Looking at decision-making processes for climate adaptation with the Met Office and assessing how the HACID (hybrid human artificial collective intelligence in open-ended domains) tool could support such processes
Cities are under pressure to minimise the impacts of climate change on urban life. From local councils to utility and transport companies - many different organisations are responsible for managing roads, preparing infrastructure for extreme weather and responding to the other localised impacts caused by the changing climate. But navigating the scientific evidence and the evolving data landscape on climate projections to understand their practical impacts isn’t easy. Organisations often turn to expert climate services providers to help them develop climate adaptation plans.
A climate service is the provision of climate information to help end users make climate-smart decisions. Climate information can range from seasonal forecasts (eg, forest fire outlooks) to long-term projections (eg, sea-level rise).
Climate service providers (such as the Met Office) face four main challenges that impact their ability to support public organisations with climate adaptation.
HACID is a European project to develop novel tools that leverage collective human and machine intelligence to support decision-making in open-ended domains. One key application domain focuses on climate adaptation and is being developed in close collaboration with our project partners Met Office, CNR and MPG. Our approach combines three key methodological innovations:
Knowledge graphs (KGs) organise data from multiple sources. Unlike typical databases, they capture information about data points of interest in a given domain or task (like people, places or events), and the connections between them. They are also “smart” in the sense they can infer new connections between data points that weren’t originally connected. AstraZeneca, for example, built a biomedical knowledge graph with heterogeneous biomedical data such as genes, proteins, compounds, and diseases. The knowledge graph supports them in finding connections in the data and predicting new disease targets for drug discovery.
Through our approach, we hope to combine crowdsourced input and other sources of data to uncover new solutions for climate questions and accelerate climate adaptation.
We spoke to three public organisations to understand their needs (and their partner’s needs) when using climate data and the role that the data plays in their climate adaptation decision-making. We found that climate decisions are important for business operations and asset management. Within these areas of focus, public organisations are facing three main challenges.
“Very few people (within the organisations) have the expertise (in climate science) and need intermediaries to interpret the data. The challenge is how to make it accessible at scale” Interview with Climate Resilience Manager working on a regional Adaptation Programme
Some organisations report that climate data is currently inaccessible - hard to find, download and work with (because of non-standard data formats). This challenge coupled with a lack of in-house specialists, means they rely on costly, externally provided climate services to interpret data, rather than being able to explore the data independently themselves.
“How often will the temperature go above 40’C in the London area over the next 25 years? Should we replace the cooling system on the trains?” Examples of relevant climate adaptation questions surfaced during an interview with the Strategy & Planning Manager of a transport organisation
Climate models typically refer to large areas (10km areas) and lack the spatial granularity necessary to make local decisions (city scale). In addition to this, it’s difficult to get information on when and where thresholds will be surpassed (eg, when and where flooding will happen). These two factors make decisions about procurement and adaptation of current assets difficult.
“The challenge is how to consistently use emissions scenarios or adaptation plans. Most of the value would come from helping stakeholders to deal with uncertainty, interpret [climate data] and enable collaboration between different organisations and stakeholders.” - Interview with Climate Resilience Manager working on a regional Adaptation Programme
There is currently no official guidance on what emission scenarios to consider for climate projections (emission scenarios are possible future development pathways of human greenhouse gas and aerosol emissions). Participants reported that inconsistencies in processes and the use of climate data make collaboration with other partners difficult.
The process of developing a climate service typically involves three high-level phases: planning, creation of content, and final delivery. Research with internal Met Office employees on the high-level process of delivering climate services (data provision, briefing documents and training) revealed three main challenges in current practice.
This is one of the three user journeys created as an output of a Met Office workshop. The workshop focused on capturing the end-to-end process of delivering climate services and the challenges at each step.
The challenge is "understanding the actual problem at hand (eg, misunderstandings caused by our customer themselves being a step removed from the user)" – Met Office workshop
In the planning phase, the largest challenge lies in meeting broad and changing user requirements. This often requires a long customisation process to ensure each climate service is helpful and supports the organisations in making climate adaptation decisions.
“... scientists are sometimes reluctant to say that the data is good enough to use. It would be preferable if they could provide the uncertainty characteristics of that data” – Delivery team member, Met Office workshop
Climate projections are uncertain because it is hard to determine the path of future emissions (whether emissions will increase, stabilise or decrease). As our understanding of climate systems evolves due to new information, data, and models - new uncertainties can also emerge. Finally, the tolerance for uncertainty can vary between different stakeholders.
“Explaining the data properties and the uncertainty related to the data is challenging. It is important that the user understands what the data should and should not be used for.” – climate scientist, Met Office workshop
Explaining where the data and models come from and how the data should (and should not!) be used is difficult. It requires that climate service providers act as translators between the very technical models and the very practical needs of external organisations. For example, the level of accuracy needed for a public awareness campaign will be different from the accuracy needed to calculate the elevation of a new hospital building to avoid flooding.
We set out to create a decision-support tool that centralises knowledge, supports climate scientists dealing with data uncertainty and promotes explainability. The HACID tool will support climate scientists by allowing them to:
We hope that, through the creation of this expert community and centralised knowledge base, in future, we can open the platform to public organisations. In HACID’s future vision, public organisations would be able to ask questions directly and receive valuable information to support their climate decision-making.
We developed a concept design for a new tool and outlined the envisioned user journey.
The service blueprint documents the interactions between actors, actions and functionality within the HACID-DSS concept. It also includes wireframes from the prototype.
Our partner the National Research Council of Italy (CNR) will lead the technical development of a HACID prototype for user testing. The prototype will be based on the user research and concept design described above.
During this phase of the project, our team will be working on designing and delivering the participatory evaluation process and helping partners to adapt the tool development process to be more participatory (eg, testing new methods for eliciting inputs from experts during the development of the knowledge graph)
This prototype development and the evaluation of the process will help with further iteration of the participatory AI methodology first developed by Nesta in the context of humanitarian innovation.
We’ll be sharing what we learn along the way, so look out for more project updates. In the meantime, if you’d like to learn more about this project, get in touch at co[email protected] using the subject line HACID.