| title | tags | authors | affiliations | date | bibliography | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
`unfold`: removing the barriers to sharing and reproducing prospective life-cycle assessment databases |
|
|
|
05 January 2023 |
paper.bib |
unfold is a Python package that allows reproducing life-cycle databases which
partially build on a data source that cannot be shared. It produces
data packages that contain the differences between the databases to
share and the licensed data source. The data package also includes a
metadata file describing the databases, the author, and other helpful
information. unfold allows one to pack and unpack any databases from
a single data package to ease the sharing and reproducibility of prospective
or scenario-based life-cycle assessment databases.
Life-cycle assessment (LCA) consists of quantifying the environmental footprint of a product or a process by considering impacts along each of its relevant life-cycle phases [@ISO:2006]. It relies on databases describing thousands of industrial processes, which interdependency and interactions with the greater environment form large input-output matrices. However, these matrices -- often modified to fit the need of a study -- are difficult to share when they contain data covered by a restrictive End-user license agreement, such as that of the ecoinvent database [@Wernet:2016].
The difficulty in sharing these databases prevents the reproducibility of results and impedes the scientific validity of the work produced by the LCA community.
Using the LCA framework brightway2 [@Mutel:2017], unfold allows sharing modified
LCA databases without:
- exposing the data under license,
- going through the steps of data modification,
but sharing instead data packages used to reproduce the databases locally (provided the End-users also have access to the licensed data source).
unfold is initially conceived to share heavily modified LCA databases, such as
those generated with the premise package [@Sacchi:2022] or those regularly produced
within the field of prospective LCA (where the need to modify the source database
extensively is often required) – see the work of Cox [@Cox:2018], Mendoza [@Mendoza:2018],
Joyce [@Joyce:2022] and colleagues.
LCA relies on input-output matrices representing product and service exchanges that are typically sparse, with a density inferior to 1%. Sometimes, for the need of sensitivity or scenario analysis, up to hundreds of thousands of exchanges are added, modified, or removed from the technosphere matrix, potentially reflecting significant changes in certain industrial sectors. [@Mendoza:2018] gives an example where exchanges related to power generation are modified throughout the database to reflect changes in generation efficiencies, regional electricity mixes, etc.
unfold calculates the scaling factors for each modified exchange with
reference to its value before modification (i.e., the reference database) and
stores them into a data package.
When another user reads the data package, the scaling factors are multiplied by the exchange value found in the local reference database to back-calculate the value of the modified exchange.
The example illustrated in Figure 1 represents a use case where User 1
(in blue) would like to share databases A and B, which are initially derived
from database C – which cannot be shared. unfold creates a data package (zip file)
where the difference across exchanges in databases A and B relative to the reference
database C are stored as scaling factors. Additional metadata (e.g., author, description)
is included in the data package, per Frictionless's specifications [@Walsh:2022].
User 2 (in green) reads the data package using unfold on its local computer,
which back-calculates the value of each exchange by multiplying the original values
found in the reference database with their corresponding scaling factors.
This assumes User 2 also has local access to the reference database C.
unfold offers two functions: pack and unpack LCA databases. The packing
function calculates the differences between one or several modified databases
and a reference database and stores them in a data package ready to be shared.
The unpacking function creates and registers one or several LCA databases by
reading the data package and the local reference database.
unfold can also produce superstructure databases [@Steubing:2021], which can
integrate several scenarios into a single brightway2 database, to be further
used with the LCA software activity-browser [@Steubing:2020].
unfold allows sharing LCA databases that partially build on proprietary data
without exposing the data under license. This eases the process of
reproducibility, which is an essential prerequisite to scientific validity.
Sharing scenario-based or prospective LCA databases is already possible through
other means, such as exchanging Futura's recipes [@Joyce:2022] or
premise's input parameters [@Sacchi:2022]. However, the diversity of shared
databases is eventually limited by the range of tools that generate them -- most
of the existing tools focus on modifying energy-intensive sectors.
On the other hand, unfold allows sharing databases regardless of the tool or
method used to generate them, as long as they present differences with the reference
database. For example, a database with manually adjusted leakage rates of ozone-depleting
gases can be shared just as easily.
Doing so allows sharing of the work of the different authors of these databases to be reused within the community. To that effect, a repository that lists several data packages already exists [@Sacchi:2022b], allowing users to reproduce and use prospective LCA databases without executing the steps or operating the tools that led to their generation.
unfold offers a way to exchange and reproduce LCA databases that build partially
on licensed data by sharing scaling factors in a data package.
Financial support was provided by the Kopernikus Project Ariadne (FKZ 03SFK5A), funded by the German Federal Ministry of Education and Research