|
Development of a real-time flood-prone area detection model using the Random Forest algorithm
|
Maedeh Mosalla tabari   , Hamid Ebadi , Zahra Alizadeh * |
| Khajeh Nasir Toosi University of Technology |
|
|
Abstract: (123 Views) |
| The identification of flood-prone areas is a crucial step in flood crisis management. In this research, a combination of the Particle Swarm Optimization (PSO) algorithm and the Random Forest (RF) algorithm is employed to improve the selection of hyperparameters in the RF method through parallel and simultaneous parameter space search. The objective is to identify locations that are susceptible to river flooding. The proposed method utilizes both ground data, such as precipitation data from ground stations, distance from roads and rivers, as well as remotely sensed data, including digital elevation models, slope, aspect, and vegetation cover index, extracted from radar remote sensing imagery such as Sentinel-1A, and optical satellite images like Landsat-8 and Sentinel-2. The results demonstrate that this algorithmic combination enhances the modeling accuracy compared to using the RF algorithm alone and improves the capability of generating real-time flood-prone maps using both ground and remote sensing data. Case studies conducted in the Ottawa-Gatineau and Gonbad-Kavus regions, where river floods occur, yielded kappa coefficients of 74/31 and 73/21, respectively, when modeling using both ground and remote sensing data. Moreover, modeling using only remote sensing data in these regions resulted in kappa coefficients of 69/1 and 67/45, respectively. These findings indicate that while utilizing both ground and remote sensing data improves modeling accuracy, modeling solely based on remote sensing data also yields acceptable performance. Additionally, modeling using high-resolution TerraSAR-X radar satellite data without the use of ground data was performed in the Ottawa-Gatineau region, resulting in a kappa coefficient of 80. This suggests that with increased spatial resolution of radar images, the need for ground data in modeling the flood-prone areas is significantly reduced. |
|
| Keywords: Ground data, Remote sensing data, Random Forest, PSO algorithm, Flood crisis managment |
|
|
|
Type of Study: Research |
Subject:
RS
Received: 2023/04/16 | Accepted: 2024/03/4 | ePublished ahead of print: 2025/02/2
|
|
|
|
|
|
|
| Send email to the article author |
|
|
