Climate change has prolonged droughts and intensified heatwaves worldwide, increasing the frequency, intensity, and spatial extent of forest fires. In South Korea, an increasing incidence of large forest fires has created a need for rapid, accurate forest fire-damaged area boundary delineation to support post-fire recovery and restoration planning. This is particularly important in South Korea, where the forest-fire season is followed by the summer rainy season, increasing the risk of post-fire soil erosion and slope instability and thereby increasing the need for rapid stabilization and restoration measures. However, previous studies have reported varying delineation performance depending on the input configuration, suggesting the need to examine how different spectral inputs and their integration affect boundary delineation. To address this limitation, we propose a hierarchical ensemble U-Net framework for forest fire-damaged area boundary delineation using Sentinel-2 imagery. To design the input data structure, differences between burned and unburned areas were examined using the Blue, Green, Red, RedEdge, and NIR bands and the NDVI and dNDVI indices, and four three-channel combinations (Blue–Green–Red, Blue–Green–RedEdge, Green–Red–NIR, and Green–RedEdge–NIR) and two single indices (NDVI and dNDVI) were chosen as inputs. Three U-Net models were then evaluated: a single-channel dNDVI model, a three-channel NIR-RE-G model, and a hierarchical ensemble model, which integrated the outputs of six independently trained U-Nets. Among the three models, the single-channel dNDVI model showed higher overall accuracy (84.1%), producer’s accuracy (89.0%), and F1-score (0.80) than the three-channel NIR-RE-G model (82.9%, 70.7%, and 0.77, respectively), whereas the NIR-RE-G model achieved higher user accuracy (91.8%); the hierarchical ensemble model yielded the best overall performance, with the highest overall accuracy (85.7%), producer’s accuracy (92.2%), and F1-score (0.83). This suggests that integrating complementary spectral responses can mitigate the limitations of single-input models and support more robust boundary delineation across the six tested wildfire cases in South Korea.
Aviation kerosene leaked during airport storage, refueling, and ground-support operations can infiltrate soil pores and form oil-contaminated soil systems with combustion behavior different from that of free liquid pool fires. To clarify the regulating effect of soil physical properties on fire hazards, humus soil, lateritic soil, loess, and sandy soil were contaminated with 10, 30, 50, 70, and 90 g of No. 3 aviation kerosene and tested using a cone calorimeter under an external heat flux of 35 kW/m2. For humus soil, lateritic soil, and loess, the time to ignition generally ranged from 3.5 to 7.0 s. In contrast, sandy soil showed a distinctly delayed and nonmonotonic ignition response, with the longest time to ignition reaching approximately 18 s. Heat release increased with fuel loading, but the response varied markedly among soils. At 90 g, the peak HRR values of loess, humus soil, lateritic soil, and sandy soil were 1021, 964, 806, and 461 kW/m2, respectively, and their THR values were 380, 354, 356, and 334 MJ/m2. The mass-loss results further showed that MLR was not directly equivalent to effective heat release: sandy soil had a relatively high peak MLR of approximately 0.40 g/s at 90 g, but its HRR remained the lowest because part of the fuel loss was associated with downward migration or inefficient near-surface flaming. In contrast, loess combined the highest peak MLR, approximately 0.43 g/s, with the highest HRR, indicating efficient conversion of released fuel into flaming heat output. Smoke production was also substrate-dependent, with loess showing the strongest combined smoke hazard and humus soil exhibiting sustained smoke production. Overall, permeability, together with pore structure and fuel retention characteristics, was associated with the distribution of aviation kerosene between the near-surface region and deeper soil layers. These soil-dependent differences were reflected in ignition, heat release, mass loss, and smoke generation. Under the present test conditions, loess showed the highest combined fire hazard, followed by humus soil, lateritic soil, and sandy soil.
Immersion liquid cooling is a thermal runaway protection strategy for lithium-ion batteries; however, the regulatory mechanism of immersion depth on battery thermal runaway remains unclear. In this study, 18650-type ternary lithium-ion batteries with capacities of 2000, 2600, and 3200 mAh were selected as research objects, and HFE-7200 dielectric fluorinated liquid was used as the cooling medium. A self-built immersion battery thermal runaway test platform was employed to investigate the influence of immersion depth on thermal runaway characteristic parameters. The kinetic laws of temperature variation, thermal runaway mass loss, and toxic gases, including CO, NO2, NH3, and HF, were discussed. The results show that immersion depth exerts a nonlinear regulatory effect on thermal runaway. Shallow immersion of 1–2 cm cools the lower part of the cell, prolongs the thermal runaway incubation time, and may intensify thermal runaway, whereas deep immersion of 3–4 cm can effectively suppress thermal runaway. When thermal runaway was effectively suppressed, the maximum system temperature was 264.3 °C. The average mass loss rates of the 2000, 2600, and 3200 mAh batteries under different immersion depths were 16.2%, 18.6%, and 38.8%, respectively. The release sequence of the four major toxic gases during battery thermal runaway was determined, and their peak concentrations were compared. A three-zone heat transfer model was established along the battery axial direction, clarifying that the proportional changes of the fully cooled zone, vapor-liquid coupling transition zone, and unimmersed runaway zone jointly govern the heat generation and heat transfer balance of the cell. This study can provide a quantitative experimental basis and theoretical support for safety liquid level calibration of immersion liquid cooling battery systems.
This study tested whether fine particulate matter (PM2.5) emissions from forest fuel combustion in the Xiao Xing’an Mountains are controlled by common or vegetation-category-specific factors. Six conifer, nine broad-leaved and seven shrub species were burned in a laboratory chamber (6966 measurements). Because the factors tested differed among categories, effects were analyzed within each category and compared across categories only for shared factors, using ANOVA, multiple linear regression, Random Forest and XGBoost. Fuel load was a primary factor in all three categories. In addition, PM2.5 was controlled by moisture content, plant part and combustion mode in shrubs; by plant part, season and species in broad-leaved species; and by plant part and wind speed in conifers. Among the shared factors, moisture content had a clear effect only in shrubs (partial η2 = 0.16 vs. ≤0.03 in trees). Smoldering tests produced 34% more PM2.5 than flaming tests in shrubs (p < 0.001), and broad-leaved bark and branches had 89% higher PM2.5 concentrations in April than in July. The mean shrub emission factor was 12.5 g/kg, without significant inter-species variation (p = 0.668). Under grouped five-fold cross-validation, Random Forest and XGBoost reached R2 of 0.77–0.96, compared with 0.21–0.74 for linear models. The results support vegetation category as a stratification variable in regional forest-fire emission inventories.
Wooden site cabins and modular units are combustible enclosures in which a single furniture fire can rapidly involve adjacent rooms, yet full-scale experimental data on multi-compartment fire development involving combustible interior linings remain scarce. This paper presents a full-scale fire experiment in a two-room wooden site cabin with combustible linings, supported by a simplified numerical investigation using the Fire Dynamics Simulator (FDS). An F21-type armchair, ignited using heptane as an accelerant in the corner of the fire room, served as the primary fire source. Temperature development was monitored using 19 thermocouples distributed across the compartments, doorways, and window. During the experiment, the fire developed rapidly from the armchair, reaching flashover in the primary room after approximately 2.5 min before spreading through the internal doorway into the adjacent compartment and subsequently causing window failure. The FDS model employed a simplified combustible-lining representation, with surface ignition at 340 °C and a prescribed heat release rate per unit area (HRRPUA) of 150 kW m−2. Following an empirically determined temporal alignment of 46 s, the simulation reproduced the general temperature development during the growth phase and early fully developed stage. The interval between the earliest flashover threshold crossings in the two rooms was 36 s experimentally and 24 s in the simulation, indicating a more rapid transition between compartments in the model. This difference may partly reflect the simplified treatment of combustible surface ignition and fire spread. However, the temporal alignment and simplified modelling assumptions limit conclusions regarding independent predictive accuracy, particularly for event timing and later-stage temperature development. The primary contribution is a full-scale experimental reference case for investigating fire development and spread in lightweight, combustible multi-compartment structures. The experimental data and supporting numerical materials are openly available through Zenodo, facilitating further model assessment and research.
