(If you would rather watch a video of this post, see this TEDx talk!)
When I was 11 years old and living in Taupō, here in Aotearoa New Zealand, the nearby volcano, Mt Ruapehu, erupted. I remember seeing a giant grey ash cloud cover the otherwise blue sky. Volcanic ash fell in a light layer on our town, and I collected a sample into a little jar off our car bonnet. We filled up an old pool in our basement, and the bath, for drinking water. I thought it was a great adventure!
However, I still had to go to school. My pregnant teacher was not so excited - she was anxious about the ash and the potential health impacts. I had a friend who lived on a sheep farm, where many of the animals had died after eating the grass covered in ash. My uncle was meant to be visiting us, but his flight was cancelled.
We are all aware of how devastating natural hazards can be, from volcanic eruptions, to storms and floods, earthquakes, tsunami, landslides, and wildfires. But looking back on Ruapehu’s eruption now, you can see that everyone has different things going on in their lives that means we are affected in different ways by hazards. Humans are complex.
What if there had been a way for all of us to receive a warning before the eruption occurred that reflected our different situations and helped us to take the most appropriate actions? Would we have found this useful?
Globally, and here in New Zealand, some areas have been hit really hard by cyclones, typhoons and hurricanes in the past decade. With a changing climate, we can expect weather-related hazards to become more severe. One of the best ways to mitigate the impacts of events like these is to take action before it arrives, using forecasts and warnings.
Forecasts have been improving
The forecasting of hazards has come a long way. Weather forecasts issued days to a week before an event are now much more accurate than they used to be, giving us more time to get prepared.
In my role as a social scientist, I conduct research on how we can make more effective forecasts and warnings. I have seen how forecasts have improved for earthquakes, tsunami, landslides and volcanic eruptions. Over the past decade, we (while I was working at GNS Science) produced forecasts for aftershocks following the Canterbury and Kaikoura earthquake sequences, eruptions at Whakaari White Island in 2019, Ruapehu, and Taupō supervolcano, and landslides for Cyclone Gabrielle in 2023.
These forecasts are mostly based on the severity of the hazard, not the potential impacts. Weather warnings have also traditionally been triggered by the severity of the hazard. For example, if the wind speed was forecast to be more than 100km per hour (60 miles per hour), then a weather warning would be issued. This is whether the area has a city on it, or is rural farmland.
A move towards impact-based warnings
But things are changing. The World Meteorological Organization has led a call for more meaningful warnings, that help people to understand what the weather forecast means for them, their families, livelihoods and properties.
Some countries, including New Zealand, are starting to develop warnings that are triggered by the level of expected impacts. These are referred to as impact-based warnings. Instead of a blanket threshold triggering the warning, the threshold is different between different locations, depending on how resilient they are. The threshold for issuing a warning in Wellington city might be 120 km/hr for wind gusts, in comparison to 100km/hr for Auckland city (these thresholds change and are just used to illustrate the point). This is because Wellington gets so much more wind, and so the trees, infrastructure and people can handle the 100km/hr wind gusts better.
In future, the threshold could be further refined to reflect how people’s movements change throughout the day and seasons. The threshold could be lower in a city during rush hour, while more people are out on the roads, and higher at other times. It could take into account whether it is summer holidays, or there is an outdoor football match or concert planned, so more people are outside and in specific locations. The potential damage to infrastructure, like electricity lines, and disruption to transport, could also be considered.
How effective are impact-based warnings?
Would warnings based on the level of expected impacts be better at helping people to respond? I have conducted research with New Zealand’s MetService on whether impact-based warnings are more likely to change people’s risk perceptions and prompt them to act, than traditional hazard-based warnings. (Here is the link to the paper.) The findings are promising. People who received the impact-based warning had a greater understanding of the possible consequences of the storm. They also had higher levels of concern about the storm than the survey participants who received the hazard-based warning.
This shows us that warnings that describe the potential impacts in more detail are better at changing people’s risk perceptions. Whether impact-based warnings are actually more likely to prompt people to take actions to prepare for the storm is a bit less clear, so globally, we are working to understand this some more.
Following our research, MetService went on to introduce red weather warnings, which are triggered when significant impacts are expected. For example, in February 2023, Cyclone Gabrielle was forecast to sweep past Auckland. Only two weeks earlier, Auckland was hit by significant floods, so they knew that the catchment was already soaked. This meant that further floods were more likely, so they issued a red warning. These red warnings are issued on a regional scale. Many people in Auckland were affected by the floods, but they were affected in different ways.
Tailored warnings
What if we could find a way to give more targeted and tailored warnings to individual people in the future? This way, when a cyclone is arriving on our shores, or a volcano is erupting, we will know whether we may be affected or not, how, and what we should do about it.
Below is an example of a future warning. Let’s say I am in my kitchen and my device alerts me that in three days' time, a cyclone is going to hit my area. It tells me that the resulting flooding will cut off the road to my house, isolating my family for a week. It suggests a shopping list for food and water supplies that it knows I am running low on, including for my cat and children. It suggests that I may wish to reschedule the haircut I had finally got around to booking. It has a link to advice on how to prepare my weatherboard house for the storm, and could also link to accommodation should I wish to evacuate instead.
Which agency sent this warning? Can it be trusted? How did it know that I have a cat and children – or that I had a haircut booked? Did my elderly neighbour get the same warning, at the same time?
Using digital footprints
National science or emergency management agencies don’t tend to have real-time information about individuals, nor the time to write 5 million targeted messages, so how could tailored warnings possibly be produced?
What if our digital footprints fed into warning systems? Tech giants have a lot of our data already in their systems. They could use artificial intelligence techniques to process that data and figure out how susceptible we are to impacts from the coming cyclone. For example, they might be able to use our health-related and demographic data, our online search history, information on how robust our housing and infrastructure is, and potentially see our online photos.
They could therefore know that I am female, that I have a cat and children, live in a weatherboard house, and that our local road may get flooded in a storm of this intensity. They know if it is autumn and the leaves are clogging drains leading to a higher chance of flooding, or if it is summer with trees in full leaf and more likely to get blown over onto powerlines or block roads.
The tech giant could forecast using machine learning where we are likely to be in comparison to the cyclone forecast by looking at our mobile phone location and travel history data, our calendars and social media data.
In this way the tech giant knows about my booked haircut and any other planned trips. By combining all this information, they could calculate our individual risk levels. They could know whether my neighbour or I are most at risk, and why.
Tailored guidance messages, and timing of delivery
The tech giant could also know what guidance information is best suited to each of us. A smart fridge would know what food I do and don’t have, and could cross check this against a list of recommended emergency food. An online shopping list could then be prepared for me to approve in my targeted warning, supporting me to take action.
By knowing the location, material and design of my house, they could provide me with tailored advice on whether I can clear the spouting and shut the windows and I’ll be alright, or whether I can expect problems and should be evacuating. My accommodation booking and search histories could be used to provide me with tempting options.
As well as figuring out our levels of risk, our tolerance to that risk could be known. This is the point when we think it’s worth doing something to make ourselves a bit safer. Our many and varied responses to the Covid-19 outbreak demonstrated how different our levels of concern were.
Our purchase histories show whether we bought masks, and digital advertising boards in malls with facial recognition software might know if we actually wore them. Our location data shows whether we were happy to frequent crowded night clubs or preferred outdoor dining or staying home.
Our social media activity and which articles we read informs the tech giants how we think and feel.
By overlaying our individual risk with our risk tolerance levels, personalised warnings could be issued at the time that is most useful to us. AI-powered large language models could write those warning messages, tailoring the content based on all that underpinning data.
This is the way I could receive a warning about flooding cutting off the road to my house before the cyclone arrives, with options provided for actions that are tailored to me.
Problems with this approach
This could be the future of warnings. But I’m not necessarily saying it should be. There are a lot of potential issues with this sort of approach.
Commercial gain could drive the warnings, instead of public good.
Tech giants could suddenly axe the system if they decide it’s no longer wanted.
Multiple alerts could be issued by a range of providers. This could lead to information overload, increased anxiety or warning fatigue, which is cry wolf syndrome.
There are bound to be privacy issues, and biases in the algorithms.
Whether or not all the digital footprint data I described are held by an individual tech giant, or can be shared between them, we don’t know. But we do know that the power that these corporations have over us is immense, meaning we often accept their terms and conditions, and allow cookies, in order to be able to use our phones, apps, and platforms.
Our data is out there and is already being used to target advertising, predict products that we might be interested in, and give us Netflix suggestions.
Alternative solution?
This scenario of using our digital data to inform warnings is not currently happening, as far as I’m aware, but I think the chances are high that it will occur. We need to research every aspect of it to see whether tailored warnings are wanted and useful, and explore other options that could inform useful warnings, beyond using our personal data.
Perhaps instead, and this is VERY IMPORTANT, we could define our own preferences directly with warning agencies for when we want to be warned, what impacts might be relevant for us, and what sort of message content would be useful - without our personal data being used.
If tech giants do issue personal data-driven warnings, I think they would need to be issued alongside and be consistent with a regular more generic system to ensure everyone gets a warning, even if they don’t have a smartphone or digital footprint.
Agreements need to be sorted between tech giants who hold all of our data and could potentially use it for warnings, and the agencies and governments of countries who have existing warning systems.
We need to investigate how far down this digital footprint path we should go. The tricky thing is, the path is still under construction and I’m not sure anyone knows where it is going. As stated by Dennis Gabor, the future cannot be predicted, but it can be invented.
Into the future
Perhaps one day in future when Mt Ruapehu erupts, we will be able to produce forecasts that provide more tailored information about how people will be impacted, and what they can do about it.
A family with an 11-year-old budding scientist might receive safety information alongside how it is a good opportunity for curious kids to learn about volcanoes. Pregnant teachers could receive targeted advice about health risks from the ash. Farmers could get information about how their animals will be affected and what to do. And travel disruption information can be given out to those with flight bookings before the eruption happens.
When you receive a forecast or a warning for any hazards, I encourage you to pay attention to it. Maybe one day you will notice that your warning refers to your personal situation, and is different to your neighbours. Think about how you would prefer to be warned in future, so that you can take action, and better weather the storms of tomorrow.
Comments