|
High-Precision Wildfire Risk Mapping Using an Ensemble Machine Learning Model: Performance Assessment of Sentinel‑2 and Landsat‑8 Data
|
Mohaddeseh Mesvari  , Reza Shah-Hosseini * |
| Faculty of Surveying and Spatial Information Engineering, University of Tehran & - |
|
|
Abstract: (73 Views) |
Forest fires are among the most destructive natural disasters, leading to irreparable ecological and economic consequences. Developing accurate fire risk maps is a vital tool for the management and prevention of this crisis. The present study was conducted with the aim of developing an integrated framework based on machine learning algorithms and evaluating the impact of satellite data spatial resolution on the accuracy of wildfire risk modeling. To this end, spectral and thermal data from two sensors, Landsat-8 (with 30m resolution) and Sentinel-2 (with 10m resolution), were utilized along with topographic and climatic variables. In this paper, five different base machine learning models, which are among the well-known machine learning models, were trained, and finally, the results from these models were trained using a meta-ensemble model to predict fire-prone areas. This approach allowed for the utilization of all the diverse capabilities inherent in machine learning models, better modeling of complex non-linear patterns in the input data, and the creation of a more stable model whenever possible. Exploratory analyses showed that NDVI and NBR indices, along with elevation and land surface temperature variables, play the most significant role in distinguishing fire-affected areas. Quantitative evaluation results demonstrated that the meta-ensemble model implemented on Sentinel-2 data achieved a high performance among all base models, reaching an accuracy of 94.52%, an F1-Score of 94.76–94.87%, and an IoU index of 80.12%. This indicates that instead of relying on a single machine learning model alone, the combination of different capabilities from various models can be used to improve model accuracy and stability. Furthermore, comparing the results of the two sensors showed that the enhancement of spatial resolution in Sentinel-2 led to improved accuracy in wildfire risk estimation, and its output maps possess far more precise clarity and delineation due to the reduction of spatial averaging effects. The results of this study indicate that integrating meta-ensemble models with high-resolution satellite data can be utilized as an efficient system in the process of proactive natural resource management.
|
|
| Keywords: Wildfire Risk Modeling, Machine Learning, Meta-Ensemble Model, Sentinel-2, Landsat-8, Spatial Resolution, Remote Sensing. |
|
|
|
Type of Study: Research |
Subject:
Special
Received: 2026/03/26 | Accepted: 2026/05/6
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
