pith. sign in

arxiv: 2505.17556 · v1 · submitted 2025-05-23 · 💻 cs.LG · cs.CV

Wildfire spread forecasting with Deep Learning

Nikolaos Anastasiou , Spyros Kondylatos , Ioannis Papoutsis This is my paper

Pith reviewed 2026-05-19 14:02 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords wildfire spread forecastingdeep learningspatio-temporal modelingburned area predictionMediterranean firesremote sensingtemporal ablationignition-day baseline
0
0 comments X

The pith

A deep learning model predicts final wildfire burned areas more accurately when given four days of pre-ignition and five days of post-ignition observations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors built a deep learning system that forecasts the final extent of a wildfire using satellite imagery, weather records, vegetation maps, land cover, human factors, topography, and thermal data collected around the moment of ignition. They ran an ablation study on a large Mediterranean dataset covering 2006 through 2022 to measure how much extra accuracy comes from adding days before and after ignition. The strongest model, which uses a nine-day temporal window centered on ignition, raises both the F1 score and the Intersection over Union by nearly five percentage points over a baseline that sees only the ignition day. These improvements matter for emergency services that must decide where to send crews and equipment in the first hours of a fire. The team has released the full dataset and trained models so others can build on the work.

Core claim

The paper shows that a spatio-temporal deep learning model trained on remote-sensing, meteorological, vegetation, land-cover, anthropogenic, topographic, and thermal-anomaly inputs achieves substantially higher accuracy in forecasting final burned-area extent when the input window is expanded from a single ignition-day snapshot to four days before ignition through five days after ignition, delivering an approximately 5% gain in both F1 score and Intersection over Union on a held-out test set drawn from the Mediterranean region.

What carries the argument

A deep learning model that ingests a variable-length spatio-temporal stack of remote-sensing and environmental layers centered on ignition time and outputs a predicted burned-area mask.

If this is right

  • Emergency planners can allocate firefighting resources with greater confidence in the first days of a new ignition.
  • Forecasts that incorporate post-ignition observations can be updated daily to refine containment strategies.
  • The same temporal-window approach can be applied to other regions once local data become available.
  • Public release of the dataset and code lowers the barrier for hybrid physics-plus-machine-learning wildfire models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Operational systems could ingest live satellite feeds to extend the post-ignition window in real time.
  • Testing the same architecture on climate-projection data might reveal how forecast skill changes under altered vegetation and weather regimes.
  • The five-percent gain may compound when the model is combined with physics-based spread simulators that supply additional constraints.
  • Similar temporal-context ablations could improve forecasts for other rapidly evolving hazards such as floods or oil spills.

Load-bearing premise

The held-out test fires and labeling decisions in the Mediterranean dataset will continue to represent the conditions the model will encounter in future operational use.

What would settle it

Running the released model on fires that occurred after 2022 in the same region or on fires in a different continent and measuring whether the five-percent margin over the ignition-day baseline still appears.

Figures

Figures reproduced from arXiv: 2505.17556 by Ioannis Papoutsis, Nikolaos Anastasiou, Spyros Kondylatos.

Figure 1
Figure 1. Figure 1: FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: presents the cumulative monthly distribution of fire events aggregated across all years. As expected in a Mediterranean climate, the majority of fires occur during the summer months, with July and August exhibiting the highest frequency, followed by March and September, though at notably fewer occurrences. To assess the variability in wildfire sizes, fire events were grouped using k-means clustering [45] b… view at source ↗
Figure 2
Figure 2. Figure 2: displays the annual distribution of fire events from 2006 to 2022. An upward trend is observed in recent years, particularly after 2017 (excluding 2018), where the number of recorded events exceeds 650 per year. In contrast, earlier years contain significantly fewer samples. FIGURE 2. Yearly distribution of fire events in the dataset [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIGURE 5 [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIGURE 6 [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIGURE 9 [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
read the original abstract

Accurate prediction of wildfire spread is crucial for effective risk management, emergency response, and strategic resource allocation. In this study, we present a deep learning (DL)-based framework for forecasting the final extent of burned areas, using data available at the time of ignition. We leverage a spatio-temporal dataset that covers the Mediterranean region from 2006 to 2022, incorporating remote sensing data, meteorological observations, vegetation maps, land cover classifications, anthropogenic factors, topography data, and thermal anomalies. To evaluate the influence of temporal context, we conduct an ablation study examining how the inclusion of pre- and post-ignition data affects model performance, benchmarking the temporal-aware DL models against a baseline trained exclusively on ignition-day inputs. Our results indicate that multi-day observational data substantially improve predictive accuracy. Particularly, the best-performing model, incorporating a temporal window of four days before to five days after ignition, improves both the F1 score and the Intersection over Union by almost 5% in comparison to the baseline on the test dataset. We publicly release our dataset and models to enhance research into data-driven approaches for wildfire modeling and response.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents a deep learning framework for forecasting the final extent of burned areas in wildfires, using a spatio-temporal Mediterranean dataset (2006-2022) that combines remote sensing, meteorological, vegetation, land cover, anthropogenic, topographic, and thermal data. It emphasizes use of inputs available at ignition time and reports an ablation study on temporal windows, where a model incorporating four days before to five days after ignition improves F1 score and Intersection over Union by nearly 5% over an ignition-day baseline on held-out test data. The dataset and trained models are released publicly.

Significance. If the performance gains can be demonstrated using only pre-ignition and ignition-time data, the work would provide a useful empirical demonstration of the value of multi-day context for data-driven wildfire spread prediction and support further research through the public data release.

major comments (2)
  1. [Abstract] Abstract: the headline empirical result (best model with four days before to five days after ignition yielding ~5% F1/IoU gain) incorporates five days of post-ignition observations. This directly conflicts with the stated forecasting setup that uses only data available at ignition time; the ablation therefore does not establish that the reported improvement is achievable under operational constraints.
  2. [Ablation study] Ablation study design: because the temporal window is varied explicitly to include post-ignition periods, the 5% gain cannot be taken as evidence for the central claim without additional results that restrict inputs to pre-ignition and ignition-day data only.
minor comments (2)
  1. [Abstract] Provide explicit definitions of the exact input features and preprocessing steps for each temporal-window configuration in the ablation.
  2. Report error bars, exact train/validation/test splits, and model architecture details to allow assessment of whether the 5% gain is robust.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We acknowledge that the current presentation of the abstract and ablation study creates an inconsistency with the stated goal of forecasting using only data available at ignition time. We will revise the manuscript to resolve this and strengthen the alignment between the empirical results and the operational forecasting setup.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline empirical result (best model with four days before to five days after ignition yielding ~5% F1/IoU gain) incorporates five days of post-ignition observations. This directly conflicts with the stated forecasting setup that uses only data available at ignition time; the ablation therefore does not establish that the reported improvement is achievable under operational constraints.

    Authors: We agree that the abstract as currently worded highlights a result relying on post-ignition data while the paper's central framing is forecasting with ignition-time inputs. This creates a mismatch with operational constraints. We will revise the abstract to lead with performance using only pre-ignition and ignition-day data as the primary forecasting result. The ablation will be reframed to show the incremental value of additional temporal context, with explicit discussion of which configurations are feasible at ignition time versus those requiring post-ignition observations. revision: yes

  2. Referee: [Ablation study] Ablation study design: because the temporal window is varied explicitly to include post-ignition periods, the 5% gain cannot be taken as evidence for the central claim without additional results that restrict inputs to pre-ignition and ignition-day data only.

    Authors: The referee is correct that the reported ~5% gain corresponds to a window that includes post-ignition data and therefore cannot directly substantiate the ignition-time forecasting claim. We will add a dedicated comparison in the ablation study (or as a new table/figure) that restricts all inputs to pre-ignition and ignition-day data only. This will provide the missing evidence for the central claim while retaining the existing results to illustrate the potential benefit of extended temporal windows when such data become available. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation or claims.

full rationale

The paper describes a standard supervised deep learning pipeline: models are trained on a spatio-temporal Mediterranean dataset (2006-2022) and evaluated on a held-out test set using F1 and IoU. The ablation varies the temporal window (including post-ignition observations) and reports an empirical ~5% gain for the widest window versus an ignition-day baseline. No derivation reduces by construction to its inputs; there are no self-definitional equations, no fitted parameters renamed as predictions, and no load-bearing self-citations. The reported improvement is a direct empirical measurement on external test data and does not collapse to a tautology or re-labeling of training inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard supervised learning assumptions plus the representativeness of the 2006-2022 Mediterranean fires; no new physical entities or ad-hoc constants are introduced beyond typical neural-network hyperparameters.

free parameters (1)
  • neural network weights and hyperparameters
    Learned from training data; the ablation result depends on these fitted values.
axioms (1)
  • domain assumption Training and test fires are drawn from the same distribution
    Required for the reported test-set improvement to indicate future performance.

pith-pipeline@v0.9.0 · 5727 in / 1332 out tokens · 79253 ms · 2026-05-19T14:02:06.632896+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 2 internal anchors

  1. [1]

    J. G. Pausas and J. E. Keeley, ‘‘Wildfires as an ecosystem service,’’ Frontiers in Ecology and the Environment, vol. 17, no. 5, pp. 289–295,

    work page

  2. [2]

    Available: https://esajournals.onlinelibrary.wiley.com/doi/ abs/10.1002/fee.2044

    [Online]. Available: https://esajournals.onlinelibrary.wiley.com/doi/ abs/10.1002/fee.2044

    work page doi:10.1002/fee.2044 2044

  3. [3]

    H. A. Kramer, M. H. Mockrin, P . M. Alexandre, and V . C. Radeloff, ‘‘High wildfire damage in interface communities in california,’’International Journal of Wildland Fire, vol. 28, no. 9, pp. 641–650, 2019. [Online]. Available: https://doi.org/10.1071/WF18108

    work page doi:10.1071/wf18108 2019

  4. [4]

    S. E. Finlay, A. Moffat, R. Gazzard, D. Baker, and V . Murray, ‘‘Health impacts of wildfires,’’PLoS Curr, vol. 4, p. e4f959951cce2c, 2012, pubMed-not-MEDLINE, accessed 2024-03-25. [Online]. Available: https://doi.org/10.1371/4f959951cce2c

    work page doi:10.1371/4f959951cce2c 2012

  5. [5]

    Bautista Vicente, N

    F. Bautista Vicente, N. Carbajal, and L. F. Pineda Martínez, ‘‘Estimation of total yearly co2 emissions by wildfires in mexico during the period 1999–2010,’’Advances in Meteorology, vol. 2014, no. 1, p. 958457, 2014. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1155/2014/ 958457

    work page doi:10.1155/2014/ 1999

  6. [6]

    M. W. Jones, A. Smith, R. Betts, J. G. Canadell, I. C. Prentice, and C. Le Quéré, ‘‘Climate change increases the risk of wildfires,’’ 2020

    work page 2020

  7. [7]

    M. M. V alero, O. Rios, C. Mata, E. Pastor, and E. Planas, ‘‘An integrated approach for tactical monitoring and data-driven spread forecasting of wildfires,’’Fire Safety Journal, vol. 91, pp. 835–844, 2017

    work page 2017

  8. [8]

    Mandel, S

    J. Mandel, S. Amram, J. Beezley, G. Kelman, A. Kochanski, V . Kon- dratenko, B. Lynn, B. Regev, and M. V ejmelka, ‘‘Recent advances and applications of wrf–sfire,’’Natural Hazards and Earth System Sciences, vol. 14, no. 10, pp. 2829–2845, 2014

    work page 2014

  9. [9]

    Marjani, M

    M. Marjani, M. Mahdianpari, and F. Mohammadimanesh, ‘‘Cnn-bilstm: A novel deep learning model for near-real-time daily wildfire spread prediction,’’Remote Sensing, vol. 16, no. 8, p. 1467, 2024

    work page 2024

  10. [10]

    Alizadeh, M

    N. Alizadeh, M. Mahdianpari, E. Hemmati, and M. Marjani, ‘‘Fusion- firenet: A cnn-lstm model for short-term wildfire hotspot prediction uti- lizing spatio-temporal datasets,’’Available at SSRN 4963223, 2024

    work page 2024

  11. [11]

    Richardson, A

    D. Richardson, A. S. Black, D. Irving, R. J. Matear, D. P . Monselesan, J. S. Risbey, D. T. Squire, and C. R. Tozer, ‘‘Global increase in wildfire potential from compound fire weather and drought,’’NPJ climate and atmospheric science, vol. 5, no. 1, p. 23, 2022

    work page 2022

  12. [12]

    Shadrin, S

    D. Shadrin, S. Illarionova, F. Gubanov, K. Evteeva, M. Mironenko, I. Levchunets, R. Belousov, and E. Burnaev, ‘‘Wildfire spreading prediction using multimodal data and deep neural network approach,’’Scientific Reports, vol. 14, no. 1, p. 2606, 2024

    work page 2024

  13. [13]

    Afghah, A

    F. Afghah, A. Razi, J. Chakareski, and J. Ashdown, ‘‘Wildfire monitoring in remote areas using autonomous unmanned aerial vehicles,’’ inIEEE INFO- COM 2019-IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS). IEEE, 2019, pp. 835–840

    work page 2019

  14. [14]

    F. Huot, R. L. Hu, N. Goyal, T. Sankar, M. Ihme, and Y .-F. Chen, ‘‘Next day wildfire spread: A machine learning dataset to predict wildfire spreading from remote-sensing data,’’IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–13, 2022

    work page 2022

  15. [15]

    J. G. Pausas and J. E. Keeley, ‘‘Wildfires and global change,’’Frontiers in Ecology and the Environment, vol. 19, no. 7, pp. 387–395, 2021

    work page 2021

  16. [16]

    K. G. Hirsch,Canadian forest fire behavior prediction (FBP) system: user’s guide., 1996, no. 7

    work page 1996

  17. [17]

    M. P . Kale, S. S. Meher, M. Chavan, V . Kumar, M. A. Sultan, P . Dongre, K. Narkhede, J. Mhatre, N. Sharma, B. Luitelet al., ‘‘Operational forest- fire spread forecasting using the wrf-sfire model,’’Remote Sensing, vol. 16, no. 13, p. 2480, 2024

    work page 2024

  18. [18]

    Reichstein, G

    M. Reichstein, G. Camps-V alls, B. Stevens, M. Jung, J. Denzler, and N. Carvalhais, ‘‘Deep learning and process understanding for data-driven Earth system science,’’Nature, vol. 566, no. 7743, pp. 195–204, 2019, publisher: Nature Publishing Group

    work page 2019

  19. [19]

    Radke, A

    D. Radke, A. Hessler, and D. Ellsworth, ‘‘Firecast: Leveraging deep learn- ing to predict wildfire spread.’’ inIJCAI, 2019, pp. 4575–4581

    work page 2019

  20. [20]

    Kondylatos, I

    S. Kondylatos, I. Prapas, G. Camps-V alls, and I. Papoutsis, ‘‘Mesogeos: A multi-purpose dataset for data-driven wildfire modeling in the mediter- ranean,’’Advances in Neural Information Processing Systems, vol. 36, pp. 50 661–50 676, 2023

    work page 2023

  21. [21]

    P . Jain, S. C. P . Coogan, S. G. Subramanian, M. Crowley, S. Taylor, and M. D. Flannigan, ‘‘A review of machine learning applications in wildfire science and management,’’Environmental Reviews, vol. 28, no. 4, pp. 478–505, Dec. 2020, arXiv: 2003.00646. [Online]. Available: http://arxiv.org/abs/2003.00646

    work page arXiv 2020

  22. [22]

    H. V . Le, D. A. Hoang, C. T. Tran, P . Q. Nguyen, V . H. T. Tran, N. D. Hoang, M. Amiri, T. P . T. Ngo, H. V . Nhu, T. V . Hoang, and D. Tien Bui, ‘‘A new approach of deep neural computing for spatial prediction of wildfire danger at tropical climate areas,’’ Ecological Informatics, vol. 63, p. 101300, Apr. 2021. [Online]. Available: https://www.scienced...

    work page 2021

  23. [23]

    Zhang, M

    G. Zhang, M. Wang, and K. Liu, ‘‘Forest Fire Susceptibility Modeling Using a Convolutional Neural Network for Y unnan Province of China,’’ International Journal of Disaster Risk Science, vol. 10, no. 3, pp. 386– 403, Sep. 2019. [Online]. Available: http://link.springer.com/10.1007/ s13753-019-00233-1

    work page 2019

  24. [24]

    ——, ‘‘Deep neural networks for global wildfire susceptibility modelling,’’ Ecological Indicators, vol. 127, p. 107735, Aug. 2021. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1470160X21004003

    work page 2021

  25. [25]

    J. R. Bergado, C. Persello, K. Reinke, and A. Stein, ‘‘Predicting wildfire burns from big geodata using deep learning,’’Safety Science, vol. 140, p. 105276, Aug. 2021. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0925753521001211

    work page 2021

  26. [26]

    Bjånes, R

    A. Bjånes, R. De La Fuente, and P . Mena, ‘‘A deep learning ensemble model for wildfire susceptibility mapping,’’Ecological Informatics, vol. 65, p. 101397, Aug. 2021. [Online]. Available: https://www. sciencedirect.com/science/article/pii/S1574954121001886

    work page 2021

  27. [27]

    Kondylatos, I

    S. Kondylatos, I. Prapas, M. Ronco, I. Papoutsis, G. Camps-V alls, M. Piles, M.-Á. Fernández-Torres, and N. Carvalhais, ‘‘Wildfire danger prediction and understanding with deep learning,’’Geophysical Research Letters, vol. 49, no. 17, p. e2022GL099368, 2022

    work page 2022

  28. [28]

    C. E. V an Wagneret al.,Structure of the Canadian forest fire weather index. Environment Canada, Forestry Service Ottawa, ON, Canada, 1974, vol. 1333

    work page 1974

  29. [29]

    Bouguettaya, H

    A. Bouguettaya, H. Zarzour, A. M. Taberkit, and A. Kechida, ‘‘A review on early wildfire detection from unmanned aerial vehicles using deep learning-based computer vision algorithms,’’Signal Processing, vol. 190, p. 108309, 2022. VOLUME 11, 2023 9 Anastasiouet al.: Wildfire spread forecasting with deep learning

    work page 2022

  30. [30]

    Ghali, M

    R. Ghali, M. A. Akhloufi, and W. S. Mseddi, ‘‘Deep learning and trans- former approaches for uav-based wildfire detection and segmentation,’’ Sensors, vol. 22, no. 5, p. 1977, 2022

    work page 1977

  31. [31]

    N. T. Toan, P . T. Cong, N. Q. V . Hung, and J. Jo, ‘‘A deep learning approach for early wildfire detection from hyperspectral satellite images,’’ in2019 7th International Conference on Robot Intelligence Technology and Applications (RiTA). IEEE, 2019, pp. 38–45

    work page 2019

  32. [32]

    Aminou, ‘‘Msg’s seviri instrument,’’ESA Bulletin(0376-4265), no

    D. Aminou, ‘‘Msg’s seviri instrument,’’ESA Bulletin(0376-4265), no. 111, pp. 15–17, 2002

    work page 2002

  33. [33]

    Y . Zhao, Y . Ban, and J. Sullivan, ‘‘Tokenized Time-Series in Satellite Image Segmentation With Transformer Network for Active Fire Detection,’’IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–13, 2023. [Online]. Available: https://ieeexplore.ieee.org/document/10155171/

    work page arXiv 2023

  34. [34]

    Escuin, R

    S. Escuin, R. Navarro, and P . Fernández, ‘‘Fire severity assessment by using nbr (normalized burn ratio) and ndvi (normalized difference vegetation index) derived from landsat tm/etm images,’’International Journal of Remote Sensing, vol. 29, no. 4, pp. 1053–1073, 2008

    work page 2008

  35. [35]

    Knopp, M

    L. Knopp, M. Wieland, M. Rättich, and S. Martinis, ‘‘A deep learning approach for burned area segmentation with sentinel-2 data,’’Remote Sensing, vol. 12, no. 15, p. 2422, 2020

    work page 2020

  36. [36]

    Sdraka, A

    M. Sdraka, A. Dimakos, A. Malounis, Z. Ntasiou, K. Karantzalos, D. Michail, and I. Papoutsis, ‘‘Floga: A machine learning ready dataset, a benchmark and a novel deep learning model for burnt area mapping with sentinel-2,’’IEEE Journal of Selected Topics in Applied Earth Observa- tions and Remote Sensing, 2024

    work page 2024

  37. [37]

    Gerard, Y

    S. Gerard, Y . Zhao, and J. Sullivan, ‘‘Wildfirespreadts: A dataset of multi- modal time series for wildfire spread prediction,’’ inAdvances in Neu- ral Information Processing Systems, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36. Curran Associates, Inc., 2023, pp. 74 515–74 529

    work page 2023

  38. [38]

    Radke, A

    D. Radke, A. Hessler, and D. Ellsworth, ‘‘FireCast: Leveraging Deep Learning to Predict Wildfire Spread,’’ inProceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence. Macao, China: International Joint Conferences on Artificial Intelligence Organization, Aug. 2019, pp. 4575–4581. [Online]. Available: https://www.ijcai.org/...

    work page 2019

  39. [39]

    Burge, M

    J. Burge, M. Bonanni, M. Ihme, and L. Hu, ‘‘Convolutional LSTM Neural Networks for Modeling Wildland Fire Dynamics,’’ Apr. 2021, arXiv:2012.06679 [cs]. [Online]. Available: http://arxiv.org/abs/2012. 06679

    work page arXiv 2021

  40. [40]

    J. L. Hodges and B. Y . Lattimer, ‘‘Wildland Fire Spread Modeling Using Convolutional Neural Networks,’’Fire Technology, vol. 55, no. 6, pp. 2115–2142, Nov. 2019. [Online]. Available: https://doi.org/10.1007/ s10694-019-00846-4

    work page 2019

  41. [41]

    S. R. Coffield, C. A. Graff, Y . Chen, P . Smyth, E. Foufoula-Georgiou, and J. T. Randerson, ‘‘Machine learning to predict final fire size at the time of ignition,’’International journal of wildland fire, vol. 28, no. 11, pp. 861– 873, 2019

    work page 2019

  42. [42]

    Liang, M

    H. Liang, M. Zhang, and H. Wang, ‘‘A neural network model for wildfire scale prediction using meteorological factors,’’IEEE Access, vol. 7, pp. 176 746–176 755, 2019

    work page 2019

  43. [43]

    Nastos, ‘‘The mediterranean climate,’’Soil Protection in Sloping Mediterranean Agri-Environments: lectures and exercises, p

    P . Nastos, ‘‘The mediterranean climate,’’Soil Protection in Sloping Mediterranean Agri-Environments: lectures and exercises, p. 7, 2009

    work page 2009

  44. [44]

    Garcia-Hurtado, J

    E. Garcia-Hurtado, J. Pey, M. J. Baeza, A. Carrara, J. Llovet, X. Querol, A. Alastuey, and V . R. V allejo, ‘‘Carbon emissions in mediterranean shrub- land wildfires: An experimental approach,’’Atmospheric Environment, vol. 69, pp. 86–93, 2013

    work page 2013

  45. [45]

    Ruffault, T

    J. Ruffault, T. Curt, V . Moron, R. M. Trigo, F. Mouillot, N. Koutsias, F. Pimont, N. Martin-StPaul, R. Barbero, J.-L. Dupuyet al., ‘‘Increased likelihood of heat-induced large wildfires in the mediterranean basin,’’ Scientific reports, vol. 10, no. 1, p. 13790, 2020

    work page 2020

  46. [46]

    Likas, N

    A. Likas, N. Vlassis, and J. J. V erbeek, ‘‘The global k-means clustering algorithm,’’Pattern recognition, vol. 36, no. 2, pp. 451–461, 2003

    work page 2003

  47. [47]

    Prapas, S

    I. Prapas, S. Kondylatos, I. Papoutsis, G. Camps-V alls, M. Ronco, M. Ángel Fernández-Torres, M. P . Guillem, and N. Carvalhais, ‘‘Deep learning methods for daily wildfire danger forecasting,’’ 2021. [Online]. Available: https://arxiv.org/abs/2111.02736

    work page arXiv 2021

  48. [48]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P . Fischer, and T. Brox, ‘‘U-net: Convolutional networks for biomedical image segmentation,’’ 2015. [Online]. Available: https: //arxiv.org/abs/1505.04597

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  49. [49]

    Wang, ‘‘A review on 3d convolutional neural network,’’ in2023 IEEE 3rd International Conference on Power , Electronics and Computer Appli- cations (ICPECA)

    C. Wang, ‘‘A review on 3d convolutional neural network,’’ in2023 IEEE 3rd International Conference on Power , Electronics and Computer Appli- cations (ICPECA). IEEE, 2023, pp. 1204–1208

    work page 2023

  50. [50]

    J. He, L. Li, J. Xu, and C. Zheng, ‘‘Relu deep neural networks and linear finite elements,’’arXiv preprint arXiv:1807.03973, 2018

    work page arXiv 2018

  51. [51]

    Gaussian Error Linear Units (GELUs)

    D. Hendrycks and K. Gimpel, ‘‘Gaussian error linear units (gelus),’’arXiv preprint arXiv:1606.08415, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  52. [52]

    Galdran, G

    A. Galdran, G. Carneiro, and M. A. G. Ballester, ‘‘On the optimal combi- nation of cross-entropy and soft dice losses for lesion segmentation with out-of-distribution robustness,’’ inDiabetic F oot Ulcers Grand Challenge. Springer, 2022, pp. 40–51. VIII. APPENDIX Table 4 lists all the variables used, along with their initial spatial and temporal resolutio...

    work page 2022