Flood Susceptibility Mapping in Yola Metropolis, Nigeria, Using Random Forest, Terrain-Derived Predictors, and Nested Spatial Cross-Validation
This study develops a Random Forest-based flood susceptibility map for Yola Metropolis, Nigeria, using terrain and land-use predictors with nested spatial cross-validation to identify elevation as the dominant factor and classify 92.24% of the area as low-to-moderate risk, providing a reproducible screening tool for disaster planning in data-constrained urban settings.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Flooding is a persistent threat in cities that are growing quickly, especially in regions where detailed weather records and drainage maps are scarce. When heavy rains fall or rivers swell, water seeks the lowest ground, but in rapidly expanding urban areas, new buildings and paved roads often block the natural paths water takes to drain away. To protect people and property, planners need to know which parts of a city are most likely to flood. This is where flood susceptibility mapping comes in. Instead of trying to predict exactly when a flood will happen or how deep the water will get, these maps identify areas that have the physical characteristics—such as low elevation or specific types of land cover—that make them prone to flooding. By using satellite images and computer models, scientists can create these maps even in places where ground-level data is hard to collect, providing a crucial first step for disaster planning.
In the city of Yola, located in northeastern Nigeria, researchers recently applied this approach to create a detailed map of flood risk. Yola sits in a low-lying area near the Benue River, making it vulnerable to rising waters, yet it lacks the extensive local data usually required for precise engineering studies. A team of scientists from the University of Pretoria and other institutions set out to build a model that could identify flood-prone zones using only the information available from satellite imagery and digital maps. Their goal was not just to produce a map, but to test a rigorous method for ensuring the map was accurate and reliable, even with limited data. They focused on three main factors that influence flooding: how high the land is, the slope of the ground, and what covers the surface, such as trees, crops, or buildings.
To build their model, the researchers used a powerful computer learning technique called Random Forest. This method works by creating thousands of decision trees that look at different combinations of data to spot patterns. The team fed the computer data from a digital elevation model, which acts like a 3D map of the ground's height, along with a land-use map from 2022 that showed what covered the ground in that year. They also included a record of 832 specific locations where a flood had been observed in 2022, alongside locations that remained dry. This record served as the "answer key" for the computer to learn from. The researchers were careful to avoid a common mistake in mapping: they ensured that the computer did not use data points that were right next to the ones it had already learned from for testing. Instead, they divided the city into large blocks and tested the model on entire blocks of land it had never seen before, a method that gives a much truer picture of how well the model would work in the real world.
The results showed that the most important factor in predicting flooding was simply the height of the land. The computer model determined that knowing the elevation was far more critical than knowing the exact slope of the ground or how far a location was from a river. While adding the slope data improved the model slightly, adding the distance to streams actually made the model slightly worse at correctly identifying flooded areas. The best-performing model, which combined elevation, slope, and land cover, correctly identified flooded and non-flooded areas with a high degree of accuracy. It successfully classified 91.7 percent of the actual flood locations as high-risk and 93.1 percent of the dry locations as safe. This level of precision suggests that in this specific landscape, the shape of the land is the dominant force controlling where water accumulates.
Using these findings, the team created a final map that divides Yola into three categories: low, moderate, and high susceptibility. They set the boundaries for these categories based on how the model performed during its testing phase, rather than using arbitrary lines. The map revealed that the vast majority of the city, about 82.8 percent of the area where the model could make a prediction, falls into the low-susceptibility category. This means most of the city is on ground high enough to stay dry under typical conditions. About 9.4 percent of the area is classified as moderate, and a smaller portion, roughly 7.8 percent, is marked as high susceptibility. These high-risk zones are concentrated in the lowest-lying areas, which aligns with the physical reality of how water flows. However, the researchers also noted a small area, about 5.26 square kilometers, that they could not map. This was because the land cover in that specific spot was a type of vegetation that the computer had never seen in its training data, so it wisely refused to guess rather than making up an answer.
The study concludes that this map is a powerful tool for planning, but it is not a complete solution. It serves as a screening tool to help officials decide where to look more closely. If a developer wants to build in a high-susceptibility zone, this map signals that a detailed engineering study is needed to check drainage and water depth. The researchers were careful to state that their map does not predict how deep the water will be or how often floods will occur; it only shows which places are most likely to get wet based on their physical setting. By proving that a simple combination of elevation and land cover can produce such a clear picture, the study offers a reliable workflow for other cities in data-poor regions. It demonstrates that even without complex local measurements, a careful analysis of satellite data can provide the essential guidance needed to protect communities from the rising waters of a changing climate.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.