Upstream Brightway LCA

This repository demonstrates two ways to run the same multi-method life-cycle assessment in Brightway 2.5. The baseline implementation is intentionally repetitive; the optimized implementation avoids repeating activity-search and inventory work that is unchanged between impact methods.

The example creates or reuses a foreground activity named Manual pre-clean, connects it to 1 kg of ecoinvent soap production for the RoW location, and calculates 25 EF v3.1 impact methods, excluding the no LT variants.

Repository structure

upstream-bw-lca/
|-- main.py
|-- requirements.txt
|-- env_bw25.yml
|-- Init/
|   `-- db.py
|-- Searcher/
|   |-- find_flow.py
|   `-- methods.py
|-- Comparison/
|   |-- lca_model.py
|   |-- run_lca_baseline.py
|   `-- run_lca_optimized.py
`-- Outputs/
    |-- methods_list.txt
    |-- results_baseline.csv
    `-- results_optimized.csv

The root main.py is the primary entry point: it runs the baseline and optimized implementations, verifies that their LCIA scores match, and compares their runtimes. Database setup, discovery utilities, implementation details, and generated artifacts are kept in separate directories.

Baseline and optimized implementations

Both programs use the same project, database, foreground activity, input, and impact methods from Comparison/lca_model.py. They differ only in how they find the input activity and perform the calculations:

Phase	Comparison/run_lca_baseline.py	Comparison/run_lca_optimized.py
Activity lookup	Iterates through every activity in the ecoinvent database	Uses Brightway's local search index, followed by exact filtering
Inventory calculation	Creates a new LCA object and runs LCI for every impact method	Creates one LCA object and runs LCI once
Impact calculation	Runs LCIA after every repeated LCI	Switches characterization method and reuses the same inventory
Output	Outputs/results_baseline.csv	Outputs/results_optimized.csv

The life-cycle inventory is the calculated list of elementary flows caused by the functional unit. It does not change when only the impact method changes. The optimized implementation therefore calculates this inventory once and applies each characterization method to it.

Conceptually, the baseline does:

for each impact method:
    create LCA
    calculate LCI
    calculate LCIA

The optimized implementation does:

create LCA
calculate LCI once

for each impact method:
    switch characterization method
    calculate LCIA

Measured runtime comparison

Run the reproducible comparison without overwriting either results CSV:

conda activate env_bw25
python main.py --repeats 3

The comparison alternates execution order between rounds, reports median timings, and fails if the implementations calculate different method sets or scores.

The benchmark was run in the following environment:

Component	Tested configuration
CPU	12th Gen Intel Core i7-12700H, 14 physical cores and 20 logical processors
RAM	31.68 GiB installed (nominally 32 GB)
Storage	WD PC SN810 SDCPNRY-1T00-1006, 954 GiB NVMe
Operating system	Microsoft Windows 11 Home, 64-bit, version 10.0.26200 (build 26200)
Python	3.11.14 in Conda environment env_bw25
Brightway	bw2data 4.0.dev56, bw2calc 2.0.dev22, bw2io 0.9.dev38
Numerical stack	numpy 1.26.4, scipy 1.17.1, pandas 3.0.1
Brightway project	e10bio
Brightway database	ecoinvent-3.10-cutoff
Compared workload	25 EF v3.1 methods, excluding no LT variants

On this configuration, the median of three alternating-order rounds was:

Phase	Baseline	Optimized	Speedup
Activity search	1.503 s	0.032 s	46.42x
LCI and LCIA	8.759 s	0.364 s	24.09x
End to end	10.382 s	0.408 s	25.44x

Both implementations returned identical scores for all 25 methods. These figures describe this machine and warm local Brightway project; hardware, filesystem cache, database size, and installed package versions can change the absolute timings.

Requirements

• Conda or Miniconda
• A licensed copy of the ecoinvent 3.10 ecoSpold2 datasets
• Access to the required Brightway biosphere project

The ecoinvent database is licensed data and is therefore not included in this repository.

Recreate the Conda environment

The recommended installation method is the supplied env_bw25.yml, because the project was developed with specific Brightway development versions:

conda env create -f env_bw25.yml
conda activate env_bw25

A pinned requirements.txt is also provided as a pip alternative:

python -m pip install -r requirements.txt

Database configuration

The two LCA implementations and runtime comparison expect:

Brightway project:  e10bio
Brightway database: ecoinvent-3.10-cutoff

Init/db.py is a separate import helper configured for the APOS system model and project e10bio-apos. Ensure its project and system model match the calculations you intend to run. Then provide the location of your licensed ecoSpold2 datasets through an environment variable and run the initializer from the repository root:

$env:ECOINVENT_DATASETS_DIR = "C:\path\to\ecoinvent\datasets"
python -m Init.db

The local path remains outside the source code and is not committed. Database import should normally be performed only once.

Search activities and LCIA methods

Search the configured Brightway database by activity name, with optional unit, location, and partial-name filters:

python -m Searcher.find_flow --name "soap production" --location RoW --unit kilogram

Export every available LCIA method identifier for the project configured in the script:

python -m Searcher.methods

The method list is written to Outputs/methods_list.txt.

Run the calculations

Run the comparison from the repository root:

python main.py --repeats 3

The comparison does not overwrite the saved CSV files. It alternates execution order, checks that both implementations return equivalent scores, and reports median runtimes and speedups.

Run only the baseline implementation:

python -m Comparison.run_lca_baseline

This writes Outputs/results_baseline.csv and prints the full-scan, calculation, and end-to-end timings.

Run only the optimized implementation:

python -m Comparison.run_lca_optimized

This writes Outputs/results_optimized.csv and prints the indexed-search, calculation, and end-to-end timings. Use this implementation for normal calculations; the baseline is retained to make the optimization measurable and educational.

Supporting files

main.py

Runs both implementations without writing result files, verifies that their scores are identical, and reports median runtime and speedup.

Comparison/lca_model.py

Defines the shared Brightway project, source and foreground databases, soap input, foreground exchange, EF v3.1 method selection, and result format used by both comparison implementations.

Init/db.py

Installs the configured biosphere project and imports an ecoinvent ecoSpold2 database into Brightway. It should normally be run only during initial setup.

Searcher/methods.py

Writes every available LCIA method identifier to Outputs/methods_list.txt and prints method counts for the project configured inside the script.

Searcher/find_flow.py

Searches the configured Brightway database by activity name, with optional unit, location, and partial-name filters.

Generated files

• Outputs/results_baseline.csv: results from the repeated-LCI baseline.
• Outputs/results_optimized.csv: results from the one-LCI optimized implementation.
• Outputs/methods_list.txt: LCIA method identifiers generated by Searcher/methods.py.

Generated results depend on the selected Brightway project, ecoinvent version and system model, and installed LCIA methods. Do not commit licensed ecoinvent datasets or local Brightway project data.