New England is a complicated place, especially when it comes to flooding.
Samuel Muñoz, an associate professor of marine and environmental science at Northeastern University, said the region's complex network of small, interconnected rivers, diverse topography, and atmospheric movement in the Atlantic Ocean all make it extremely difficult to model mathematically.
Muñoz and new research student Lindsey Lawrence used machine learning to build “self-organizing maps” to reveal how atmospheric and land conditions interact and identify four patterns that lead to New England flooding. This breakthrough in climate modeling is expected to help predict floods before they occur, especially in a warming climate.
complexity
Muñoz said previous flood research focused on large rivers like the Mississippi River. “In other systems we work with, the mechanisms that cause flooding are not simple, but they are simpler than in New England,” he says. New England has a more complex topography, more rivers (albeit smaller), and more rain.
Precipitation can occur in New England in a variety of ways, from snow to the occasional hurricane, he says.
Lawrence points out that rain itself is very difficult to model, which is one reason why weather apps aren't as accurate as you might hope. Her “nerdy answer” to why precipitation is so difficult is that “the actual microphysics of clouds is really, really incredibly difficult.”
Munoz says it's also a matter of scale. Current weather models divide the world into a grid with cells 100 kilometers on a side, about 62 miles. The size of these grids means the models are good at predicting things like pressure and temperature, but the actual mechanics of condensation and the resulting precipitation occur on much smaller scales.
Machine learning helped overcome this problem.
map
By aggregating data since 1979, Lawrence developed so-called self-organizing maps that condense the data into modelable groups and clusters.
Self-organizing maps and the machine learning protocols needed to create them have been used since the 1980s, Lawrence points out. They are called maps, she continued, because they reduce data while preserving “so-called topology, data shape, data characteristics, and grouping features.”
Lawrence and Muñoz were interested in grouping floods together, which are similar events. Once Lawrence's map was completed, he was able to identify four specific patterns in surface conditions such as pressure, temperature, and soil moisture that lead to flooding.
The map also shows seasonality and when it is most likely to occur, Lawrence said. Three of the maps tend toward late winter or early spring, and one features summer features. And because each map has notable patterns in barometric phenomena and soil moisture, meteorologists can use them to issue early warnings and save lives, she says.
what happens next
Muñoz points out that they were also motivated to think about climate change and how it will affect future river flows, floods and droughts in the region.
This has historically been difficult as Earth models struggle with precipitation. But, he continued, “we can look at how those conditions change over time as the composition of the atmosphere changes” through global warming, linking “flooding events to patterns of pressure and temperature” and soil moisture.
reference: Lawrence L, Armando Z, Muñoz SE. Connecting large-scale atmospheric and surface patterns to New England river peak flow events. Geophys Res Lett. 2025;52(19):e2025GL116899. doi: 10.1029/2025GL116899
This article has been reprinted from the following material: Note: Materials may be edited for length and content. Please contact the citation source for details. You can access our press release publishing policy here.
