Building a Climate Resilience Demonstrator

How can digital twins help us respond to climate change? We asked Hartree Centre Data Engineering Specialist and Technical Architect on the Climate Resilience Demonstrator project, Tom Collingwood, to tell us more. 

With all the recent storms causing damage and flooding across the UK, it couldn’t be a more relevant time to talk about climate change!

To start us off, can you explain to us what climate resilience means?

It is very topical right now with all the storms we’d had recently and it seems to be a developing issue that is worth looking at. Most people will be familiar with the idea of climate change causing all kinds of disruption – from flooding to droughts to storm damage. On the extreme end of the scale, they can pose a threat to our safety whether directly or indirectly by disrupting an essential service or system – for example, the power going out in a hospital, or emergency services losing signal on their way to an accident.

This kind of work is distinct from trying to slow/stop climate change – which is incredibly important too – instead, we’re trying to help inform the understanding around what might happen to our infrastructure if/when more severe climate events do occur, and hence how we might prioritise resilience planning around assets which are crucial to whole-of-system resilience.

This means there is huge potential in terms of damage prevention, cost savings and service reliability for immediate services like telecoms, energy, water and utilities – but these also cascade down to any industry that relies heavily on or would be affected by disruption to those services. Which is pretty much all industries!

So that’s where the Climate Resilience Demonstrator (CReDo) comes in?

The Climate Resilience Demonstrator, CReDo, is a digital twin demonstrator project to improve climate and extreme weather resilience across infrastructure networks – the first of its kind in the UK. We narrowed down our focus to look specifically at the effects of extreme flooding on the communications, power and water networks in a specific area of the UK. We have developed a prototype digital twin that takes in data from climate, water, utilities, telecoms and energy industries and applies flood impact models which predict where flooding will form in the UK, how those floods might affect the equipment they touch and how knock-on impacts spread out to the rest of the networks outside the immediate flood zones.

We wanted to demonstrate how those who own and operate infrastructure can use secure, resilient, information sharing, across sector boundaries, to mitigate the effect of flooding on network performance and service delivery to customers, so we’ve been developing reliable approaches and frameworks for secure data sharing and information management that can inform this kind of model and be scaled up.

Why is a digital twin useful when tackling the challenge of climate resilience?

Lots of niche areas of utilities and telecoms will have specific experts or teams who have been responsible for that same machine or equipment for the last 50 years and if you lose that person or team – all that operational knowledge goes with them. If you start using digital twins and connected data, you need to get that specialist information out of their head and turn it into models that can run in AI and machine learning systems, providing 24 hour access to that information.

Many people think of digital twins as operational tools streaming live data from sensors and adjusting ongoing processes accordingly, such as in a manufacturing facility. With climate change, the feedback loop we are looking at might take 100 years to complete, so in this specific use case our digital twin isn’t streaming live data and instead is operating in a way which provides resilience planners with predicted outcomes for a given set of inputs, so they can use the information when making decisions about the future networks they’re supporting. We’re only scraping the surface of what digital twins can do for climate resilience with this specific use case. Bringing operational sensor data into the mix (from river levels to real-time asset monitoring) would broaden the application out to explore current and potentially upcoming failures via predictive maintenance modelling, or branching out into other climatic effects such as wind and extreme heat to inform how we make the whole network more resilient to a variety of new challenges over the coming decades. You’re building the foundations for a digital decision support, and potentially future decision-making, assistant that always gives consistent advice to actively support the experts making vitally important decisions about our countries’ infrastructure.

“That’s the thing – if you get this kind of work right, basically no one will ever hear about it. Life goes on as normal, the power stays on, the communications don’t go down and the damage is minimal.”

Tom Collingwood – CReDo Technical Architect

How do you teach a computer to do that?

You have a structured conversation with the experts, you ask them to tell you how things might break – even in strange or temperamental ways you wouldn’t expect – and you incorporate all those cases to develop a model that provides more accurate predictions. The more you know, the more data you have to keep running through the system to refine it and make better decisions and better decisions in future.

Can you talk us through an example to illustrate what kind of scenarios you’re modelling?

One of the examples we looked at was a water pumping station. So we had to factor in variables like knowing what will break if the water reaches a specific depth because that would submerge the electronics and potentially start a fire. Or if the fuel has been stolen from a backup generator, which will mean everything switches off in an emergency – but imagine cases where there are no sensors detecting whether the fuel is still there.

Our approach means that in the short-term we look at the statistical probability and frequency of those factors to make more accurate predictions of when and how failures might occur. In the long term we’ve discovered what data would be useful so you can put the technologies in place – in this case, you’d install fuel level sensors in the tanks.

So the process goes something like this:

  • Learn from experts what variables affect potential failures or faults
  • Make a plan for which data you need to start collecting
  • Create a model that uses that data to make predictions, and provides sensible approximations where the data aren’t readily available to the system yet
  • Keep feeding in new data to refine the models over time
  • Review the outputs of the models with the experts running the machines/assets, and tweak as necessary to ensure the models give sensible outputs using the current information at hand
  • Use the predicted outputs to inform plans to mitigate the failures

So with the flooding example, you can’t stop the weather but you can predict when it’s likely to happen and put up defences in time to minimise damage or disruption?

Exactly. And the next stage is to look at what knock-on effects happen when a fault or failure occurs – so if it’s a power plant that went down, everything it supplies power to has now lost its primary power supply. What would that mean for vital infrastructure, like healthcare? This was what the short film we funded through the project was exploring – that something like loss of power – even over a short period – can actually be life or death.

The ultimate impact of a single asset going down isn’t something which is immediately apparent – we have to cascade those failures across multiple networks throughout the system if we want to understand the real impact, and with complex network interdependencies that’s not an easy thing for humans to resolve quickly, whereas the right computational models can be very well suited to doing this quickly and repeatably.

Short film “Tomorrow Today” was produced by the National Digital Twin programme and Climate Resilience Demonstrator to explore the potential impact of digital twins.

What was the Hartree Centre’s role in the project?

The Hartree Centre was brought into the consortium originally to provide leadership of technical delivery, and I was given the role of Technical Architect accordingly. This meant my job was to oversee the successful delivery of a technical plan, so I had to do a bit of planning first and then ensure we could make it happen. We also had several other members of our Data Science and Research Software Engineering teams working on different aspects of data analysis and code optimisation for the project.

I’ve had oversight of what’s being done across the consortium of project partners: STFC’s Hartree Centre and DAFNI, CMCL Innovations, the Joint Centre of Excellence for Environmental Intelligence (JCEEI), the National Digital Twin Hub and the Universities of Edinburgh, Warwick and Newcastle.

On the industry side, Anglian Water, BT and UK Power Networks provided infrastructure data and Mott MacDonald supported us with domain expertise in infrastructure and flood modelling.

That’s lots of pieces to bring together!

Yeah, it’s a massive and quite complicated stakeholder map with a lot of moving pieces! So I’ve spent a lot of the last year joining the dots and doing agile programme planning. We’ve approached it with telecoms, water and utilities providers as the “customers” we had in mind as they’re the ones who would ultimately be able to benefit from the outputs of the project and use them to increase reliability and functionality.

A bunch of very talented people were put in front of me and I had to figure out how we could deliver as much as possible simultaneously and get it all done in time for the close of the project. We set up a secure cluster on DAFNI to put data from the asset owners all in one place, so that the scientists working on the project could access it and connect it up to develop models, without it being shared or accessed by anyone else.

What are the next steps?

The project comes to a close in March 2022, so we’re currently writing up the reports and planning a webinar to present our experiences, talk about technical achievements and lessons we’ve learned along the way so that hopefully others can learn from them too and continue to develop our ideas.

The project partners are going to collate reports and write executive summaries so we have something to help us engage with business leadership audiences that are less technical but have decision-making authority to try implementing these concepts at scale. The technical reports are there in more detail so that technical staff can understand what needs to be done.

We’re also going to continue working with the partners on this project to seek funding for the continuation of development, and hopefully further scaling up of this project. Watch this space!

Find out more about the Climate Resilience Demonstrator.

Missed the show-and-tell webinar? Watch it now

Read the technical reports

Caught in the data loop?

Fresh from the Open Data Institute (ODI) Summit 2019 and bursting with questions, Holly Halford, Science and Business Engagement Manager for the STFC Hartree Centre, explores the use of personal data for online marketing and asks: how do we stop ourselves getting stuck in the data loop?

So, your friend is getting married. You post a few harmless pictures on Instagram, throwing in a few #wedding tags for good measure. The next day, you’re scrolling through your social media feeds and perusing news sites only to find that every sponsored post, every inch of ad space is now trying to sell you wedding dresses. Wedding venues. Wedding fayres. Decorative wedding trees. Things you didn’t even know existed – all useless to you and, presumably, the advertiser – but the ads are still there, taking up precious mindshare.

But you asked for this – you were the one who carelessly hashtagged your way into the echo chamber… right?

From targeted advertising to political persuasion, whether to help or hinder us, our personal data is being used on a daily basis to effect changes in our behaviour. From the extra purchase you didn’t really need to make, to the life milestones you are forced to start thinking about because your data fits a certain demographic.

New research, conducted by the ODI and YouGov and published to coincide with the recent ODI Summit 2019, concluded that nearly 9 in 10 people (87%) feel it is important that organisations they interact with use data about them ethically – but ethical means different things in different contexts to different people. In discussion at the conference, Prof. Nigel Shadbolt and Sir Tim Berners-Lee highlighted that research shows people are reasonably accepting of personal data being used for targeted advertising, but less amenable to it being used for political advertising. Tim proposed a possible reasoning for this, positioning himself as in favour of targeted commercial advertising – at least towards himself – as it generally helps to find the things you want faster, and also helps companies to make the sales that keep them in business. A “win-win” for both consumer and economy, then.

Sir Tim Berners-Lee in conversation with Professor Nigel Shadbolt and Zoe Kleinman at the ODI Summit 2019.

He suggested that political advertising is different in nature because it may make people act in a way that isn’t truly in their own personal best interest due to a manipulation or misrepresentation of information. It’s of course, possible to argue that the same can be true of misleading commercial advertising but the potential impacts are almost always limited to being purely financial – spending money you didn’t need to, getting into debt etc – and these ramifications are not significantly different to the pitfalls of marketing via any other route. Traditional print media, billboards or television advertising have all probably promised you a better life at some point, if you just buy that car, that smartphone or that deodorant.

Tim has a point – targeted advertising can be useful and makes some logical sense, especially if we have actively searched for related terms or shown our interest in a certain product or service by interacting with content related to it. Despite how 1984 it can feel sometimes, I’m actually personally much more comfortable with data-driven advertising based around our active behaviors as opposed to the other option – the demographic based approach, which I feel has the potential to be far more insidious.

There’s a beauty product advert in my Facebook feed. If I click on the “why am I seeing this” feature, I am quickly informed that Company X “is trying to reach females aged 25 to 54”. Whilst the transparency is a welcome change, it doesn’t fill me with hope that a significant proportion of the media thrust upon us each day is tailored based on nothing more than gender or other divisive demographics. I often wonder how many men have beauty product adverts showing up in their feeds compared to say… cars, watches, sporting equipment? (I unscientifically and anecdotally tested this theory on a colleague recently, a man in a similar age bracket to myself. He reported an unusually high capacity of DIY ads.)

Credit: Death To Stock

The data bias is there, entrenched in historic trends that have potentially damaging consequences in the perpetuation of gender stereotypes and more – if your demographic fits the initial (and undoubtedly biased) statistical trend, do we now, via data-driven marketing, perpetuate it for all eternity?

But how do we address the very fundamentals of marketing and communications without perpetuating stereotypes and pushing conformity to social norms? As a marketing and communications professional, I confess that the commonly used concept of developing “personas” to describe your target audience and help articulate your message more clearly to them has never sat well with me, because those personas by nature are based on stereotypes and assumptions. Knowing your audience is an absolutely crucial pillar of marketing, but if you only ever acknowledge an existing or expected audience, how do you access new markets and prevent alienating potential customers outside of that bracket? Not to mention the ethical concerns this approach flags up. We need to take a more creative approach to get messages heard without excluding anyone. It may not be the easiest route but I’m certain that it is possible, more ethical and when executed successfully, more effective.

So, what can we, as consumers, do to prevent trapping ourselves with our own hashtags and search terms? The current options seem fairly lacking. Perhaps we can turn to AI-driven discovery of “things you might enjoy”. Features like this can be found on most common media platforms, with varying degrees of success. But as the algorithms get more accurate, the tighter the loop closes. As Tim purported, the intention is to be helpful and save us time – if only to provide a good user experience that keeps you invested in using the platform, of course – but everything it suggests will be based on existing tastes and activity. If you’re predisposed to playing Irish folk music, good luck getting Spotify to suggest you might have an undiscovered a passion for post-progressive rock.

Credit: Death To Stock

This presents a bigger problem when considering the landscape of opinions, causes and politics. The idea of social media curating our own personal echo chambers and arenas of confirmation bias is not a new one. It’s true that we can subscribe to contrasting interest groups, a tactic some journalists have been using – but how many of us have the patience to subject ourselves to a cacophony of largely irrelevant content, if it’s not a professional requirement? A more pressing question is: if we don’t interact positively (or at all) with that “alternate” content, does another algorithm begin to de-prioritise it until we no longer see it anyway and we’re back where we started?

Is the answer in a change of algorithms, then? The tactic of ignoring trends and demographics seems to be entirely at odds with the notion of creating better, more accurate AI algorithms and data-driven technologies. Whether we like it or not, they are meant to do exactly that – generate accurate predictions based on statistically evidenced trends and demographics. I feel quite strongly that a great deal more creative thought is required to ensure that ethical practices and regulations are instigated in line with the pace of technological advancement, and prevent data-driven marketing from driving us round in circles for the foreseeable future.

Afterword: I wrote the majority of this blog post before the launch of the Contract for the Web recently announced by Sir Tim Berners-Lee. It presents an encouraging and much needed first step towards safeguarding all the opportunities the internet presents and championing fairness, safety and empowerment. Now, let’s act on it.