Why My LCA Scores Radically Changed After The ecoinvent Update
With each upgrade of an LCI database, LCA practitioners need to be aware that their results might change. PRé’s LCA expert Tommie Ponsioen researches the extent of the changes caused by updating from ecoinvent 2.2 to 3.1.
Now that ecoinvent 3.1 is available, which includes recycled content, SimaPro users who want to work with the latest data can update the links in their life cycle assessment models from ecoinvent 2.2 to ecoinvent 3.1. As both versions of the background database are based on the same allocation methodology, the results can be compared. However, the results may have changed significantly. It may be good news if the calculated impacts are lower after the update, but it can be difficult to communicate this. People may be sceptical that your impact has actually been reduced even if the product hasn’t changed. If the calculated impacts become higher, you have even more explaining to do. This article helps you with these explanations by looking at different examples of ecoinvent products that were updated, and how that changes the results of the models they’re used in.
I used the ReCiPe single score method to compare products that were updated from version 2.2 to 3.1. This comparison did not include waste treatment and products with changed units. About 40% of the scores changed less than 10%, about 50% increased by more than 10%, and about 10% decreased by more than 10%. This graph gives a more detailed picture:
I chose twelve products to investigate further:
After comparing each product’s process contributions and networks in the old and the new version, I found four potential explanations for the changes:
1. The product’s allocation method was changed from physical to economic.
2. A process was removed from the product’s network.
3. The product’s markets of important inputs were extended to include more countries and technologies.
4. The product used land transformation, which was modelled in a completely different way.
1. Alternative Allocation
How products are allocated can make a big difference in their impact on the final results, especially if they are by-products. For example, if your model has links to ammonium chloride from animation of chlorosilane or chlorine, gaseous from lithium chloride electrolysis, your results may decrease dramatically, because the ReCiPe scores of these products are reduced to less than 1% after the update. These by-products are produced in large quantities compared to the main products, but are priced much lower than those of the main products. Therefore, their impact is high in case of physical (mass) allocation, which was used in version 2.2, and low in case of economic allocation, as used in version 3.1.
The cases of acetic acid from acetaldehyde oxidation and nitrogen fertilizer from ammonium nitrate phosphate production are less extreme but follow similar lines. Conversely, the allocation factor of the main product hydrogen, liquid from chlor-alkali electrolysis, increased considerably because the database switched from physical to economic allocation.
I believe that it was a good choice to switch to economic allocation in version 3.1, because the use of by-products can have several economic and environmental benefits and should therefore be stimulated rather than punished. For example, even if the market for by-products collapses because of to the supposed high environmental burden (as is the case with physical allocation), the producer of these by-products will probably not change the production rate, because that is driven by the revenue from the main products. The by-products then become residues or waste and need to be disposed of. When economic allocation is used, the environmental burden of a by-product gradually increases when its market value goes up, to reflect that the revenue from by-products may partially motivate production.
2. Removed Processes
In the case of glass cullet, sorted, the process for glass collection is not included in the network of glass cullet anymore, resulting in a dramatic lowering of the score. However, this extreme difference becomes insignificant when comparing the old and new scores of packaging glass made from glass cullet, because the glass cullet has a minor contribution. I did not find any other products where processes were removed from the network.
3. Changed Markets
In three cases, the scores increased by about 10% to 20% because the markets of important products were extended.
- In case of sulphite (v 2.2) or sulfite (v 3.1), the sodium hydroxide market was extended by Quebec (Canada) and “the rest of the world” (grouping of all other sulphite producing regions that are not represented in the database), where version 2.2 only included average European production.
- The network of petrol, unleaded changed so drastically that it is difficult to find parts that have not changed.
- The most interesting case of changing markets is the nitrogen fertilizer from calcium ammonium nitrate production. The score of this nitrogen fertiliser increased by 20%, whereas the score of another nitrogen fertiliser (from ammonium nitrate phosphate production) decreased by 50% due to a switch in allocation method as discussed before. In the case of calcium ammonium nitrate, the amount of ammonia in the top process did not change, but the origin of the ammonia did. In the old version, the input of ammonia was only from steam reforming production, while in the newest version only 5% originates from this technology and the remaining 95% of the ammonia comes from partial oxidation production, which has a higher impact.
Conversely, in the case of sawnwood, hardwood, raw, air dried, the score was reduced by more than 50%, due to specification of production countries and technologies in the hardwood market.
Many products are affected by this type of changes. Changing scores for electricity from various countries and technologies, for example, are generally explained by changed markets, but also by the use of newer data in version 3.1.
4. Different Land Transformation Approach
All products that originate from palm fruit or soybeans, such as soap and soybean meal, have scores that are much lower in version 3.1. This is the result of a revised modelling approach for land transformation and associated greenhouse gas emissions. There has been much discussion about land transformation in life cycle assessments, and as a consequence there are now several different approaches. If your scores are influenced by land transformation, I recommend you do a sensitivity analysis with alternative data, e.g. from the Agri footprint database or with ecoinvent processes that you adapt according to an alternative approach for land transformation.
Getting Used To The New Situation
The large changes in the ecoinvent data may make it challenging to interpret your LCA results after the update to version 3.1. It can be difficult to explain the differences, because you need to look through long process contribution lists, analyse complex networks, go through the process comments and check the unlinked source data on the ecoinvent website. On the other hand, SimaPro has many features to help you analyse the results with ecoinvent unit processes in detail. After investigating twelve different ecoinvent products, I came up with the four different types of explanations. You will likely find the same.
After doing the update and explaining the differences, a new phase of LCA modelling begins. Links in your projects need to be revised as the new database contains much more products, with different technologies and country sources and completely new products. For new projects, you also need to get used to the new structure and level of detail.
Although working with background data is never ideal, it gives you an idea of the potential impact, whereas working with no data at all just leaves you in the dark. The advantage of working with ecoinvent is that it is the most transparent life cycle inventory database available. It allows detailed analysis of the data and continuous improvement.
Tommie joined the Consultancy Team in 2012, working with databases and methods. He collaborated in projects such as Prosuite and improving the ReCiPe method. He worked at PRé from 2012 until 2015.