If we want to know how our climate will behave in the future, we need emissions projections for a long list of different chemicals. However most socio-economic models do not provide the full range. Our latest work provides python code that allows modellers to fill out incomplete models.
In order to project climate features, sociologists and economists must make assumptions about technological developments, international norms, national policies, lifestyle choices and population changes to produce what is called an Integrated Assessment Model (IAM). These require many assumptions and are not firm predictions of the future, but internally consistent pathways that the world may take.
Collections of IAM scenarios with common social or political features from different models can then indicate what the effect of a certain type of policy might be. However most models do not provide estimates for many of the substances we emit – beyond just CO2 and methane, there are a host of less-known substances, from aerosols that make smog and change how clouds form, to a huge range of fluoridated gases that gram-for-gram can exert hundreds or even thousands of times the impact of carbon dioxide.
Our infilling tool, Silicone, looks for relationships between a commonly modelled emission (e.g. CO2) and a rarer emission (e.g. NO2) in all the scenarios that model these two emissions. It then uses that result to recommend a value for the rarer emission for IAMs that only project the more common emission.
It provides a wide range of possible relationships to choose between. These include assuming direct proportionality between the emission types, interpolating between the results in complete scenarios and performing a type of quantile regression. It also provides a set of specialised tools for breaking an aggregate value (like Kyoto gas total) into components in ways that ensure the total is conserved, or for infilling many emissions in a similar way. It’s important to note that this is a toolkit, not a magic wand – care should be taken choosing the infilling technique and the range of complete scenarios to perform the infilling, and if the scenario being infilled is radically different to those that have a complete set of data, then we cannot have confidence in the results.
In this figure, you can see our code in action, infilling the volatile organic compounds (VOC) results (unknown for the dotted line) from the CO2 results, known for all pathways, using a smooth quantile regression technique called Quantile Rolling Windows. This technique always produces responses within the limits of known scenarios, hence on the right the output dashed line is always inside the coloured lines.
The code is all open-source so additional infilling techniques can be added as required. Please let us know if you want to contribute!
For further information contact firstname.lastname@example.org.