Dynamic mapping of flood susceptibility and exposure associated with the El Niño phenomenon using machine learning and geospatial modeling in Trujillo, Peru

Floods are among the most powerful and destructive forces of nature, capable of reshaping landscapes and upending lives in a matter of hours. For communities living in low-lying areas or near river valleys, the threat is constant, but the danger is not uniform across the land. Some spots are naturally prone to holding water, while others remain dry even during heavy storms. Traditionally, scientists have mapped these risks using static pictures that show which areas are generally vulnerable based on their shape, height, and soil. However, these maps often fail to capture a crucial reality: the risk changes when the weather changes. A hillside that is safe during a light drizzle might become a torrent during a week of heavy rain, and a flat plain might only flood if the rainfall is intense enough to overwhelm the ground. Understanding where people and buildings sit in relation to these shifting risks is essential for keeping communities safe, yet most existing maps do not update themselves as the weather does.

In the arid coastal province of Trujillo, Peru, this challenge is particularly acute. The region is generally dry, but it is periodically struck by intense rainfall events linked to the El Niño phenomenon, a climate pattern that warms the Pacific Ocean and alters weather across the globe. These rare storms can turn dry ravines into rushing rivers, flooding urban neighborhoods and damaging infrastructure. Researchers from the National University of Trujillo and the National University of Altiplano set out to solve the problem of static mapping by creating a new way to track flood risk that changes with the weather. They wanted to build a system that could not only identify which parts of the city are naturally vulnerable but also update that risk in real time as new rain falls, while simultaneously showing exactly where the people and buildings are located within those danger zones.

To achieve this, the team looked back at three major flood events that hit Trujillo in 2017, 2023, and 2025. Instead of relying on traditional cameras that cannot see through clouds, they used radar images from satellites that can scan the ground day or night, regardless of the weather. By comparing images taken before and after each storm, they were able to trace the exact footprints of the water as it spread across the land. They combined this historical data with detailed information about the terrain, such as the slope of the land, the height of the ground, and the location of drainage channels, as well as geological data about the rocks and soil. They also fed in data about the rainfall itself, breaking down the storms into different durations to see how short, intense bursts of rain compared to longer, steady downpours.

The researchers then used a type of computer learning called machine learning to find the patterns connecting the landscape and the weather to the actual floods. They tested several different computer models to see which one could best predict where water would go. One model, known as Random Forest, proved to be the most accurate. It learned to recognize that certain combinations of factors—such as a specific direction the land faces, the steepness of a slope, and the amount of rain—were strong indicators of flooding. This model did not just memorize the past events; it learned the underlying rules of how water moves across the Trujillo landscape. The analysis revealed that the direction the land faces, the amount of rain, the elevation, and the slope were the most important factors in determining which areas would flood.

What makes this study distinct is its ability to move beyond a single, unchanging map. Once the computer learned the rules from the past storms, the researchers showed that they could update the risk map simply by plugging in new rainfall data, without having to teach the computer all over again. This means that if a new storm hits, the system can instantly generate a fresh map showing where the risk is highest right now. They tested this by creating a scenario for September 2026. In this updated view, the risk was not spread evenly; instead, a large portion of the territory showed a moderate level of risk, covering more than half of the area. However, the areas with the highest risk were smaller but still significant, and they overlapped with specific parts of the city where many people live.

The study also brought a human dimension to the numbers by layering population data and building density onto the flood maps. This allowed the team to see not just where the water might go, but who would be affected. In the 2017 event, for example, more than half of the exposed population was located in areas of high or very high risk. By contrast, in the 2025 event, the distribution shifted, with a larger share of the population sitting in areas of moderate risk. When looking at the dynamic update for September 2026, the researchers found that while the moderate risk zone covered the majority of the land and the vast majority of the population, the smaller zones of high risk still contained nearly 18 percent of the people and nearly 20 percent of the built-up surface. This highlights that even a small area of high risk can be critical if it is packed with homes and businesses.

The researchers concluded that their approach offers a powerful tool for decision-makers. By combining satellite radar, machine learning, and real-time weather data, they created a system that can track flood susceptibility as it evolves. This dynamic view helps identify priority areas for prevention and response, moving away from the idea that flood risk is a fixed feature of the landscape. While the study relies on computer simulations and historical data rather than real-time hydraulic measurements, it provides a robust method for understanding how the interaction between the terrain and the weather shapes the danger to communities. For a city like Trujillo, where the threat of sudden, intense rain is a recurring reality, having a map that can update itself offers a clearer path toward protecting lives and property.