Researchers created flood vulnerability maps for Tokmok, Osh and Jalal-Abad. They used radar satellite images from Sentinel-1, which can detect flooded areas even under cloudy conditions, and processed the data using machine learning algorithms.

The largest share of areas classified as having a very high flood susceptibility was found in Tokmok, accounting for an average of 37.8 percent of the city’s territory.

In Osh, this figure is 12.01 percent, while in Jalal-Abad it is 8.44 percent. Thus, the share of the most vulnerable areas in Tokmok is more than three times higher than in Osh and about four and a half times higher than in Jalal-Abad.

To create the maps, scientists used three machine learning models — Random Forest, XGBoost and AdaBoost. The models were trained to distinguish areas where flooding had previously been recorded from areas without such events, and the results were then compared with natural and territorial factors affecting water accumulation and movement.

All three models demonstrated prediction accuracy and performance indicators above 84 percent. The researchers also applied the SHAP method, which helps identify what factors had the greatest influence on the final assessment of each area.

The authors noted that such maps can be used for urban planning, the placement of roads and engineering infrastructure, modernization of drainage systems and preparation of civil protection measures.

At the same time, the study assesses the susceptibility of territories to flooding, rather than the full risk to the population.

A comprehensive assessment of potential damage would also require taking into account population density, the condition of buildings, infrastructure value and the readiness of emergency services.

The study involved Eldiyar Duulatov, a researcher at the Institute of Geology of the National Academy of Sciences of Kyrgyzstan, as well as scientists from Poland and India.