Skip to content

relatorio-busca-koopman-pignn-fogo.md

Latest commit

 

History

History
117 lines (90 loc) · 5.53 KB

relatorio-busca-koopman-pignn-fogo.md

File metadata and controls

117 lines (90 loc) · 5.53 KB

Relatorio de Busca: Neural Koopman Operator, PI-GNN e Propagacao de Fogo

Data: 13/06/2026 Projeto: QueimandasGemeosDigitais/ceara-queimadas Escopo: Base de conhecimento local + Projeto + Web (arXiv)

SUMARIO EXECUTIVO

A combinacao Neural Koopman Operator + Physics-Informed Graph Neural Networks (PI-GNN) para propagacao de fogo e um gap absoluto na literatura: ZERO publicacoes combinam essas tres tecnicas. O unico concorrente identificado e Tang Sui (2026), que usa GNN com fisica mas sem Koopman e sem dados VIIRS/GOES-16 do Brasil.

PARTE 1 - REFERENCIAS DO PROJETO

KOOPMAN OPERATOR THEORY

  1. Koopman (1931) Hamiltonian Systems - PNAS 17(5), 315-318
  2. Mezic (2013) Analysis of Fluid Flows via Koopman Operator - Annu. Rev. Fluid Mech. 45, 357-378
  3. Rowley et al. (2009) Spectral Analysis of Nonlinear Flows - J. Fluid Mech. 641, 115-127. DOI: 10.1017/S0022112009992059
  4. Schmid (2010) Dynamic Mode Decomposition - J. Fluid Mech. 656, 5-28. DOI: 10.1017/S0022112010001217
  5. Williams et al. (2015) Extended DMD - J. Nonlinear Sci. 25, 1307-1346. DOI: 10.1007/s00332-015-9258-5
  6. Brunton et al. (2016) SINDy - PNAS 113(15), 3932-3937. DOI: 10.1073/pnas.1517384113
  7. Lusch et al. (2018) Deep Learning for Universal Linear Embeddings - NeurIPS 31
  8. Takeishi et al. (2017) Learning Koopman Invariant Subspaces - NeurIPS 30
  9. Proctor et al. (2016) DMD with Control - SIAM J. Appl. Dyn. 15(1), 142-161. DOI: 10.1137/15M1013857
  10. Baddoo et al. (2023) Physics-Informed DMD - Proc. R. Soc. A 479(2271), 20220576. DOI: 10.1098/rspa.2022.0576
  11. Lorenzo-Sanchez et al. (2026) PI-Koopman for ENSO - arXiv:2606.09369. URL: https://arxiv.org/abs/2606.09369

PHYSICS-INFORMED NEURAL NETWORKS

  1. Karniadakis et al. (2021) Physics-Informed ML - Nat. Rev. Phys. 3(6), 422-440. DOI: 10.1038/s42254-021-00314-5
  2. Lu et al. (2021) DeepXDE - SIAM Rev. 63(1), 208-228. DOI: 10.1137/19M1274067
  3. Cuomo et al. (2022) Scientific ML Through PINNs - J. Sci. Comput. 92(3), 88. DOI: 10.1007/s10915-022-01928-w
  4. Wang et al. (2021) Gradient Flow Pathologies in PINNs - SIAM J. Sci. Comput. 43(5), A3055-A3081. DOI: 10.1137/20M1380068
  5. Mao et al. (2024) PINNs for Wildfire Spread - J. Comput. Phys. 507, 112978. DOI: 10.1016/j.jcp.2024.112978
  6. Vogiatzoglou et al. (2024) PINNs for Wildfire Spreading - arXiv:2406.14591. URL: https://arxiv.org/abs/2406.14591

PHYSICS-INFORMED GNNs + WILDFIRE

  1. Esparza et al. (2025) GraphFire-X: PI-GNN for WUI - arXiv:2512.20813. DOI: 10.1016/j.cacaie.2026.100085
  2. Davalas et al. (2024) Causal GNNs for Wildfire Danger - arXiv:2403.08414 (ICLR 2024 ML4RS)
  3. Michail et al. (2025) FireCastNet - arXiv:2502.01550
  4. Lam et al. (2023) GraphCast - Nature 625, 559-567. DOI: 10.1038/s41586-023-06776-6
  5. Zhenirovskyy et al. (2026) Neural-Parameterized CA for Wildfire - arXiv:2606.11676

PARTE 2 - BUSCA NA WEB (arXiv API)

Neural Koopman Operator (13 resultados)

  • "Learning Neural Koopman Operators with Dissipativity Guarantees" (arXiv:2509.07294, CDC 2025)
  • "Scaling Law of Neural Koopman Operators" (arXiv:2602.19943, 2026)
  • "Learning Hamiltonian Neural Koopman Operator" (arXiv:2406.02154, 2024)
  • "Dynamic Neural Koopman Distillation" (arXiv:2605.24924, 2026)
  • "Physics-Informed Neural Koopman Machine" (arXiv:2512.06134, 2025) GAP: ZERO papers aplicam Koopman a propagacao de fogo.

PI-GNN + Wildfire (1 resultado: GraphFire-X)

GAP: Koopman + PI-GNN + fogo = ZERO publicacoes.

GraphCast + Fire (1 resultado: FireCastNet, arXiv:2502.01550)

PARTE 3 - INOVACAO PROPOSTA

Gap: Neural Koopman Operator + PI-GNN + Wildfire = 0 publicacoes

Matematica Central: dz/dt = f(z, x, t) -> Koopman: g(z_{t+1}) = K * g(z_t) K = argmin ||G+A - K||^2_F (EDMD + PINN-Rothermel regularization)

Arquitetura: GNN -> parametros fisicos -> PDE solver -> previsao

Periodicos Alvo A1:

  • Environmental Modelling & Software (Qualis A1, JCR Q1)
  • ISPRS J. Photogrammetry & Remote Sensing (A1, Q1)
  • IEEE T. Geoscience & Remote Sensing (A1, Q1)
  • Remote Sensing of Environment (A1, Q1)

PARTE 4 - REFERENCIAS ADICIONAIS

Digital Twins para incendio

  • Morsali & Khajavi (2026) IVSR - arXiv:2602.08949
  • Webb et al. (2026) FIRE-VLM - arXiv:2601.03449
  • Raha et al. (2025) FIRETWIN - arXiv:2510.18879
  • Zhou et al. (2025) AIMNET - arXiv:2512.06148

Deep Learning + Wildfire

  • Chowdhury et al. (2023) IEEE JSTARS 16, 5753-5765
  • de Silva et al. (2023) IEEE Trans. AI 4(5), 1207-1225
  • Schmidt et al. (2025) Environ. Model. Softw. 182, 106203

Sensoriamento Remoto Fogo

  • Giglio et al. (2016) Remote Sens. Environ. 178, 31-41
  • Schroeder et al. (2014) Remote Sens. Environ. 143, 85-96
  • Xu et al. (2020) Remote Sens. Environ. 245, 111818

Ecologia do Fogo Brasil

  • Pereira et al. (2024) ISPRS J. Photogramm. 210, 179-196
  • Barbosa et al. (2024) Global Change Biol. 30(4), e17265
  • Cano et al. (2024) Environ. Res. Lett. 19(4), 044005

PARTE 5 - FONTES CONSULTADAS

Local (projeto)

  • /Users/naubergois/gois/references.bib (59+ entradas)
  • refs-queimadas-v3.2-expanded.bib (86+ entradas)
  • research-report.md (35+ papers)
  • project_memory.json (knowledge base)
  • INOVACAO-A1-PESQUISA.md (inovacao matematica)
  • ceara-queimadas/docs/refs-queimadas.bib (831 linhas)

Web (arXiv API)

  • "neural koopman" -> 13 resultados
  • "physics-informed graph attention" AND wildfire -> 1 (GraphFire-X)
  • "PI-GNN" OR "physics-informed graph neural" -> 53 resultados
  • "graphcast" AND fire -> 1 (FireCastNet)

Arquivo gerado

  • /Users/naubergois/gois/relatorio-busca-koopman-pignn-fogo.md

Relatorio gerado em 13/06/2026 pelo cron job Hermes Agent.