Russian Federation
Russian Federation
from 01.01.2013 until now
Russian Federation
UDC 614.8
This article examines the problem of probabilistic wildfire forecasting with a 24-hour and 7-day lead time based on ground-based meteorological observations and satellite monitoring. A feature space was constructed consisting of lagged meteorological, exponentially weighted average, calendar, and satellite features. To solve the problem, a gradient boosting model (CatBoost), as a representative of classical machine learning, and a recurrent neural network (LSTM), as a representative of deep learning architectures, were compared. Model performance was assessed using standard metrics such as ROC-AUC, PR-AUC, F1, Precision, and Recall. Statistical robustness was assessed using bootstrap analysis with 95% confidence intervals. For additional evaluation, probabilities were calibrated using isotonic regression and Platt scaling, and an analysis of seasonal quality indicators was conducted. A study of the interpretability of the models was conducted using SHAP analysis for CatBoost and the Integrated Gradient attribution method for LSTM. It was found that CatBoost provides a more balanced forecast quality and demonstrates higher sensitivity, but is inferior in the accuracy and reliability of probabilistic estimates.
wildfires, machine learning, CatBoost, LSTM, fire forecasting
1. Konstantinov A.V., Morkovina V.V. Forest fires as the most significant threat to the economic security of the forestry sector // Current areas of scientific research in the 21st century: theory and practice. - 2016. - Vol. 4, No. 2 (22). - Pp. 319-325. - EDN UOWSNE.
2. Belov A.N. Forest fires as a threat to fire safety: current state and prevention // Bulletin of Economics, Management and Law. - 2023. - Vol. 16, No. 1. - Pp. 62-70. - EDN DHJWFB.
3. Medvedev D.V., Matveev A.V. Algorithms for intelligent support of management decisions in the event of forest fire threats // Scientific and analytical journal "Bulletin of the St. Petersburg University of the State Fire Service of the Ministry of Emergency Situations of Russia". – 2025. – No. 2. – P. 35-48. – DOIhttps://doi.org/10.61260/2218-130X-2025-2-35-48. – EDN OKGHLE.
4. Medvedev D.V., Matveev A.V. Information model for supporting decision-making on responding to landscape fires // Siberian Fire and Rescue Bulletin. –2025. – No. 1(36). – P. 117-125. – DOIhttps://doi.org/10.34987/vestnik.sibpsa.2025.22.36.011. – EDN PKSBTR.
5. Prokhorenkova L. et al. CatBoost: unbiased boosting with categorical features // Advances in Neural Information Processing Systems. – 2018. – Vol. 31.
6. Matveev A.V., Bogdanova E.M. Classification of methods for forecasting emergency situations // National Security and Strategic Planning. -2018. - No. 4 (24). - P. 61-70. - EDN YTPZYL.
7. Matveev A.V., Medvedev D.V., Smirnov A.S. Possibilities of applying artificial intelligence technologies to improve the efficiency of emergency management // Informatization and Communications. - 2025. - No. 4. - P. 98-111. - DOIhttps://doi.org/10.34219/2078-8320-2025-16-4-98-111. - EDN KGTJCQ.
8. Medvedev D.V., Matveev A.V., Smirnov A.S. Application of the logistic regression model in decision-making on determining the number of forces involved in extinguishing forest fires // Fire and Explosion Safety. - 2024. - Vol. 33, No. 4. - Pp. 84-96. - DOIhttps://doi.org/10.22227/0869-7493.2024.33.04.84-96. - EDN MJLVTY.
9. Medvedev D.V., Matveev A.V. Possibilities of using an intelligent decision support system in responding to forest fire threats // Technologies of technosphere safety. - 2025. - No. 3 (109). - Pp. 74-89. - DOIhttps://doi.org/10.25257/TTS.2025.3.109.74-89. - EDN GZXXSU.
10. Rodionov A.M., Ivanov S.A. Analysis of modern models and information systems in forecasting and monitoring forest fires // Economy. Informatics. - 2023. - Vol. 50, No. 4. - P. 913-923. - DOIhttps://doi.org/10.52575/2712-746X-2023-50-4-913-923. - EDN QYFBQJ.
11. Chen T., Guestrin C. XGBoost: A Scalable Tree Boosting System // Proc. 22nd ACM SIGKDD. – 2016. – P. 785 794. – DOIhttps://doi.org/10.1145/2939672.2939785
12. Ke G. et al. LightGBM: A Highly Efficient Gradient Boosting Decision Tree // Advances in Neural Information Processing Systems. – 2017. – Vol. 30. – P. 3146 3154.
13. Raparthi M. et al. Implementation and Performance Comparison of Gradient Boosting Algorithms for Tabular Data Classification //International Conference on Deep Learning and Visual Artificial Intelligence. – Singapore : Springer Nature Singapore, 2024. – P. 461-479. DOIhttps://doi.org/10.1007/978-981-97-4533-3_36
14. Prapas I. et al. Deep Learning Methods for Daily Wildfire Danger Forecasting // arXiv preprint arXiv:2111.02736. – 2021. – DOIhttps://doi.org/10.48550/arXiv.2111.02736
15. Ku C. Y., Liu C. Y. Predictive modeling of fire incidence using deep neural networks // Fire. – 2024. – V. 7. – No 4. – P. 136. – DOIhttps://doi.org/10.3390/fire7040136
16. Liu Z. et al. NASA global satellite and model data products and services for tropical meteorology and climatology // Remote Sensing. – 2020. – V. 12. – No 17. – P. 2821. – DOIhttps://doi.org/10.3390/rs12172821
17. Adler A. I., Painsky A. Feature importance in gradient boosting trees with cross-validation feature selection // Entropy. – 2022. – V. 24. – No 5. – P. 687. – DOIhttps://doi.org/10.3390/e24050687
18. Sakai T. Evaluating evaluation metrics based on the bootstrap // Proceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval. – 2006. – P. 525-532. – DOIhttps://doi.org/10.1145/1148170.1148261
19. Sundararajan M., Taly A., Yan Q. Axiomatic attribution for deep networks // Proceedings of the 34th International Conference on Machine Learning. – 2017. – P. 3319‑3328.
20. Matveev A.V. Comparative analysis of methods of interpretability of artificial intelligence models in decision-making processes // Modern science-intensive technologies. - 2026. - No. 5. - P. 131-138. - DOIhttps://doi.org/10.17513/snt.40785. - EDN KKJORY.
21. Ancona M. et al. Towards better understanding of gradient-based attribution methods for deep neural networks // arXiv preprint arXiv:1711.06104. – 2017. DOIhttps://doi.org/10.48550/arXiv.1711.06104
22. Du M. et al. On attribution of recurrent neural network predictions via additive decomposition // The world wide web conference. – 2019. – P. 383-393. – DOIhttps://doi.org/10.1145/3308558.3313545
23. Posocco N., Bonnefoy A. Estimating expected calibration errors // International conference on artificial neural networks. – Cham : Springer International Publishing, 2021. – P. 139-150. – DOIhttps://doi.org/10.1007/978-3-030-86380-7_12
24. Raina R. et al. Impact of platt scaling on calibration in ML-based wireless resource allocation // 2025 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN). – IEEE, 2025. – P. 1-5. – DOIhttps://doi.org/10.1109/ICMLCN64995.2025.11140554
