Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Building Environment and Energy Engineering | en_US |
| dc.contributor.advisor | Huang, Xinyan (BEEE) | en_US |
| dc.creator | Li, Yizhou | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14427 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Intelligent cross-scale modeling of wildland fire risk and development | en_US |
| dcterms.abstract | Wildfires are a natural ecological regulator, shaping ecosystems for millennia by managing biomass and nutrients. However, human expansion, especially into wildland-urban interface (WUI) zones, has heightened risks. Climate change exacerbates this by extending dry periods, increasing vegetation flammability. Critically, historical fire suppression policies have backfired, accumulating dangerous fuel loads and fostering soils prone to intense fires, making ecosystems more vulnerable. This challenge is acute in Asia, where research lags despite unique risks: complex WUI landscapes, fragmented vegetation, dense populations, monsoonal climates, fuels and topography. Models from temperate and arid regions fail here. Rapid urbanization further increases exposure. Hong Kong exemplifies this: 70% vegetated land adjacent to hyper-urban areas creates a high-risk, fragmented WUI with frequent hill fires threatening infrastructure, biodiversity, and air quality. Existing models fail to capture its complex multi-scale dynamics involving micro-terrain, human ignitions, and monsoonal winds. This thesis addresses these limitations by pioneering a multi-scale deep learning modeling framework for Hong Kong, integrating high-resolution satellite data, wildfire modeling and deep learning to resolve fire spread from micro-terrain to regional scales. | en_US |
| dcterms.abstract | This thesis is to address this need and presented with 6 chapters. A brief introduction of the contents in each chapter is given as follows: | en_US |
| dcterms.abstract | Chapter 1 introduces the global wildfire crisis, emphasizing unique challenges in Asian urban-proximate regions and the critical need for multi-scale predictive tools. It reviews gaps in understanding Asian wildfire regimes, wildfire modeling simulation limitations, coarse-resolution forecasting constraints, and cross-scale dynamics, outlining the thesis objectives. | en_US |
| dcterms.abstract | Chapter 2 analyzes the spatiotemporal distribution of 12,545 wildfires in Hong Kong (2010-2024), revealing a human-dominated, high-frequency and low-intensity regime distinct from common wildfires in the North America. It quantifies anthropogenic drivers (e.g., 24.9% surge during grave-sweeping), identifies key risk factors (WUI proximity, vegetation fraction, density), and establishes pyrogens. A mitigation framework for fire-resilient Asian megacities is proposed. | en_US |
| dcterms.abstract | Chapter 3 tackles micro-scale wildfire prediction by integrating CFD physics with deep learning. We generate a high-fidelity database of 64,000 grassfire simulations and develop a Long Short-Term Memory (LSTM) network predicting firefront movement. Achieving an indicator of coefficient of determination R²=0.98 agreement with CFD, this approach drastically reduces computational cost while preserving physics-based accuracy for homogeneous fuels, establishing a benchmark for data-driven wildfire research. | en_US |
| dcterms.abstract | Chapter 4 develops a dual-model deep learning framework (U-Net + ConvLSTM) for super real-time (>10²-10⁴ speedup), high-resolution (5m) mid-scale wildfire forecasting. Addressing satellite data limitations, it forecasts burning area 5 hours ahead and refines predictions in real-time. Trained on 210 Sunshine Island cases (12,600 samples), it achieves >90% accuracy, enabling rapid operational forecasting where traditional methods fail. | en_US |
| dcterms.abstract | Chapter 5 develops a cross-scale deep learning framework for high-resolution wildfire forecasting, explicitly integrating micro- to macro-scale dynamics (25 m² to 20 km²). Using adaptive resolution transitioning (5-m resolution for early-stage fires <12 h and 320-m for large-scale propagation), it achieves efficient predictions without full-domain high-resolution modeling costs. Trained on 240 Hong Kong wildfire cases (8,640 samples), the model attains >75% accuracy across 2–72 h lead times. The framework is operationalized through the Intelligent Wildfire Forecast Tool (IWFTool) for real-time emergency response. | en_US |
| dcterms.abstract | Chapter 6 summarizes the conclusions and suggests feasible ideas for future research. | en_US |
| dcterms.extent | ix, 96 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2026 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Wildfires | en_US |
| dcterms.LCSH | Wildfires -- Risk assessment | en_US |
| dcterms.LCSH | Deep learning (Machine learning) | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14427