Exploring Flooding & Sea Level Rise through AI
Can AI image generation inform our views of sea level rise & climate change?
Over the past month, I’ve been diving into a rabbit hole of AI image generation and exploring its ability to help us visualize the potential impacts of flooding and sea level rise in Charleston and the Lowcountry. As climate change poses significant challenges to coastal cities, can exploring the future through AI-generated images allow us to better understand the consequences of flooding and potential solutions to it? My MeanHighWater Project aims to document flooding in the Lowcountry in hopes of educating and inspiring action. Could incorporating AI into it help?
On June 4th, we were in the middle of a round of tidal flooding in Charleston Harbor. I was preparing to go out and photograph the flooding later that evening, which inspired me to begin asking Midjourney to ‘imagine’ various flood scenarios around Charleston and the Lowcountry. I realized quickly that while Midjourney has an idea of what Charleston is and looks like, it cannot recreate and depict actual places. That poses challenges for generating scenarios that people can easily relate to, and it creates a disconnection between AI-fantasy and reality. The results are also sometimes quite odd (in one instance, church steeples became lighthouses). As I tweaked prompts, I finally generated a few images that generally represented the scenes I had imagined, and I posted a few to Instagram. The responses were interesting.
“beautiful but disturbing”
“these almost glorify/beautify the impacts of sea level rise”
“if only we adapt that well”
“AI can’t row”
Later that evening, the tide ended up peaking higher than predicted, which flooded significant portions of the peninsula. As I was photographing the flooding, I kept saying to myself, “This isn’t AI. This is real!” Perhaps the AI-generated images weren’t just glorifying sea level rise, they were making it seem as a future and distant threat.
I kept adjusting prompts and testing ideas on how to make these images more impactful. What could a future Charleston, abandoned and decaying, look like? These images are hauntingly beautiful and likely plausible if we don’t begin to incorporate flood resilient strategies and implement flood resilient infrastructure faster. But without narrative, these could be interpreted of as images of Charleston or Savannah or New Orleans flooded and slipping into ruin.
I decided to add people and imagine various events to create different narratives. How will students navigate flooded streets to get to classes at the College of Charleston and MUSC? Will locals and tourist still shop on King Street? The results are plausible scenarios (some of these scenes can be seen after a downpour floods streets around the campuses), but do they serve as a call to action?
Finally, I started to test prompts to generate images that show potential mitigations to the impacts of flooding (e.g., canals, sea walls). I really struggled to get AI to understand the concept of a sea wall around Charleston. The resulting images showed a wall, but the land on the inside of the wall was elevated to the top of the wall.
I also prompted AI to generate images of new water management strategies around the peninsula. It’s a topic I covered in my Imagining Our Future with Water talk last year, and finding ways to incorporate surface water into our urban landscape is something I think we must do if we want to become more resilient to flooding. I think the AI results with this are pretty powerful, as they show how water could become an asset and feature, making areas of our City more unique and desirable. It won’t be easy, but I think these images do a good job depicting the possibility.
I’ve only scratched the surface of the abilities of Midjourney and other AI applications with image generation, and I’m sure there are ways to further refine and harness it’s abilities. I see potential in it as a tool to help us imagine and communicate the risks of and mitigations to sea level rise and flooding in the Lowcountry, but I’m not there yet. Do these serve as a call to action or help you imagine what the future could be? Perhaps the images create a conversation, and maybe that conversation will lead to meaningful mitigations.
Meanwhile, I’ll continue to test and refine this process. Stay tuned for more.