SASEL Lab
SASEL Lab
SASEL Lab
Sustainability assessment for African food value chains, built for the FCI4Africa project and powered by Ambit, the regionalised life cycle assessment engine developed at SASEL.
Green Means Go is a sustainability assessment platform for the African agri-food value chain, built for the Food Convergence Innovation for Africa project. Farms, processors and exporters use it to measure the environmental footprint of their operations and to carry that evidence into markets that now ask for it. Underneath the application sits Ambit, a regionalised life cycle assessment engine developed at SASEL, which performs the science.
The separation matters. Green Means Go is the interface, the workflow and the African market focus. Ambit is the assessment core: the inventory database, the supply chain solver, the emission models, the characterization methods and the uncertainty analysis. The engine is written to serve any region and any product system, and Green Means Go is its first application.
The engine takes its name from the Latin ambitus, a going round, which also gives English its word for the scope and boundary of a thing. Both meanings sit at the centre of life cycle assessment: the cycle a product travels from field to gate, and the boundary an analyst draws around it before any number is computed.
An environmental footprint is the sum of everything consumed to make a product, traced back through every supplier. Ambit computes that the way life cycle assessment requires: it assembles the technosphere and biosphere matrices for the entire supply chain reachable from the product under study and solves them as a linear system. The distinction is practical. Real inventory databases contain loops, since making electricity needs steel and making steel needs electricity, so walking the chain step by step would run forever. Solving the system terminates and gives the exact answer.
Background data comes from ecoinvent 3.11, held in a canonical store that the engine queries directly. Each purchased input a farm reports is matched to a background process, and the resulting upstream inventory is merged with the emissions produced on the farm itself. The merged inventory is then characterized through a single validated path, so on-farm and upstream contributions are reported separately while resting on identical treatment.
On-farm emissions follow the IPCC 2019 refinement, with emission factors selected by the climate zone of the farm. Characterization runs on ReCiPe 2016 and EF 3.1, resolving ten canonical midpoint categories including climate change, water consumption, land use, acidification, freshwater and marine eutrophication, fossil and mineral resource depletion, particulate matter formation and photochemical oxidation. Results roll up into a normalised single score expressed in micro-person-equivalents, with each category contribution shown so a user can see what drives the number.
Ambit treats region as data. A registry entry supplies the ranked background locations the matcher should prefer, the AWARE water scarcity factor, the IPCC climate zone parameters that drive on-farm emission factors, the default characterization method and the currency. Ghana, Nigeria and Canada are entries in that table today, and a Canadian farm automatically draws Canadian grid backgrounds and the EF 3.1 method. Adding a country is a registry entry, which is what allows one engine to serve African value chains and comparative work elsewhere at the same time.
A number alone changes little. Ambit ranks abatement measures against the assessed operation, then screens each one economically using the farm quantities and a regional price feed. It estimates annual revenue, annualises the cost and saving of each measure, computes payback, assigns an affordability tier, and sequences the result into a phased plan covering what to do now, this year and later.
That layer holds a deliberate discipline. Every number in a recommendation comes from explicit arithmetic, and the language model above it is limited to explaining what the arithmetic returned. Estimates carry their assumptions and their gaps, because a screening assessment often lacks farm size, season count or available capital, and a confident wrong payback figure does real harm.
The application wraps the engine in the workflow an operator needs. Farms and processing facilities are set up once and assessed repeatedly. Assessments stream their progress as they run. Results are organised into studies that carry versions and revisions, so an assessment can be revisited and its history defended. Scenarios let an operator test a change before making it, a review workflow supports sign-off, and finished work can be shared or exported for research use.
Results are ISO 14040 and 14044 reports carrying a provenance stamp that records the engine build, the dataset editions, the characterization method and the field model used. That traceability is what turns an assessment into evidence for EU carbon border and deforestation reporting, and for market access under the African Continental Free Trade Area.
Green Means Go is available on the web and was developed at the SASEL Lab, McGill University, for the Food Convergence Innovation for Africa project.
The canonical store holds processes, flows, exchanges and characterization methods imported from openLCA-format and GLAD-published datasets. For a given product system the solver assembles the technosphere matrix and the biosphere matrix over every process reachable from the target, then solves for the scaling vector and multiplies through to obtain the elementary flow inventory. Sign conventions were validated against openLCA and are documented explicitly: the reference flow sits on the diagonal signed by direction, waste treatment activities carry their reference as an input, and biosphere amounts are stored as given so that flows encoding direction in their own identity, such as carbon dioxide taken from air, keep the sign their characterization factor expects.
Field emissions are computed by a Rust kernel covering direct and indirect nitrous oxide, methane, and fuel combustion carbon dioxide, with emission factors drawn from the IPCC 2019 refinement and selected by climate zone. The kernel returns an inventory of elementary flows which is mapped onto the canonical store nomenclature and merged with the upstream inventory before characterization, so the whole footprint passes through one characterization path.
Impact per category is the sum over elementary flows of quantity multiplied by characterization factor. A midpoint map reconciles ReCiPe and EF category names into ten canonical categories and preserves the unit each method actually reports, since the same category carries different units under different methods. The single score normalises each midpoint by its per-capita reference to obtain person-equivalents, equal-weights them, and reports the total in micro-person-equivalents per kilogram together with each category share. Qualitative bands are calibrated empirically against a benchmark basket, with a separate calibrated set for facility assessments.
Each contribution source, whether a matched purchased input or the modelled field emissions, is scored on the Weidema pedigree matrix and assigned its own geometric standard deviation. Monte Carlo sampling then draws an independent lognormal multiplier per source, with a single draw scaling that source across every impact category at once, so uncertainty shared within a source stays correlated while remaining independent between sources. Category totals are re-summed from the sampled sources. The scope is disclosed plainly: the screen covers inventory magnitude, characterization factor uncertainty stays outside the propagation, and the inventory solve is reused across iterations. Those limits appear in the basis string and in the ISO limitations section of the report.
A matcher ranks abatement measures against the assessed operation and its hotspots. The economics pass then screens feasibility with plain arithmetic over the farm quantities and a regional price book, producing annual revenue estimates, annualised measure costs and savings, payback periods, affordability tiers and a phased sequence. Assumptions and data gaps travel with every estimate.
Reports follow ISO 14040 and 14044 structure, covering goal and scope, inventory, impact assessment and interpretation, with limitations stated. A provenance stamp records the engine build, the dataset editions in use, the characterization method and the field model, so a result can be reproduced and audited later. A method lab allows one inventory to be re-characterized under different methods for comparison, and research export produces the underlying data for external analysis.
The engine is reached through a FastAPI service backed by a relational store with Alembic migrations, covering authentication, workspaces, farms, facilities, inventories, assessments, studies, scenarios, reviews, sharing and batch jobs. Long-running assessments stream progress to the client. The Green Means Go frontend is a Next.js and TypeScript application with dashboards for farms and facilities, guided farm and processing assessment flows, and results presentation.
Tech Stack
Categories