Calculating the "additionality" of deforestation avoidance interventions

Prof Srinivasan Keshav ( @ProfKeshav ) is using computer science to move towards a sustainable future through clean energy and environmental conservation. He is a co-director of 4C - the Cambridge Centre for Carbon Credits - and the Robert Sansom Professor of Computer Science in the Department of Computer Science and Technology at the University of Cambridge.

When evaluating forest protection projects, the concept of additionality is crucial. A landowner might claim to have saved a patch of forest from destruction, and ask to be paid on that basis. However, it could be that the forest was not saved as a result of their actions – it might be that the forest was never going to be destroyed in the first place, whether that landowner managed it or not. We want to pay only for services rendered – for "additional' value conferred.

How can we determine whether or not that was the case? It is impossible to prove what would have happened in the absence of an intervention, but we can gather evidence to make a case using counterfactual scenarios.

For example, if there are two identical patches of forest adjacent to a road, one of which was managed by the landowner, one of which was not, and after twenty years only the forest managed by the landowner remained unfelled, then that would make it more likely that the landowner had made a positive difference to the fate of the forest.

This concept can be generalised – we can take samples from within a project area, called ‘pixels’ and pair them up with other pixels taken from outside the project area, ‘counterfactuals’. These pairs of pixels are matched in all major characteristics – forest cover, distance from roads, elevation, soil type, rainfall etc. If we do this well, we expect the paired pixels to experience similar rates of deforestation.

These pixel pairings can then be monitored throughout time using satellite images taken over many years. If there is no meaningful difference in the way that the project area is being managed, then we would expect the pixels in the project area to experience very similar deforestation trajectories to the counterfactuals, on average. However, if the protected area is conferring truly additional protective value, then we will see the trajectories diverge – the counterfactual pixels will experience greater rates of deforestation compared to the project pixels. The difference between those trajectories represents the additionality of the project.

Using algorithms, we can perform hundreds or thousands of these pixel matches in seconds, providing a very reliable evidence base for any differences between the project area and its matched controls.

Thanks for this summary to James Miller, student environmentalist and film-maker.