Four steps to choose the right LCA database

Most data errors happen before you open the database

Building an LCA model is a layered problem. You select software, collect primary data from your own supply chain, and then back it up with secondary data from a database. Within that database, you choose a methodological version, and within that version you can select between market and transformation datasets. Each layer narrows your options, and each one can go wrong quietly.

The instinct for most new practitioners is to open SimaPro, search for the material, and take whatever looks closest. Laura Schumacher’s first piece of advice is to resist that.

“Never go straight into the databases. It’s very easy to drown in all of the possibilities, in all of the processes that you can select.” 

Laura Schumacher, LCA expert at PRé Sustainability

The alternative costs ten minutes and saves rework later: before opening anything, write down what you are modeling, the level of detail you need, the geography you are modeling for, whether you need unit or system processes, and whether you know the specific production route or will use a market average. These questions belong in the goal and scope definition stage of the ISO 14044 framework, which is where your data requirements are set.

1 – Match the library type to the data you actually have

SimaPro holds two kinds of data libraries, and the right one depends on the units your primary data uses.

1. Process data libraries
   • What they are: they collect data at the level of a single activity, expressed in physical quantities, like one kilogram of cotton yarn, or one kilowatt-hour of medium-voltage electricity.
   • When to use: if your foreground data is physical. They are usually more granular.
2. Input-output data libraries: 
   • What they are: these map economic output and environmental impact per sector across a whole economy, including financial flows. The unit is monetary: one euro of paper, for example. 
   • When to use: when your data arrives as spend rather than mass, for example euros spent on materials or invested in a sector. EXIOBASE, the EU and Denmark input-output database, the Japanese IDEA inventory, and the Social Hotspots Database all sit in this category.

Process data libraries are often sectoral. Agri-footprint and AGRIBALYSE serve agriculture and food. Carbon Minds and Industry Data 2.0 cover chemicals and industrial materials. Checking whether a database genuinely covers your sector before you commit to it is one of the cheapest error checks available.

2 – Choose the version that matches your methodology

Once you decide to go with process data, ecoinvent is the most widely used option, and it comes in four versions. Practitioners often assume these are editions of the same thing, but they are not. Each applies a different methodological approach, and each represents a different school of thought about how to divide environmental burdens.

The difference comes down to allocation, defined in ISO 14044 as partitioning input or output flows between the product system under study and other product systems. Two situations trigger it.

1. Multifunctionality: a single operation produces more than one output: a cow yields milk, meat, and leather. Allocation decides how the inputs and outputs of that cow are split across the three.
2. End of life: a product reaches the end of its useful life and enters another product’s life cycle through recycling or energy recovery. A PET bottle returned to the supermarket gets shredded and becomes part of another bottle. Allocation decides where the first life cycle ends and the second begins, and who carries the burdens and the benefits.

ecoinvent’s four system models resolve these differently. Allocation cut-off by classification and the EN 15804 cut-off version both follow the ISO hierarchy for multifunctionality, but cut the end-of-life boundary at different points, with the EN 15804 version adjusted for Environmental Product Declaration requirements. 

Allocation at the point of substitution (APOS), avoids end-of-life allocation using system expansion. The consequential model uses substitution rather than allocation, and is the one version built on marginal rather than average data.

Which version you need depends on your goal and scope. If you are producing an EPD, you must use a cut-off allocation model. The specific EN15804 makes it easiest to produce an EPD due to the addition of specific indicators.

3 – Check the geography

Geographies in ecoinvent vary in resolution. India alone appears under several geographical scopes. Matching the geography of the dataset to the geography of the product you are modeling is a direct opportunity to catch an error early.

The most common trap is “rest of world.” It reads as a safe default, but your specific location may not be inside it for that particular process. The ecoinvent documentation lists what each geography covers, and checking takes a minute.

If you know your supply splits across three countries and you know the shares, Laura recommends building that mix yourself rather than accepting a broader average.

4 – Decide between market and transformation processes

A market process gives you an average of production across locations, with average transport already included. Soybean oil sourced from Brazil, the US, Europe, and the rest of the world arrives as one blended dataset.

A transformation process is specific to one production route in one location. Soybean oil produced in Brazil, with its actual inputs and outputs visible. But it does not include transport, and that’s where errors can creep into your model. If you select a transformation process, you have to add the transport dataset from the suppliers’ location to yours. 

If you do not know where your supplier sits, use the market process and accept the average. If you do know, use the transformation process for that geography and add the transport yourself.

Choosing between the two deliberately, rather than taking whichever appeared first in the search results, removes a whole category of silent error.

When nothing fits exactly, adapt the dataset

Sometimes the closest available dataset is close but not right. SimaPro lets you copy a background process and edit the copy, which moves you up the accuracy ladder from generic secondary data toward value chain specific data. The result is hybrid data: secondary data adapted to reflect your actual supply chain.

Laura Schumacher illustrates this with coconut oil: an available dataset covers the Philippines, but the model needed India. Going input by input, the two water flows and the electricity input were switched to Indian geographies and the Indian grid mix. The chemical factory input was already global, and transport and heat were already modeled as rest of world, so all three applied to India unchanged. Only the dehusked coconut stayed Philippine, because no Indian dataset exists.

That is a total of four edits, and the result is substantially more representative than the generic Philippine dataset. SimaPro desktop gives LCA experts and consultants full control over this kind of adaptation, while SimaPro cloud applies the same models across product and sustainability teams with reusable, standard-aligned structures.

Mixing databases is sometimes unavoidable

No database covers everything. A poultry model built in ecoinvent may need a specific feed ingredient that only Agri-footprint carries. Mixing is acceptable, but first you need to check that the methodological assumptions match.

Specifically, check how each database handles multifunctional allocation and end-of-life allocation. If the two differ, you risk double-counting an impact or missing one entirely. If they use the same approach, you can use both databases and document the choice and the reasoning.

A checklist before you trust a dataset

Three tips from Laura Schumacher for keeping a study defensible under review:

1. Use proxies honestly and document all. When no exact match exists, select something close in function, dimensions, geography, or material, then refine later. If nothing is close enough, mark it as a data gap rather than forcing a substitute.
2. Stay consistent in your sources. Prioritize commonly accepted LCA databases, report them transparently, and avoid mixing where you can. When mixing is unavoidable, document what you used and why, and account for it in your uncertainty assessment.
3. Flag data gaps clearly. Cut-offs are generally allowed below a defined threshold of total mass, energy, or impact, but the exclusion still needs documenting so a reviewer can factor it in when reading the results.

On whether a proxy is good enough, Schumacher’s answer is to decide empirically rather than in advance. Run the results with the proxy in place, then look at the hotspot analysis. “It’s always good to check first how important it is to your results,” she said. “How sensitive are my results to this specific data point? And only then try and improve on it.”

For a more formal assessment, the European Commission’s Environmental Footprint method scores data against criteria covering reliability and geographical, time-related, and technological representativeness, which combine into a single data quality rating. SimaPro also includes the pedigree matrix, which lets you record quality indicators per input and calculate a score from them.

Your results will change even when your model does not

Databases update regularly. That means a model you have not touched can return different numbers next year, which can be surprising for a client who has built a decision on last year’s figure.

The change itself is not a problem. Datasets update because they are getting closer to reality, which is what you want. What causes trouble is the unexpected change. Here five habits to manage database updates:

1. Indicate which database version you used.
2. Monitor changes on a schedule.
3. Give stakeholders advance notice of what will change.
4. Update your models on a predictable cadence.
5. Keep version control on both models and results.

When results move after an update, find the reason. ecoinvent publishes documentation on what changed in each release, which usually explains the difference and often shows it as an improvement in granularity rather than a real increase in impact.

If a dataset is flagged obsolete after an update, Schumacher recommends switching to a current one rather than holding on to the old version. Processes are frequently marked obsolete because the geography has been split into finer regions, which means the newer database can model your situation more precisely once you select the right replacement.

Bonus: a cheat sheet for handling LCA data

• Always define your goal and scope first. This sets the frame for everything that follows.
• Collect foreground data and note whether it is physical or monetary.
• Choose the library type: process data for physical quantities, input-output for monetary values.
• Choose the database that fits your sector and meets your quality standards.
• Choose the dataset: geography, unit or system process, market or transformation.
• Model, and document every choice as you go.
• Iterate. Use the hotspot analysis to find where better data would actually change the answer, and improve it.

Start with a proxy, document it, look at the first results, and target your data collection at the points where it matters. Build the maintenance habit in from day one rather than retrofitting it when a client asks why the number moved.

FAQ – Frequently asked questions

How accurate is a “rest of world” geography in LCA? 

Rest of world is a reasonable first approach when no country-specific dataset exists, but accuracy depends on whether your location is actually included in that scope for that process. Check the ecoinvent documentation to confirm coverage, and use the data quality indicators in SimaPro to record how close the match is.

Should I use a unit process or a system process? 

Both return the same results. A unit process keeps every background link visible so you can trace inputs down through the supply chain. A system process aggregates those links into a single set of inputs and outputs, which is less detailed but faster to calculate and easier to interpret. Use unit processes when you need traceability. In SimaPro, you can easily switch between the two versions using Library Switch in your calculation set-up.

Is hybrid data classified as primary or secondary data? 

Hybrid data sits between the two. It starts as a secondary dataset and is adapted in SimaPro using primary data from your own value chain, so it is more specific than generic secondary data but cannot be called primary. Report it as adapted secondary data and document which inputs you changed and why.

Why do supplier-specific EPDs often show lower emissions than generic datasets?

Suppliers usually publish an EPD because their product performs well, while generic datasets tend to be more conservative estimates. Supplier-specific data is preferable because it reflects your actual supply chain, but check the methodological choices and system boundaries behind the EPD to confirm they match your model before using it.

How do I choose an impact assessment method? 

The method follows your goal and scope. SimaPro supports around 40 life cycle impact assessment methods. If the study covers climate change only, a greenhouse gas focused method works. Make sure the method matches the data you collected, since one covering water and land use returns nothing useful if you never modeled those flows.

What should I do when SimaPro flags a dataset as obsolete? 

Switch to a current dataset rather than reverting to the older database version. Datasets are often marked obsolete because the geography has been split into more specific regions, which means you can now select a closer match manually. Document which replacement you chose and the reason for the choice.