All Issue

2025 Vol.35, Issue 4 Preview Page

Research Article

Wildfire Susceptibility Assessment Using AutoML and Geo-Similarity Sampling

AutoML 및 Geo-Similarity 샘플링을 활용한 산불 취약성 평가

Kounghoon Nam1
,
Youngkyu Lee2
,
Sungsu Lee3
,
Gyo-Cheol Jeong4
,
Jongtae Kim5*
남 경훈1
,
이 영규2
,
이 승수3
,
정 교철4
,
김 종태5*
1Researcher, Fire Insurers Laboratories of Korea
2Principal Researcher, Fire Insurers Laboratories of Korea
3Professor, Department of Civil Engineering, Chungbuk National University
4Emeritus Professor, Department of Earth and Environmental Sciences, Gyeongkuk National University
5CEO, Nature and Tech Inc.
1방재시험연구원 전문위원
2방재시험연구원 책임연구원
3충북대학교 토목공학부 교수
4국립경국대학교 지구환경과학과 명예교수
5㈜자연과기술 대표이사
*Corresponding Author
31 December 2025. pp. 717-729
PDF XML Citation
Abstract

We developed a wildfire susceptibility prediction model using GeoAI-based automated machine learning (AutoML) and geo-similarity sampling, and evaluated its accuracy using the spatial classification accuracy index (SCAI). Focusing on the Daegu and Gyeongbuk regions, three AutoML-based models (GaussianNB, Random Forest, and XGBoost) were trained using the Tree-based Pipeline Optimization Tool (TPOT). The Random Forest model achieved the best performance with an area under the curve (AUC) of 0.998 and an accuracy (ACC) of 0.981, while XGBoost exhibited the highest spatial concentration in high-risk areas with an SCAI of 0.46. In all models, SCAI decreased as susceptibility levels increased, indicating improved spatial consistency between predicted values and actual wildfire occurrences. The integration of Geo-Similarity Sampling and AutoML effectively mitigated data imbalance, enhanced classification accuracy, and improved spatial reliability, demonstrating its potential applicability for wildfire risk mapping and disaster prevention decision-making.

Keywords
AutoML
Geo-Similarity Sampling
wildfire susceptibility mapping
SCAI
GeoAI
TPOT

본 연구는 GeoAI 기반 AutoML (automated machine learning)과 Geo-Similarity 샘플링을 활용하여 산불 취약성 예측 모델을 구축하고, 공간분류정확도지수(spatial classification accuracy index, SCAI)를 통해 공간적 분류 성능을 평가하였다. 대구·경북 지역을 대상으로 TPOT (Tree-based Pipeline Optimization Tool)을 이용해 GaussianNB, Random Forest, XGBoost 등 세 가지 AutoML 기반 모델을 학습하였다. 그 결과 Random Forest 모델이 AUC (area under the curve) 0.998, 정확도(accuracy, ACC) 0.981로 가장 우수한 성능을 보였으며, XGBoost 모델은 SCAI 0.46으로 고위험 지역에서 가장 높은 공간 집중도를 나타냈다. 모든 모델에서 취약성 등급이 높아질수록 SCAI 값이 감소하는 경향을 보였는데, 이는 예측값과 실제 산불 발생 지점 간의 공간적 일치도가 향상되었음을 의미한다. Geo-Similarity 샘플링과 AutoML의 결합은 데이터 불균형 문제를 효과적으로 완화하고 분류 정확도와 공간 신뢰성을 향상시켜, 산불 위험지도 제작 및 재난 예방 의사결정에 높은 활용 가능성을 보여주었다.

키워드
AutoML
Geo-Similarity 샘플링
산불 취약성 맵핑
SCAI
GeoAI
TPOT
References
1

Geofabrik GmbH, 2025, OpenStreetMap Data Extracts, Retrieved from https://download.geofabrik.de.

2

HydroSHEDS, 2025, Seamless hydrographic data for global and regional applications, Retrieved from https://www.hydrosheds.org.

3

Jeong, J.C., Son, J.H., 2025, Analysis of Andong wildfire extent using Random Forest and forest type map, Journal of Cadastre and Land Information, 55(1), 157-171 (in Korean with English abstract).

10.22640/LXSIRI.2025.55.1.157
4

Lee, S.D., Park, G.S., Oh, C.H., Cho, B.G., Yu, B.H., 2024, Determination of fire severity and deduction of influence factors through Landsat-8 satellite image analysis—A case study of Gangneung and Donghae forest fires, Korean Journal of Environment and Ecology, 38(3), 277-292 (in Korean with English abstract).

10.13047/KJEE.2024.38.3.277
5

Nam, K., Kim, J., Lee, C.J., Jeong, G.C., 2024, GeoAI-based forest fire susceptibility assessment with integration of forest and soil digital map data, The Journal of Engineering Geology, 34(2), 147-163 (in Korean with English abstract).

10.9720/kseg.2024.1.107
6

Nam, K., Lee, Y., Lee, S., Kim, S., Zhang, S., 2025, Explainable artificial intelligence (XAI) for flood susceptibility assessment in Seoul: Leveraging evolutionary and Bayesian AutoML optimization, Remote Sensing, 17(13), 2244.

10.3390/rs17132244
7

Rabby, Y.W., Li, Y., Hilafu, H., 2023, An objective absence data sampling method for landslide susceptibility mapping, Scientific Reports, 13, 1740.

10.1038/s41598-023-28991-536720965PMC9889336
8

Sim, S.M., Kim, W.H., Lee, J.S., Kang, Y.J., Im, J.H., Kwon, C.G., Kim, S.Y., 2020, Wildfire severity mapping using Sentinel satellite data based on machine learning approaches, Korean Journal of Remote Sensing, 36(5-3), 1109-1123 (in Korean with English abstract).

10.7780/kjrs.2020.36.5.3.9
9

Xu, Q., Li, W., Liu, J., Wang, X., 2023, A geographical similarity-based sampling method of non-fire point data for spatial prediction of forest fires, Forest Ecosystems, 10, 100104.

10.1016/j.fecs.2023.100104
Information
  • Publisher :Korean Society of Engineering Geology
  • Publisher(Ko) :대한지질공학회
  • Journal Title :The Journal of Engineering Geology
  • Journal Title(Ko) :지질공학
  • Volume : 35
  • No :4
  • Pages :717-729
  • Received Date : 2025-10-30
  • Revised Date : 2025-11-18
  • Accepted Date : 2025-11-27
  • DOI :https://doi.org/10.9720/kseg.2025.4.717
Journal Informaiton The Journal of Engineering Geology The Journal of Engineering Geology
Online Submission
  • NRF
  • KOFST
  • KISTI Current Status
  • KISTI Cited-by
  • crosscheck
  • orcid
  • open access
  • ccl
Journal Informaiton Journal Informaiton - close