28 October 2022 |

AI for adaptation


Over the weekend, a WSJ article made the rounds on Twitter. Posted alongside captions that ran something to the effect of “damages from flooding are less bad now than they were 100 years ago” (I’m not going to link them here), the usual litany of climate deniers happily spread it far and wide as evidence that fears about climate change are overblown. 

For one, the article did acknowledge that most of the ‘improvements’ in this area come not from floods being less significant in magnitude but from better technology and flood preparedness diminishing floods’ negative impacts. More importantly, one of the main cited statistics was damage from floods as a percentage of GDP. Considering real GDP has grown ~20x in the U.S. in 100 years, any relative ‘diminishing’ of flood damages strikes me as driven by the denominator, not the numerator. 

Further, the article only focused on the U.S. Catastrophic, quasi-biblical floods rocked Pakistan and Nigeria this year and Germany last year. Plus, the U.S. has seen its fair share this year, unfortunately, too. Over the summer, five “thousand-year” floods hit in five weeks.

I’m not a climate scientist, and I think it’s complicated to attribute any weather event, or even a set of weather events, to climate change directly. But a warmer atmosphere does retain more moisture. And an atmosphere that retains more moisture makes for heavier rains and, sometimes, floods. I’m equipped to handle that much basic science. 

To better understand flooding, the risks it poses, and what’s needed to adapt, it also helps to have an expert source. Luckily, my latest podcast recording presented an opportunity. 

Satellite imagery → better flood maps

Once in a while, you land a podcast guest who does most of the heavy lifting. Not just in driving a thoughtful conversation but in linking their area of expertise to other climate tech conversations. Recording a podcast with Subit Chakrabarti of Cloud to Street exemplified that experience. We discussed everything from the burgeoning ‘climate data’ economy to making better flood maps (Cloud to Street’s business) in a wide-reaching conversation. 

By climate data economy, I’m referring to an expansive world of data that wasn’t available a decade or two ago. Satellite imagery is an appreciable example. Last month I wrote about the growing number of companies whose mission is to make satellite imagery more available and accessible. To be sure, satellites have orbited Earth for the better part of the past century. But as Subit and I discussed, they used to point outwards at the vast infinity of space.

SpaceX satellite over one of Earth’s coastlines

Now? Not only are companies launching many more satellites than before into space, but they’re also pointing more of them back at Earth to collect observation data for a range of purposes, from expanding internet access to taking pictures. 

There’s a nice metaphor inherent to that phenomenon. While exploring space is exciting, saving the planet is paramount too! As Subit noted:

…advances in imaging science in general, which are well complemented by advances in signal processing and the utilization of imagery… are happening together at the right time to enable new and more efficient hard science applications.

AI for adaptation

To get more specific, earth observation data from satellites is useful for all kinds of things, one of which is building better flood maps (what Subit is focused on at Cloud to Street). Flood modeling and, more broadly, flood intelligence is an important area of study, as flooding across the world over the past 12-18 months has brought home. 

Flooding in Davenport, Iowa (photo credit to Kelly Sikkema)

Better flood models aren’t just for insurance providers. They tie into the much broader theme of climate adaptation that we’ve also explored in this newsletter more of late. More accurate and accessible models can help with a range of things, which Subit split into four parts of the disaster ‘life cycle,’ all of which flood models and maps can help with:

  • Prepare (understand risks by region and prepare accordingly)
  • Alert (Real-time communication with potentially impacted areas)
  • Respond (Directing disaster response to most critical areas)
  • Recovery (Ongoing monitoring and guiding how or whether you rebuild)

Back to the header above – how do artificial intelligence and machine learning help with this? This question takes us back to the question of more and better data. While data is the primary key to unlocking significant improvements in areas like flood mapping, the amount of data (petabytes per satellite) is often overwhelming. That’s where artificial intelligence comes in; to source, scour, and process the amounts of data and imagery now available, millions of humans working day and night wouldn’t suffice (hence why ‘artificial’ help is needed.) 

To go deeper on how this works, how flood maps get built, and how they inform everything from where people live to how cities are built, listen to the episode!