over two years before and two years after the expected installation of additional capacity.
For a change lasting three years or more, use range time=0 and set duration to the number of years for which the change persists.
Do not set both range time and duration to non-zero values. This combination is unsupported.
For a permanent or indefinite change, select and justify a finite duration that fits the goal and scope. An infinite IAM trend cannot be measured.

range time is a half-width, not a total interval length. For example, range time=4 selects four years before and four years after the centre of the interval, giving an eight-year difference between its endpoints.

How the supplier-observation interval is calculated

Let y be the database year and start of the demand change, r the range time, d the duration, L_i the lead time of technology i, and L_avg the production-weighted average lead time of suppliers in the market.

Supplier-observation intervals
Demand change	Perfect foresight	Myopic, average lead time	Myopic, individual lead times
Short: `range time=r`	`y-r` to `y+r`	`y+L_avg-r` to `y+L_avg+r`	For each technology: `y+L_i-r` to `y+L_i+r`
Long: `duration=d`	`y` to `y+d`	`y+L_avg` to `y+L_avg+d`	For each technology: `y+L_i` to `y+L_i+d`

Perfect foresight gives both lead-time modes the same observation interval: suppliers are assumed to start investing early enough for capacity to be available at y.

Setting both range time and duration to zero selects a legacy ecoinvent-style fallback in which lead time itself becomes the interval. With average lead time, this is y to y+L_avg for myopic behaviour and y-L_avg to y for perfect foresight. Individual mode uses the corresponding L_i for each technology. Prefer an explicit value for range time or duration in new studies.

The endpoints describe an elapsed interval. Annual IAM values, including both endpoints, may be used during the calculation. If a calculated endpoint falls outside the available IAM time series, Premise uses the nearest available IAM year and emits a runtime warning. This shortens or shifts the effective interval, so inspect the summary printed during database generation.

Four timelines showing the IAM supplier-observation intervals for short and long demand changes with perfect foresight and myopic behaviour. — IAM intervals used to identify marginal suppliers in average-lead-time mode. The blue bars show the interval over which supplier production trends are measured; they do not show when environmental impacts occur. In individual mode, `L_avg` is replaced by `L_i` for each technology.

Foresight and lead time

Lead time is the number of years between an investment decision and the installation of new production capacity. Premise reads a lead time L_i for each technology from leadtimes.yaml.

With myopic behaviour (foresight=False, the default), suppliers decide to invest only after the change in demand becomes observable. Their response is therefore shifted forward by lead time.

With perfect foresight (foresight=True), suppliers anticipate the change. They are assumed to begin investment early enough for the new capacity to be available at year. A short interval is therefore centred on year, and a long interval begins at year, irrespective of the lead-time mode.

The "lead time" argument selects how the lead-time data are applied; it does not turn lead time on or off:

False (default) uses one market-average lead time, L_avg, calculated from the technology lead times and their production shares.
True uses a separate observation interval based on L_i for every technology. This is more detailed and is recommended by Maes et al. for myopic modelling.

Measurement method 4 splits one common market interval into annual sections. It therefore cannot be combined with technology-specific lead times; Premise raises an error for measurement=4 together with lead time=True.

Important

"lead time": False means use the market-average lead time. It does not mean that lead time is zero or ignored.

Lead time describes the response of background suppliers. It is not the time between a foreground investment decision and use of the foreground product, and it does not change the year represented by the generated database.

Worked examples

Modelled marginal demand change	`year`	`range time`	`duration`	Change type
One additional bicycle is demanded in 2050	2050	2	0	Short
Additional bicycle production starts in 2046 and lasts four years	2046	0	4	Long
A factory starts production in 2050 and adds demand for twenty years	2050	0	20	Long

The second example starts in 2046: duration=4 looks forward from that year. Using year=2050 and range time=4 would instead describe a short change centred on 2050, which is a different question.

Note

A single set of system_args cannot schedule construction of a foreground factory in 2026 and its operation in 2050. That question requires a time-specific foreground model, separate databases for the relevant years, or a dynamic LCA method. The consequential parameters only determine how Premise selects marginal suppliers in the background markets it transforms.

Capital replacement

When capital replacement rate=True (the default), Premise accounts for the replacement of depreciated production capacity. For a supplier with production P and lifetime L, the replacement baseline is -P/L. The baseline is subtracted from the observed production slope.

For example, a technology with stable production has a raw slope of zero. Its adjusted indicator is 0 - (-P/L) = P/L. It may therefore be part of the marginal mix because investment is needed to replace retiring capacity even though its total production does not grow.

Capital replacement affects supplier eligibility and marginal shares. It is not a lead time and does not add the embodied environmental burden of replacing equipment to the foreground model.

Production trends evaluated with a horizontal baseline and with a capital replacement baseline. — Left: a horizontal baseline. Right: the capital replacement baseline is subtracted from the production trend.

Technology lifetimes are stored in lifetimes.yaml.

Measuring production trends

The measurement argument controls how production changes are quantified within the supplier-observation interval:

0 – endpoint slope. This is the default and is also used by ecoinvent. It is generally suitable for short intervals and approximately linear trends.
1 – linear regression. It reduces the influence of individual annual values compared with an endpoint slope.
2 – area under the curve. It gives more emphasis to developments early in the interval and can be useful when near-term consequences matter most.
3 – weighted slope. It adjusts the full-interval slope using a shorter slope defined by weighted slope start and weighted slope end. With the defaults, this shorter slope covers the last quarter of the interval and emphasizes developments near its end.
4 – annual measurement. It splits the interval into individual years and gives short-, medium-, and long-term developments equal importance. It requires lead time=False because all suppliers must share one interval.
5 – legacy production-volume weighting. It selects suppliers according to the direction of their production trend and weights the selected suppliers by their production volume at the start of the interval. It is retained for reproducing legacy consequential configurations; methods 0 to 4 are preferred for new IAM trend analyses.

Methods 2 to 4 are intended for non-linear production trajectories, which are more likely over long intervals. The choice of interval can have a larger effect than the choice of measurement method, so consequential studies should test plausible durations and measurement methods when these choices are material to the conclusions.

Comparison of methods for measuring non-linear production trends. — Comparison of methods 2, 3, and 4 for non-linear production trajectories.

Configuration reference

The supported system_args and their implementation defaults are:

Argument	Type	Default	Purpose
`range time`	integer years	`2`	Half-width of the interval for a short demand change.
`duration`	integer years	`0`	Persistence of a long demand change, forward from `year`.
`foresight`	boolean	`False`	Selects myopic or perfect-foresight timing.
`lead time`	boolean	`False`	Selects market-average (`False`) or technology-specific (`True`) lead-time intervals. `False` does not disable lead time.
`capital replacement rate`	boolean	`True`	Uses technology replacement requirements as the baseline.
`measurement`	integer, 0–5	`0`	Selects the production-trend measurement method.
`weighted slope start`	fraction	`0.75`	Start of the short slope used by method 3.
`weighted slope end`	fraction	`1.00`	End of the short slope used by method 3.

Pass the arguments explicitly so that the modelling choices remain visible and reproducible:

from premise import NewDatabase

system_args = {
    "range time": 2,
    "duration": 0,
    "foresight": False,
    "lead time": False,
    "capital replacement rate": True,
    "measurement": 0,
    "weighted slope start": 0.75,
    "weighted slope end": 1.00,
}

ndb = NewDatabase(
    scenarios=scenarios,
    source_db="ecoinvent 3.12 consequential",
    source_version="3.12",
    key="xxxxxxxxx",
    system_model="consequential",
    system_args=system_args,
)

ndb.update("electricity")
ndb.write_db_to_brightway()

During transformation, Premise prints a summary for each marginal market. Check Lead time and L avg to verify the selected mode and market-average lead time. Avg start and Avg end show the common interval or, in individual mode, the average interval used for market-level diagnostics. Also inspect Range, Duration, Foresight, Cap repl., and Vol ch.. This is especially important when an interval approaches the first or last IAM year.

Scope and limitations

The method assumes a small marginal change. A project large enough to alter market structure or the IAM pathway itself requires additional scenario modelling.
One system_args dictionary is applied to the transformed consequential markets. Different demand changes may justify different time intervals and, consequently, separate database builds.
IAM production values are interpolated to annual resolution. Requested interval endpoints outside the available IAM years are replaced with the nearest available year.
The generated database is static. The supplier-observation interval does not create temporally distributed exchanges or a dynamic impact assessment.
A long or permanent change still requires a finite, documented duration.
Premise rejects negative intervals, simultaneous non-zero range time and duration, and durations of one or two years. Use range time for a change lasting less than three years.

Some technologies are excluded from marginal markets because their feedstock availability constrains their response. This typically applies to waste-to-energy and waste-to-fuel technologies. Secondary steel is excluded from marginal steel mixes. The exclusions are defined in the consequential data files shipped with Premise.

Some imported inventories cannot be linked directly to the ecoinvent consequential database. blacklist.yaml provides alternative linking candidates.