Geographia Technica, Vol 18, Issue 1, 2023, pp. 1-18

DEVELOPING A FLOOD FORECASTING SYSTEM WITH MACHINE LEARNING AND APPLYING TO GEOGRAPHIC INFORMATION SYSTEM

Jirayu PUNGCHING , Sitang PILAILAR

DOI: 10.21163/GT_2023.181.01

ABSTRACT: Floods are natural disasters that can damage lives, property, and the economy. Therefore, it is necessary to have a reliable and accurate flood forecasting system to provide early warning in time. Although several Mathematical models have been developed and used to forecast floods continuously for decades, most require up-to-date and specific physical data, including a high experience user, to provide and interpret the result. It is an obstacle for use in remote areas with incomplete information and a lack of specialists. This study, therefore, developed a real-time flood forecasting system with Machine Learning by applying a 2-variable sliding window technique to restructure the data, which can solve the problem of data limitation. Thung Song District Nakhon Si Thammarat Province was selected to test this newly developed model. By importing the water level data of two water level observed stations, SWR025 at the upstream and NKO001 at Thung Song Municipality, into five machine learning algorithms (Linear Regression, Support Vector Machine, K-Nearest Neighbor, Decision Tree, and Random Forest) for forecasting the water level every 30 minutes for the next 5 hours. Their performance was compared by the MSE, MAE, and R2, which ranged from 0.006-0.013, 0.044-0.063, and 0.518-0.750, respectively. The Random Forest was the most efficient algorithm for the 3-hour forecast with an efficiency value of MSE 0.006, MAE 0.044, and R2 0.75. The developed ML flood forecasting model was validated by the flood data in November 2021 and showed good agreement. Then, the extent of the inundation area was evaluated by the mathematical model. Next, the water depth and surface elevation were transformed and applied to GIS. Finally, the flood risk areas on Google Maps under that specific rainfall are promptly notified to the people three hours before the flood occurs.


Keywords: Flood Forecasting, Flood Maps, Machine Learning, Linear Regression, Support Vector Machine, K-Nearest Neighbor, Decision Tree, Random Forest, Thung Song Municipality

Full article here