Flood Warning Automation – watch-outs and recommendations

One of the key roles of the Environment Agency is to provide a Warning Service to help flood communities prepare for flooding and take action when required.? This takes the form of:

·?????? Flood Alerts – flooding is possible; be prepared.

·?????? Flood Warnings – flooding is expected; immediate action is required.

·?????? Severe Flood Warnings – severe flooding; danger to life.

These warnings and alerts are critical not just for community safety, but also in helping flood communities in better protecting their homes, their livelihoods and their families.? But they are under pressure and the resilience (I bet you read that word again in this blog) of the Flood Warning Service needs safeguarding.?

The Environment Agency is currently looking at using technology to automate elements of the Flood Warning Service on a risk basis to ensure continuity, seeking to develop further the automation used during the industrial action.

As someone who works with AI technologies, I am all for using this developing technology to create efficiencies and to make systems more effective.

However, we need to pause for thought before making such a substantial change which will impact flood communities.

Substantial? I hear you ask.? It is if you look at it from the end user point of view, the very flood communities the Flood Warning Service is designed to protect.? Here are my thoughts on the watch-outs and requirements before any move to automating the Flood Warning Service is implemented.

Whilst the Flood Warning Service plays a vital role within the resilience (told you!) agenda it must be acknowledged that they can also create substantial secondary stressors for flood communities. ?Exemplified by these quotes from my research:

?“It's (living at risk of flooding) a nightmare. It, cause you know, you're on a knife edge from the time that you get the first initial contact from the EA (The Flood Warning System) saying something, you know, the water levels are rising and then you are constantly on alert, and you don't sleep, and you don't, you don't function……”

?“The minute you get a warning in, you're on tenterhooks and we've had warnings every single day for like months on end at times and it gets relentless where you can't leave your house, you can't, you're frightened to go anywhere at all”.

Ensuring that flood warnings and alerts are accurate and issued in a timely manner is therefore critical.?Not just for saving lives and trying to minimise flood damage, but for reducing the psychological impacts of warnings and living at risk of flooding.? Imagine if some of those warnings were wrong.? And this happened during the industrial action.? Imagine the human impacts of that.

Setting the correct threshold for automated flood warnings is critical.?Issuing flood warnings when they aren’t required or issuing them after the fact (both of which happened during the industrial action) at best lowers trust in the warning system and at worse causes flood communities more anxiety and greater damage to their homes and their wellbeing.

• Establishing what the correct thresholds are requires consultation with flood communities and other stakeholders.? Flood communities have not been formally consulted following the automation during the industrial period.? It is critical that this feedback is feed into any decisions about future automation and how it works.?
• This requires up front consultation with Flood Communities before any pilots (see below) are launched.

The automated system must be able identify high-risk areas beyond relying on river levels.? River levels alone are insufficient for accurate flood risk alerts and warnings.? If the Flood Warning Service is to be automated removing human expertise and experience, then the information informing the decisions it makes in terms of when and when not to issue warnings and alerts needs to be increased.

• A catchment approach is required to accurately predict flood risk and this currently requires human knowledge, experience and intervention.? Any automated service will need this source of vital information and guidance from somewhere.
• Factors like storm type, duration, asset condition and failure, the amount of impermeable surface, the geology and geography of the surrounding area and greenfield encroachment need to be factored into decisions about when to raise flood alerts and warnings.

Pilot projects are essential prior to automation to ensure that automation is effective and that flood communities receive accurate and timely warnings.?

A series of pilot projects is required in a range of areas, which involve local flood groups, the Environment Agency and other stakeholders, for example, local businesses.? These need to occur across all four seasons and a range of different catchments and forms of community, for example:

• Communities with a higher proportion of elderly, disabled or disadvantaged residents will face different challenges in responding to flood warnings compared to affluent, young and able-bodied communities.?
• Coastal communities.
• Transient communities.
• Areas with a higher proportion of young families.

Continuous and thorough community engagement must be conducted throughout the pilot period and afterwards.? The form of the consultation needs to be independent, quantitatively robust and include multiple open-ended questions to ensure that flood communities experiences are fully captured. Note: there are many AI technologies which are competent at analysing open-ended questions so this is no longer a barrier to the use of open-ended questions.

• It is also important to capture the views and experiences of Environment Agency team members who were involved with the flood warning system either issuing or responding to warnings during industrial action.? ?This information can be used to inform the development and implementation of the pilot projects.

Success criteria should be set with the involvement of flood communities prior to any pilot projects starting.

• The results of the pilots must be robustly measured against these success metrics.

Summary

• An automated service needs to be co-designed with communities, and stakeholders?including the experiences of Environment Agency team members.
• The automated service needs to be rigorously tested through the implementation of pilot projects involving communities and stakeholders.? It’s resilience (again!) needs to be challenged.
• Environment Agency team members and flood communities must be consulted about their experiences of the pilot project and this information must be used to adapt and upgrade the automation system prior to launch.

?The introduction of automation for the coming winter is premature and potentially very risky.? The National Flood Forum (https://www.dhirubhai.net/company/the-national-flood-forum/) is expressing extreme concerns about widespread automation without adequate testing, feedback and adaption.

?I would be exceedingly interested to hear people's experiences of automation whether they are an end user or someone involved with the Flood Warning Service.