Slope–Wind Coupling Effects on Fire Behavior and Emission Dynamics During Prescribed Burning in Mountainous Yunnan Pine Forests

Abstract

Prescribed burning is important for reducing wildfire risk and regulating fuel loads, but its implementation in mountainous forests is strongly influenced by the coupled effects of the wind field and topography, making it difficult to control. This study focuses on Yunnan pine (Pinus yunnanensis) forests in southwestern China. A three-dimensional Fire Dynamics Simulator (FDS) combined with measured fuel characteristics was used to simulate 21 slope (0–35°) and wind speed (0–2 m s⁻¹) combinations to quantitatively analyze the fire spread, flame structure, and gaseous emission characteristics during downslope prescribed burning. The local fire spread rate (ROS), evaluated along three lateral lines (Y = 2.5, 5.0, and 7.5 m), exhibits a non-monotonic dependence on slope over the tested range, with a minimum near 30° and a modest rebound at 35°. A downslope wind of 1 m s⁻¹ promotes near-surface heating and accelerates spread, whereas a stronger wind of 2 m s⁻¹ lifts flames away from the fuel bed and suppresses combustion. Thermal field analysis reveals that peak temperature decreases with increasing slope and that a late-stage secondary heating episode occurs at 35°. CO₂ emissions are significantly positively correlated with fuel consumption, reaching a peak of 717.5 kg under a 35° slope and no-wind conditions. CO emissions, as an indicator of combustion efficiency, reach their highest value of 2.23 kg at a 35° slope and a wind speed of 1 m s⁻¹, indicating that their trend is not entirely consistent with the ROS and temperature and that there is a certain degree of decoupling. The interaction between slope and wind speed transforms fire behavior from a cooperative to a competitive mechanism, and the topography–wind field coupling provides differentiated control over the combustion intensity and completeness. This study provides a scientific basis for the safe implementation of mountain burning programs and for regional carbon emission assessments.

1. Introduction

Climate change has significantly exacerbated global wildfire activity [1,2] and continues to reshape the spatial patterns of vegetation [3,4,5], thereby accelerating carbon cycle processes in terrestrial ecosystems [6,7]. These changes present greater challenges to sustainable forest management, including increased difficulty in fire risk assessment, the dual pressures of ecological restoration and carbon balance, and the increased complexity of cross-scale governance [8,9]. Extreme fire incidents have occurred frequently in recent years. For example, the massive wildfires that occurred in Canada in 2023 [10] burned an area of approximately 18.5 million hectares, which is 6–7 times the historical average annual area [11], and released approximately 647 Tg of carbon [12]. Such events demonstrate that extreme fires not only weaken regional carbon storage stability but also damage ecosystem structure and function, profoundly affecting climate feedback.

With the increase in wildfire activity and extreme events, relying on only emergency firefighting became insufficient to reduce risks at the source, and forest fire management shifted from post-event response to pre-event disaster reduction. Therefore, fuel management strategies that reduce fire intensity and the probability of spread by regulating fuel load and continuity have received widespread attention, and prescribed burning has become a widely adopted forest fuel management strategy in many countries [13,14,15,16,17]. In southwestern China, particularly in the Yunnan Plateau forests dominated by Yunnan pine (Pinus yunnanensis), prescribed burns have long been considered a primary measure for controlling combustibles. Studies have shown that this method helps maintain ecological stability [18,19] and effectively slows the rate of fuel accumulation [20,21]. Because Yunnan pine forests are mostly distributed in mountainous areas with complex terrain and steep slopes [22], their fire behavior is easily affected by local wind fields [23]. Topographical differences can alter the structure of slope winds, valley winds, and leeward circulation, thereby affecting fuel moisture and flame spread characteristics [20]. Therefore, revealing the role of the wind field–slope coupling mechanism in prescribed burning behavior is highly important for understanding mountain fire dynamics and optimizing regional fire management strategies.

Under coupled wind–slope conditions, the spread of forest fires is jointly controlled by the wind speed and slope [24]. Traditional observational and experimental studies have revealed the main influences of both flame morphology and spread rate [25,26,27], but owing to complex terrain, variable weather, and experimental safety limitations, obtaining the transient structure and energy transfer patterns of flame–flow field interactions is difficult. In recent years, numerical simulation methods have gradually become important approaches for studying wind–slope coupled fire behavior. Research based on computational fluid dynamics (CFD) methods, especially models using the large eddy simulation (LES) method, can reconstruct temperature, velocity, and heat flux distributions under controlled conditions, thereby characterizing multiscale turbulence features and thermodynamic feedback processes [28,29]. Studies have shown that this method has significant advantages in revealing topographic dynamics [30].

The application of CFD has facilitated the quantitative analysis of differences in flame–airflow interactions under different terrain conditions. Uphill fires often exhibit an acceleration effect caused by the superposition of buoyancy and radiant preheating [31,32], whereas downhill fires show a nonlinear propagation pattern that varies with slope [33]. In the leeward slope region, flames and airflow can form “rising–sinking boundary rotational flow”, triggering transient lateral expansion [34]. However, steeper slopes and headwinds can weaken radiative heat transfer and disrupt the Coanda effect [35,36]. These rapid and complex flow–thermal coupling processes can be accurately characterized only through high-resolution numerical simulations.

Therefore, CFD methods not only compensate for the shortcomings of experiments and observations in terms of spatiotemporal resolution and repeatability but also provide a new research approach for revealing the dynamic structure of flames under wind–slope coupling. Although existing studies have systematically analyzed the spread of fires downhill from a dynamic perspective, quantitative research on combustion product emissions remains relatively weak. During the prescribed combustion, the coupling effect of the wind speed and slope directly affects the flame stability and combustion efficiency, thus determining the emission characteristics of carbon dioxide (CO₂) and carbon monoxide (CO). Previous studies have mostly focused on overall emissions under different fuel types or fire intensities, while a systematic quantitative assessment of how changes in the wind field and slope influence the CO₂ and CO emission patterns under downhill ignition conditions is lacking.

To address the limited understanding of how wind and slope dynamics jointly influence CO₂ emissions during prescribed burns, this study aims to (1) quantify how coupled slope–wind dynamics alter the rate of spread and the temperature field using LES modeling; (2) reveal the variation patterns of the CO₂ and CO release rates and the total amounts under different wind speeds and slopes; and (3) provide a scientific basis for the assessment, management, and optimization of controllable carbon emissions during prescribed burning in mountainous regions.

2. Materials and Methods

2.1. Study Area

This study was conducted in a typical Yunnan pine forest in central Yunnan Province, southwestern China (102°00′19″–102°02′10″ E, 24°01′15″–24°02′27″ N). The terrain in the study area is generally high in the northwest and low in the southeast, with elevations ranging from approximately 422 to 3165 m. The slopes vary considerably, ranging from relatively gentle 5–10° to steep 30° or more [22]. The region has a temperate monsoon climate, with an average annual temperature of 18.1 °C, an annual precipitation of approximately 869 mm, and an annual sunshine duration of approximately 2800 h [37]. Significant climate gradients and complex topography together influence vegetation distribution and surface fuel accumulation. Yunnan pine (P. yunnanensis) is the dominant tree species, forming medium-aged stands, with a typical tree height of 10–20 m [18]. The thick layer of pine needles under the forest can easily dry out during the dry season, forming a continuous and flammable layer of fallen leaves and dead branches and providing abundant fuel for the spread of fire.

Zhaobishan Mountain was chosen as the research site primarily because of its long-term prescribed burning activities aimed at reducing combustible load and controlling wildfire risk [37]. The terrain is complex, with variable wind directions and altitudes, and the extensive Yunnan pine cover gives rise to the typical flammable characteristics of southwestern mountain forests [21,38]. The local prescribed burning system has been in place for more than 20 years and is usually carried out from late January to mid-February each year. Long-term standardized management has accumulated combustible material characteristic data and fire record archives for the region, providing a strong foundation for process-based modeling and validation. Field measurements and experimental burns for this study were conducted on 12 February 2022.

2.2. Fuel Data and Laboratory Measurements

Ten 10 m × 10 m plots were established within the studied Yunnan pine forest to obtain the field data required for model parameterization. The height and spatial coordinates of the trees were recorded in each plot, and these data were used as input variables for the model structure. On the basis of field surveys, the surface can be divided into three vertical layers: the upper layer, which consists of recently fallen pine needles; the middle layer, which is a mixture of partially decomposed litter and topsoil; and the lower layer, which is a mineral matrix. This vertical stratification structure was incorporated into the fuel layer configuration, in which pine needle litter was an important surface fuel affecting flame spread.

To quantify the thermal and physical properties of surface fuels, five 1 m × 1 m subplots were randomly selected within each 10 m × 10 m plot to collect litter samples, resulting in a total of 50 samples. The average fuel layer thickness for each subplot was recorded and calculated as a representative value for surface fuel thickness. All the samples were sealed and labelled in polyethylene bags before being transported to the laboratory for physicochemical analysis (Figure 1).

The laboratory analysis process followed the operating procedures established in our earlier studies [39,40]. After collection, the samples were first air-dried at room temperature for 48 h and then dried in an oven at 105 °C until a constant weight was reached to remove residual moisture. The dried sample was ground and sieved through a 1 mm sieve to obtain a homogeneous sample, which was used to determine the calorific value and flash point. The total calorific value was determined using an oxygen bomb calorimeter (Model XRY–1C), and the flash point was determined using a DW–02 ignition meter. Each test was repeated three times, and the average value was used for model parameter analysis and validation.

2.3. Fire Dynamics Simulator Configuration and Ignition Setup

2.3.1. Computational Domain Construction

Fire Dynamics Simulator (FDS, version 6.7.9) was used to conduct numerical simulations. The computational domain was configured to represent the plot-scale surface fire experiments while limiting boundary effects on the flow and flame development (Figure 2a). The core plot size was set to 10 m × 10 m to match the field plots. To provide sufficient buffer space for inflow development and outflow dissipation, the overall domain size was extended to 18 m × 18 m × 16 m (X × Y × Z).

Because this study focuses on within-plot surface fire behavior and does not analyze above-canopy or plume-rise outputs, the domain height was set slightly above the mean crown height (16 m vs. 12 m) while controlling computational cost. The mean canopy height was 12 m, and the mean crown base height was approximately 8 m. In the simulations, the canopy was represented as a geometric obstruction with parameterized drag rather than as a combustible fuel component. Therefore, tree crowns were idealized as cones to reduce computational costs while preserving the bulk canopy height and spatial occupancy (Figure 2b).

2.3.2. Surface and Fuel Bed Parameterization

The surface fuel bed was represented using a layered structure that reflects the observed ground strata. The surface was divided into mineral soil, a soil–litter mixture layer, and a pine needle litter layer (Figure 2c;

Table 1. Surface layer configuration and material assignment.

Surface Layer	Thickness (m)	Combustible in FDS	FDS SURF ID	Material Composition
Mineral Soil	0.15	No	Soil	Soil:1.0
Soil–litter mixture	0.05	No	Pine needle and soil	Yellow Pine: 0.75; Soil: 0.25
Pine needle litter	0.10	Yes	Pine needle	Yellow Pine: 1.0

). Thermodynamic properties for the defined materials followed the values implemented in the FDS input. Mineral soil was treated as inert and non-combustible.

Surface burning of the pine needle litter was implemented using a prescribed heat release rate per unit area (HRRPUA) approach in FDS. Specifically, the pine needle surface was assigned HRRPUA at 500 kW m⁻², and BURN_AWAY was enabled such that the surface layer could be consumed after ignition. Under this representation, the heat release rate is imposed once the ignition criterion is met, allowing the model to focus on fire flow coupling and plot-scale flame spread behavior without requiring detailed pyrolysis kinetics.

It should be noted that, under the prescribed HRRPUA approach, the specified solid phase material properties (e.g., density) primarily control the thermal response and the numerical burn-away behavior of the surface layer (

Table 2. Physical and thermal parameters of the surface strata used in the model simulations.

Material	Density (kg m−3)	Conductivity (W m−1 K−1)	Specific Heat (kJ kg−1 K−1)	Heat of Combustion (kJ/kg)	Source
Soil	1300	0.25	2.00	-	Inert material
Yellow Pine	640	0.14	2.85	20,137.37	NIST NRC validation
Soil–litter mixture	735	0.17	2.64		Mass fraction mixing

). They are not intended to serve as a direct stand-level inventory of litter mass for the field plots. Accordingly, quantities derived from the simulation are interpreted mainly for relative comparisons across wind and slope scenarios, and emission-related outputs are reported and discussed primarily in a normalized form relative to the baseline case.

2.3.3. Boundary Condition and Output Settings

Boundary conditions were specified to represent exchange between the fire-induced flow and the ambient atmosphere. The left boundary was defined as an inflow boundary with a fixed wind speed to drive the background ventilation through the domain. Canopy and stem drag were parameterized to represent momentum loss and wind speed attenuation within the stand. The remaining lateral boundaries and the top boundary were treated as open outflow boundaries, allowing air and pressure perturbations to exit the domain and reducing artificial boundary reflection. The bottom boundary beneath the fuel layers was modeled as an adiabatic, inert wall such that heat conduction into the deeper soil was neglected in the present simulations.

The simulation duration was determined based on the fire development process and therefore varied across scenarios. Each simulation was continued until the heat release rate (HRR) entered the decay stage, defined as the time when HRR remained below 80% of its peak value after the peak had occurred. A maximum simulation time of 10,000 s was imposed as an upper bound to prevent excessively long runs in cases where the termination criterion was not reached. Output intervals were configured as 1 s to capture transient flame spread and thermal evolution. The recorded outputs were used to quantify energy release behavior and to derive the fire spread metrics described in Section 2.3.5.

2.3.4. Ignition Configuration

To establish a stable flame front and reduce the impact of random disturbances, this study adopted a linear ignition strategy. The ignition zone was located on the windward edge of the model, measuring 0.5 m × 10 m, and was arranged along the Y-axis (Figure 2c). The heat release rate per unit area was set at 500 kW m⁻², a value that references and conforms to validated field-scale experimental results [39].

2.3.5. Monitoring Layout and Data Collection

Flame propagation and temperature changes were monitored using 27 thermocouples. These thermocouples were arranged along three parallel lines at Y = 2.5 m, 5.0 m, and 7.5 m. Nine sensors were installed on each observation line, spaced 1 m apart along the X-axis (Figure 2c). When the temperature recorded by a sensor exceeded 300 °C, the flame front was considered to have reached the measuring point [41]. The ROS was calculated using linear regression, with the flame arrival time as the independent variable and the sensor location as the dependent variable. The regression slope represents the flame propagation speed, whereas the coefficient of determination (R²) reflects the goodness of fit of the model.

2.3.6. Computation of the Rate of Spread (ROS)

ROS values obtained under varying wind and slope conditions were normalized to a baseline case (0° slope, 0 m s⁻¹ wind) using the following equation:

$$\mathit{rROS} = {\displaystyle \frac{{\mathit{ROS}}_i}{{\mathit{ROS}}_0}}$$

(1)

where ROS₀ is the baseline spread rate, and where ROS_i represents the spread rate under each simulated scenario.

2.4. Computation of Combustion Efficiency

Combustion efficiency (η) was quantified using an energy ratio method:

$$\eta = {\displaystyle \frac{{\int }_0^{t_1}Q\left(t\right)dt}{m_f\Delta H_c}}$$

(2)

Here,$Q\left(t\right)$ is the instantaneous HRR output from FDS (kW), $m_f$ is the initial total mass of the surface fuel bed within the 10 m × 10 m core plot (kg), and $\Delta H_c$ is the effective heat of combustion of Yunnan pine fuels (kJ kg ⁻¹) measured by bomb calorimetry. The end time $t_1$ was determined using the same threshold criterion as that used for the cumulative emission calculation (Section 2.5), defined as the first time after the peak when the CO₂ mass flux remained below 5% of its peak value for at least 60 s. Under this definition, η represents an apparent combustion efficiency.

2.5. Simulation Scenario Design

To explore wind–slope coupling, the simulation parameters were varied for both variables. In accordance with the Technical Specification for Prescribed Burning in Pine Forests, prescribed burns are conducted under wind speeds below 3 m s⁻¹. Therefore, three wind regimes (0, 1, and 2 m s⁻¹) were selected to represent calm, light, and moderate winter burn conditions. In addition, seven slope angles α (0°, 10°, 15°, 20°, 25°, 30°, and 35°) were selected to cover the operational topographic range of prescribed burns in southwestern China. All simulations represent downslope spread; for clarity, slope is reported as the absolute angle α (0–35°), and the downhill direction is described in words rather than as a negative sign. Slopes steeper than 35° were excluded because they are rarely burned deliberately.

Downslope fires were simulated by projecting the gravitational vector along the slope to maintain computational stability. Combining seven slopes and three winds produced 21 simulations. The ambient temperature (23 °C) and relative humidity (35%) were held constant to mimic field conditions.

2.6. CO₂ and CO Emission Computation

The masses of carbon dioxide (CO₂) and carbon monoxide (CO) were quantified using FDS species mass flux outputs integrated over a prescribed control surface. The control surface was defined as a horizontal plane located at Z = 0.3 m spanning the core plot area, which provides a consistent reference for near-surface smoke transport across scenarios.

Because the control surface is located within the computational domain, local recirculation and turbulent eddies may cause intermittent negative fluxes. To avoid inflating cumulative values by counting bidirectional crossing as net transport, an outflow-only metric was adopted. Specifically, the instantaneous outflow mass flow rate $m_{out}\left(t\right)$ was defined as the positive part of the signed surface integrated mass flux, and the cumulative transported masses were computed as

$$M_{{\mathit{CO}}_2,out}={\int }_{t_0}^{t_1}\dot{\mathit{max}(m{\mathit{co}}_2\left(t\right)},0)\mathit{dt} M_{CO,out}={\int }_{t_0}^{t_1}\dot{\mathit{max}(\mathit{mco}\left(t\right),0)}\mathit{dt}$$

(3)

where t₀ is the ignition time and t₁ is the effective end of the burning period. The end time t₁ was determined by a flux-based threshold criterion, defined as the first time after the peak when the CO₂ outflow rate remained below 5% of its peak value for at least 60 s. Sensitivity tests using thresholds of 3–10% and persistence windows of 30–120 s produced negligible changes in normalized cumulative results and did not affect the comparative conclusions.

It is noted that this metric represents the cumulative transported mass across the internal reference plane, rather than the net domain exit emission to the ambient atmosphere. Therefore, absolute values may differ from true total production or boundary outflow. To facilitate inter-scenario comparisons and reduce sensitivity to boundary transport variability, the cumulative masses were further expressed as normalized values relative to the baseline case (0° slope, 0 m s⁻¹ wind) when appropriate.

2.7. Grid Resolution Determination

The grid resolution was determined using the characteristic flame diameter (D*):

$$D^{* }= ({\displaystyle \frac{Q}{{\rho }_{\infty }{c}_p{T}_{\infty }\sqrt{g}}}{)}^{{\displaystyle \frac{2}{5}}}$$

(4)

where Q is the total heat release rate (kW), ρ_∞ is the ambient air density (1.2 kg m⁻³), C_p is the specific heat of the air (1 kJ kg⁻¹ K⁻¹), T_∞ is the ambient temperature (293 K), and g is the gravitational acceleration (9.81 m s⁻²).

According to Sun et al. [42], the recommended mesh cell size should be between D*/12 and D*/8. On the basis of these criteria, in this study, the grid cell size was set between 0.15 m and 0.23 m. To test the stability and mesh sensitivity of the numerical results, two different mesh schemes (0.1 m and 0.2 m) were further tested. All the simulation scenarios maintain the same physical and boundary configurations to ensure the comparability of the results.

2.8. Statistical Analysis of the Fire Spread Rate

Coupled slope–wind effects were assessed using a quadratic response surface model:

$$rROS = a+b_1\theta +b_{2}V+b_{12}\theta V+b_{11}{\theta }^2+b_{22}{V}^2$$

(5)

where $\theta$ is the slope (or slope angle/gradient; º), $V$ is the wind speed, and $b_{12}$ represents the interaction term quantifying slope–wind coupling. Model coefficients were estimated by ordinary least squares. The significance of regression terms was evaluated using t-tests within the fitted model, and overall model adequacy was assessed using the F-test, R², and residual diagnostics. Statistical significance was defined as p < 0.05. All statistical analyses were conducted in Origin 2022 (OriginLab Corp., Northampton, MA, USA). The regression was based on the average rROS values from the three lateral measurement lines (Y = 2.5, 5.0, and 7.5 m) to reduce spatial variability.

3. Results

3.1. Grid Convergence and Spatial Sensitivity Analysis

To verify that the predicted energy release was not artifacts of mesh resolution, we first evaluated mesh sensitivity by repeating the baseline simulation using two grid resolutions for the core plot region (0.1 m and 0.2 m; mesh layout and domain setup were described in Section 2.3.1). The resulting heat release rate (HRR) time series was compared to assess the sensitivity of the simulation outputs to grid refinement.

As shown in Figure 3, the HRR curves obtained with 0.1 m and 0.2 m exhibit the same temporal pattern, with the rapid increase during the early burning stage followed by a quasi-steady period. In particular, HRR rapidly increases to approximately 50,000 kW within the first 200 s and then stabilizes. The differences between the two resolutions were minor in terms of peak HRR and the overall HRR evolution. Considering both computational accuracy and cost, the 0.2 m grid significantly reduces the computational load while maintaining accuracy comparable to that of the high-resolution results. Therefore, all subsequent simulations adopted this resolution to balance computational efficiency and numerical reliability.

To evaluate model performance, the simulated rate of spread (ROS) was benchmarked against previously published experimental data on pine litter fuel (Figure 4). First, the simulated ROS of Yunnan pine (P. yunnanensis) (0.003–0.027 m s⁻¹, slope 0–30°) was compared with laboratory observation data of litter from yellow pine (Pinus ponderosa) (0.002–0.042 m s⁻¹; [24]). Although the two studies involved different pine species, both fuel beds are dominated by pine needles and share broadly similar fuel bed structures (needle-dominated, shallow litter layer, and comparable fuel loading or bed depth as reported in the literature). Therefore, the results of Beaufait were used as an order-of-magnitude benchmark rather than a strictly property-matched validation. The simulation results generally fall within the range of reported experimental values, indicating that the model can reasonably reproduce the main physical processes of fire spread.

Furthermore, the model output was benchmarked against field observation data recently reported by Wang et al. [20]. This study conducted prescribed burning experiments at the same study site as the fuel sampling performed in this work, providing a consistent benchmark for model evaluation. Under a 15° slope, the model predicts an ROS of 0.00337 m s⁻¹, which is highly consistent with the measured range of 0.00280–0.00410 m s⁻¹ reported by Wang et al. [20]. This agreement suggests that the model captures fire spread behavior under field conditions for Yunnan pine litter within the reported experimental uncertainty. These results further suggest that the FDS model configuration based on site-specific fuel parameterization can reasonably reproduce the observed flame spread behavior of Yunnan pine litter within the reported experimental uncertainty. Notably, owing to safety constraints and research priorities, the fieldwork in this study was focused mainly on investigations of fuel characteristics and did not involve direct fire behavior measurements. Therefore, the model validation relies on experimental and field observation benchmarks provided by previous literature.

3.2. Effects of the Slope Angle on Fire Propagation and Thermal Dynamics

3.2.1. Rate of Spread

To quantify how downslope angle affects the rate of fire spread (ROS) under no-wind conditions, we analyzed a set of simulations spanning slope angles from 0° to 35°. ROS was estimated from thermocouple-derived flame-front arrival times on three lateral measurement lines (Y = 2.5, 5.0, and 7.5 m). Along each line, nine thermocouples were installed at 1 m spacing; thus, each colored dataset in Figure 5 contains nine points. The flame-front arrival time at each location was defined as the first time at which the measured temperature reached 300 °C. ROS was then calculated as the slope obtained from a linear regression of flame-front position versus arrival time.

The fitted slopes differ slightly among the three measurement lines (Figure 5), but the overall trend with slope is consistent. Specifically, as the downslope angle increases from 0° to 30°, ROS decreases markedly and then increases again at 35°, indicating a non-monotonic dependence on slope under no-wind conditions. Among the centerline (Y = 5.0 m), ROS decreases by approximately 91.5%, from 0.0272 m s⁻¹ on flat terrain (0°) to 0.0023 m s⁻¹ on a 30° slope (Figure 5b). Similarly low ROS values are observed on the two flank lines at 30° (Figure 5a,c). When the slope is further increased to 35°, the decreasing trend of the ROS reverses, and the propagation speed begins to increase.

Across the measured distance, the position–time relationships are generally well represented by a linear fit (most R² > 0.9), indicating that the spread can be treated as quasi-steady for estimating a representative ROS. The two flank lines show highly consistent fitted slopes, suggesting an approximately symmetric spread pattern. The centerline exhibits slightly larger variability (R² = 0.73–0.99; Figure 5b) while remaining acceptable for regression based on ROS estimation.

To facilitate visual comparison across slopes, ROS at each slope was normalized by the corresponding 0° value on the same measurement line to obtain a relative ROS (rROS) (Figure 6). At the centerline position (Figure 6b), rROS decreases from 0.194 at 10° to 0.124 at 15°. The rROS fluctuates slightly at intermediate slopes (0.131 and 0.122 at 20° and 25°), reaches its minimum value at 30° (0.084), and then increases to 0.115 at 35°. The spread dynamics on both flanks of the fire exhibit largely similar trends, with minor variations between them. On the left flank (Figure 6a), the rROS continuously decreases from 0.193 at a slope of 10° to a minimum of 0.063 at a slope of 30° and then only slightly rebounds at a slope of 35°. In contrast, the rROS fluctuations on the right (Figure 6c) are more pronounced: 0.186 at 10°, decreasing to 0.111 at 15°, increasing to 0.141 at 25°, decreasing to 0.067 at 30°, and increasing to 0.104 at 35°. Overall, a 30° slope consistently corresponds to the minimum rROS on the centerline and both flanks, indicating that downslope spread is most strongly reduced near this slope under the present no-wind conditions.

3.2.2. Thermal Dynamics

To characterize the impact of slope angle on downstream thermal behavior during no-wind downslope spread, we used two complementary diagnostics. First, time–distance temperature maps were constructed from the thermocouple array along the downslope direction (X-axis) to visualize the spatiotemporal evolution of thermal exposure (Figure 7). Second, temperature–time histories were analyzed at monitoring points located at X = 4.0 m across three lateral positions, namely the left flank (Y = 2.5 m), centerline (Y = 5.0 m), and right flank (Y = 7.5 m) (Figure 8). From each temperature–time curve, three timing metrics were extracted: (i) time to reach the peak temperature, (ii) duration of the peak temperature plateau, and (iii) total combustion duration at the monitoring location.

For gentle slopes (0–20°), the time–distance temperature maps (Figure 7a–d) show a single continuous high-temperature band advancing downslope over time and gradually weakening as it propagates. No temporally distinct secondary high-temperature zone was observed in the downstream region. Consistent with these maps, the temperature histories at X = 4.0 m (Figure 8) show that peak heating occurs relatively early, and the duration of high-temperature exposure generally shortens as slope increases within this group. For example, on the left flank (Figure 8a,b), the 0° case reaches approximately 1100 °C at 232 s and remains above 800 °C for 1373 s. At 10° and 20°, peak temperatures remain high (approximately 1200 °C), while the duration above 800 °C is notably shorter (approximately 600–800 s). Along the centerline (Figure 8c,d), the 0° case reaches approximately 1200 °C at 137 s and remains above 800 °C for about 1428 s. For 10° and 20°, peak temperature remains high (around 1000–1100 °C), and the time to peak increases compared with 0°. The right flank exhibited similar timing behavior to the left flank within this slope group (Figure 8e,f).

For steeper slopes (25–35°), the primary high-temperature band in the time–distance maps arrives later and persists over a longer time interval (Figure 7e–g). The 35° case additionally exhibits a late-time secondary high-temperature patch in the downstream zone (approximately X ≈ 7–9 m; Figure 7g). At X = 4.0 m, this shift to later and more persistent heating is reflected in the temperature histories (Figure 8). On the left flank (Figure 8a,b), peak temperatures for 30° and 35° decrease to approximately 700–900 °C and occur much later (approximately 3000–4000 s). Along the centerline (Figure 8c,d), peak temperatures are typically around 1000 °C, but the time to peak increases markedly and is followed by a faster decline; for 35°, a late-stage temperature rise is also observed. On the right flank (Figure 8e,f), peak temperatures likewise decrease and peak times are delayed, and the 35° case displays a pronounced late-time temperature increase, stronger than that on the left flank.

Across all conditions, three patterns can be summarized. First, increasing slope generally delays the occurrence of peak heating at the downstream monitoring location, while peak temperatures tend to decrease at steeper slopes (30–35°). Second, the centerline consistently exhibits the highest peak temperature and the longest duration of high temperatures relative to the two flanks. Third, steep slope cases, especially 35°, display late-time secondary heating signatures in the temperature histories at X = 4.0 m (Figure 8) together with late-time downstream high-temperature patches in the time–distance maps (Figure 7g), indicating a multi-stage thermal exposure history under the steepest slope condition.

3.3. Interactive Effects of Slope and Wind Speed on Fire Propagation Dynamics

To reveal the coupled effect of wind speed and slope on the normalized spread rate (rROS), we compared the rROS under different combinations of wind speed and slope angle and examined the response differences between the centerline and the two flanks (Figure 9). The results show that the wind–slope coupling effect exhibits notable nonlinearity and interdependence. Under windless conditions (V = 0 m s⁻¹), the rROS is affected mainly by the slope angle, reaching a peak of 0.19 at a slope of 10° and then decreasing to 0.06 at 30°. The introduction of wind significantly amplifies the spread of the fire and fundamentally alters the relationship between the rROS and slope. At a moderate wind speed of 1 m s⁻¹, the centerline rROS spikes to 0.571 at a 10° slope, approximately three times the peak value under calm conditions. When the wind speed increases to 2 m s⁻¹, the peak position of rROS shifts from 10° to 20°, with a peak value of 0.274, indicating that changes in wind speed can trigger the migration of the “most favorable slope”. The peak rROS exhibits non-monotonic patterns as the wind spread changes, indicating that the coupling mechanism between the wind and slope is not a simple linear superposition relationship. The wind also significantly alters the spatial geometry of the fire front. Under windless conditions, the rROS distribution on the fire front is approximately symmetrical; however, under windy conditions, its distribution is clearly asymmetric. Typically, the rROS on the centerline (Y = 5.0 m) is always greater than that on both sides (Y = 2.5 and 7.5 m). This difference is most significant when the slope is 10° and the wind speed is 1 m s⁻¹, eventually forming a distinct U-shaped fire front structure.

To quantify the complex slope ($\theta$) and wind velocity ($V$) coupling, a quadratic response surface model is developed on the basis of the mean rROS value:

$$\mathrm{rROS} = 0.163 - 0.0057\mathrm{\theta } + 0.0081\mathrm{V} + 0.0021\mathrm{\theta }\mathrm{V} + 0.00014{\mathrm{\theta }}^2 - 0.020{\mathrm{V}}^2$$

(6)

The regression coefficients in

Table 3. Regression coefficients and physical interpretation of the response surface model.

Term	Symbol	Coefficient	Interpretation
Slope (linear)	b1	−0.0057	An increasing slope has a primary suppressive effect on the spread rate.
Wind (linear)	b2	+0.0081	The wind is a dominant accelerating factor for flame propagation.
Slope (quadratic)	b11	+0.00014	U-shaped response to slope, with rROSmin occurring at moderate θ.
Wind (quadratic)	b22	−0.020	Diminishing accelerative effect of wind at high V.
Interaction	b12	+0.0021	Slope and wind interact synergistically to enhance fire spread.

provide quantitative support for the above observations. The linear terms b₁ and b₂ have opposite signs, indicating that the linear effects of slope and wind speed on the rROS are in opposite directions. The negative linear slope term (b₁ = −0.0057) quantifies the suppressive effect of the downslope gradient on the rROS, whereas the positive linear wind term (b₁ = +0.0081) reflects the accelerating effect of the wind speed. The quadratic terms reflect the nonlinear response of the rROS to slope and wind speed. The positive quadratic slope term (b₁₁ = +0.00014) corresponds to a U-shaped relationship, indicating that the effect of slope on the rROS is not monotonic: it weakens in the range of smaller slopes, whereas the intensity of the effect increases again under steep slope conditions. Conversely, the negative quadratic wind term (b₂₂ = −0.020) represents a diminishing wind speed effect, indicating that excessively strong winds reduce the propagation efficiency. This pattern explains why the peak rROS at 2 m s⁻¹ is lower than that observed at 1 m s⁻¹.

Notably, the positive interaction term (b₁₂ = +0.0021) quantifies the synergistic coupling effect between slope and wind speed. This finding indicates that under simultaneous moderate slope and moderate wind speed conditions, the enhancing effect on the rROS exceeds the linear superposition of the individual effects of the two. This synergistic mechanism explains the peak rROS observed under conditions of a 1 m s⁻¹ wind speed and a 10° slope. The model performance was evaluated using the coefficient of determination. The results showed R² = 0.96, indicating a high degree of consistency between the model predictions and the observed values and explaining 96% of the observed rROS variance.

Because slope changes significantly affect the fire spread rate, to fully represent the entire temperature evolution process, the time range (vertical axis) of each subplot in Figure 10 is set according to the duration of flame propagation under different slopes. Under steeper slopes, fire spreads more slowly and burns for a longer period, so the upper limit of the vertical axis is correspondingly higher (e.g., 30° and 35° slopes). Under gentler slope conditions, fire spreads faster, and the temperature rise and fall processes are relatively shorter, thus reducing the time range. This setting ensures that the high-temperature phase is fully captured under various operating conditions while avoiding comparison bias caused by differences in time scales.

Under windless conditions (Figure 10a–e), the high-temperature zone gradually narrows, and its duration decreases as the slope increases, especially when the slope exceeds 30°, indicating that the self-sustaining combustion capability of the flames is limited. Under a wind speed of 1 m s⁻¹ (Figure 10f–j), this trend partially reversed, with a wider and longer high-temperature zone appearing on all slopes, indicating that moderate airflow can enhance the ability of the fire to sustain itself. As the wind speed increases further to 2 m s⁻¹ (Figure 10k–o), the thermal structure shifts systematically to a narrower, more intense, and much shorter high-temperature zone. This behavior is consistent with a transition from a buoyancy-assisted regime to a wind-dominated forced convection regime primarily driven by wind speed, with slope gradient modulating the buoyancy contribution and the manifestation of the transition. At this stage, although stronger wind intensifies convective transport, it can also shorten the effective burning period, yielding a more intense but shorter combustion episode.

In summary, the simulation results indicate a wind speed-driven transition from buoyancy-assisted propagation to forced convection-dominated propagation, while slope and wind interact in both synergistic and competitive ways. At moderate wind speeds, the slope-induced buoyancy effect and the wind force act synergistically to support flame spread and sustain the high-temperature zone. When the wind speed increases further, the dominance of airflow is significantly enhanced, causing a change in the propagation mechanism and forming a new spread pattern, but this change does not necessarily lead to a higher propagation rate.

3.4. Combustion Characteristics and Gaseous Emissions

To quantify how slope and wind affect combustion products and burning performance during downslope spread, we analyzed the time series of the CO₂ mass flux under different combinations of slope and wind speed (Figure 11). In addition, cumulative CO₂ and CO masses were calculated over a consistent burning period (defined by the termination criterion in Section 2.5) and are reported in Figure 12. Combustion efficiency was evaluated separately using the apparent efficiency η based on the time-integrated HRR relative to $m_f\Delta H_c$ (Figure 13).

Under windless conditions (Figure 11a), CO₂ mass flux exhibits a non-monotonic response to slope. On a flat surface (0°), the CO₂ release rate from the flames rapidly reaches a brief peak of approximately 0.16 kg s⁻¹. As the slope increases to a moderate range (15–20°), the peak flux increases to about 0.27 kg s⁻¹, and the peak occurrence time is delayed, reflecting enhanced fuel-bed preheating and increased combustion intensity. However, when the slope increases to 35°, the combustion behavior changes. The peak intensity decreases, while the combustion duration becomes considerably longer, with CO₂ release persisting for more than 10,000 s. This indicates a shift from short-duration, higher-intensity burning to a more persistent and lower-intensity combustion mode on steep slopes under calm conditions.

The effect of wind significantly alters the aforementioned dynamic processes (Figure 11b,c). At 1 m s⁻¹ (Figure 11b), the CO₂ flux peaks are generally higher and occur earlier than in the windless cases, indicating accelerated combustion and faster flame development. When the wind speed increases to 2 m s⁻¹ (Figure 11c), the effect is further amplified, manifesting as a combustion event of extremely high intensity but shorter duration. Under high wind speeds, a sharp but brief peak in CO₂ release typically occurs, indicating an accelerated fuel consumption process. This contrasts sharply with the persistent, low-intensity combustion observed under calm conditions on steep slopes.

Figure 12 summarizes the cumulative CO₂ and CO emissions. In this study, cumulative CO₂ is interpreted as a proxy for the overall extent of fuel oxidation (i.e., an indicator of the oxidized fraction of fuel consumption), whereas cumulative CO indicates incomplete combustion tendency. The cumulative responses are nonlinear because they integrate both peak intensity and burn duration. Under windless conditions (0 m s⁻¹), total CO₂ generally increases with slope and reaches a maximum of 717.5 kg at 35°, consistent with the extended low-level combustion observed in Figure 11a. Under windy conditions, total CO₂ on steep slopes tends to decrease (e.g., at 35°: 717.5 kg at 0 m s⁻¹ versus 616.5 kg at 2 m s⁻¹), which aligns with a shorter burning period resulting from faster flame passage. A local exception occurs at 30°, where total CO₂ increases to 634.0 kg at 2 m s⁻¹, implying that certain slope and wind combinations can enhance overall oxidation even as the combustion period becomes more concentrated.

Total CO emissions range from 0.87 kg to 2.23 kg (Figure 12), demonstrating that combustion completeness varies substantially across scenarios. The highest CO (2.23 kg) occurs at a 35° slope with a wind speed of 1 m s⁻¹, indicating the strongest incomplete combustion tendency among the tested cases. At a wind speed of 1 m s⁻¹, CO emissions gradually increase with increasing slope, suggesting that steep downslope burning under light wind can favor conditions that promote CO formation, such as locally limited mixing or oxygen availability relative to pyrolysis product release. Importantly, CO is not used here as a direct measure of combustion efficiency, because CO and CO₂ totals can also reflect differences in burn duration and total oxidation history. Instead, combustion efficiency is quantified explicitly by η (Figure 13). Consequently, conditions associated with high cumulative CO₂ do not necessarily coincide with low η, implying that fuel consumption and combustion quality can respond differently to the coupled effects of slope and wind. Specifically, η varies from 58.6% (10°, 1 m s⁻¹) to 73.6% (35°, 1 m s⁻¹), with near constant values around 58.6% at 0–10° and markedly higher efficiencies (often >70%) at 25–35° depending on wind speed (Figure 13).

4. Discussion

Existing studies have demonstrated a significant coupling between slope and wind speed, which can alter fire front morphology and modify spread behavior. However, at the emission level, the release process, cumulative emissions, and their relationship with the spread rate of CO₂/CO under different slope–wind combinations remain insufficient. To address this gap, this study employs high-fidelity numerical simulations to identify response modes of downslope fire propagation under different slope and wind speed conditions, elucidate the underlying heat transfer and flow drivers, and assess the sensitivity of CO₂/CO emissions to these mechanism shifts. The following discussion will focus on (1) the non-monotonic slope response of downslope spread and its risk implications; (2) the shift in the slope–wind coupling from synergy to competition; and (3) the decoupling of carbon emissions and the rate of carbon spread and its implications for carbon accounting and management.

4.1. Non-Monotonic Downslope Spread with Slope and Its Risk Implications

The no-wind simulations reveal a non-monotonic dependence of downslope ROS on slope angle. ROS decreases substantially from 0° to approximately 30° and then exhibits a modest rebound at 35° (Figure 5 and Figure 6). Such an “inhibition–recovery” pattern is difficult to represent using slope–ROS descriptions that impose a single monotonic dependence, which are commonly adopted in empirical or statistical studies of slope effects and in operational downslope correction factors [43,44,45]. The present results do not contradict those prior approaches; rather, they suggest that a monotonic trend may hold over a limited slope range, while an additional regime may become relevant for very steep negative slopes under the specific no-wind configuration studied here.

A CFD-consistent interpretation is that increasing downslope angle alters the near-surface heat feedback history experienced by unburned fuel ahead of the spreading front. From 0° to 30°, the time–distance temperature maps show that the primary high-temperature band reaches downstream locations later and becomes weaker near the surface (Figure 7a–f), and the temperature histories at X = 4.0 m show delayed peaks and reduced persistence of high-temperature exposure (Figure 8). These signatures indicate reduced and delayed thermal exposure at downstream locations, consistent with the marked reduction in ROS over the same slope range. In slope fuel spread studies, the interaction among buoyancy-driven flow, entrainment, and convection can change flame structure and redistribute heat transfer to unburned fuel [45,46]. Therefore, even under no wind, changes in slope can reorganize the flow flame configuration in ways that modify the effectiveness and timing of preheating.

At the steepest slope (35°), the thermal diagnostics indicate a qualitative change; a late-time secondary heating feature emerges. Figure 7g shows a temporally separated high-temperature patch in the downstream zone (approximately X ≈ 7–9 m), and Figure 8 shows a distinct late-stage temperature rise at X = 4.0 m. In CFD terms, this suggests that the near-surface region may experience multi-stage thermal exposure at 35°, meaning that heating is renewed after the initial passage of the primary heating episode. A physically plausible explanation is that very steep slopes modify the fire-induced near-surface flow and hot gas transport pathways, intermittently increasing the residence time and near-surface impingement of heated gases over the fuel bed, thereby partially restoring effective preheating. Prior work has reported that terrain slope can modify fire-induced flow and near-surface wind enhancement [47], supporting the plausibility of flow reorganization at steep slopes.

These findings have implications for risk assessment on steep downslope terrain. Under the present no-wind setting, the minimum ROS occurs near 25–30°, but a further increase to 35° coincides with renewed heating signatures and partial recovery of ROS. This suggests that steep downslope areas should not automatically be treated as progressively lower spread potential with increasing slope, and it motivates caution when applying monotonic downslope correction factors for hazard zoning [48]. Recent numerical work also suggests that slope effects on ROS can be nonlinear and scenario dependent [49], and the present results provide an example in which changes in the structure of thermal exposure (single-stage vs. multi-stage) accompany a non-monotonic ROS trend.

4.1.1. Shift in the Slope–Wind Coupling from Synergy to Competition

Slope and wind force regulate the relative strength between the buoyancy of the fire plume and the near-surface inflow/slope flow, thereby controlling whether the flame adheres to the fuel bed and the efficiency of heat feedback. When the momentum of both is equal, the flames tend to adhere to the ground surface, enhancing convective heat transfer and accelerating its spread [50,51]. If the external momentum is too strong or the terrain constraints are weakened, the flames rise, radiative heat becomes dominant, and the heat feedback efficiency decreases accordingly [52,53]. In this study, at low wind speeds (1 m s⁻¹), slope and wind force exhibit a synergistic effect, enhancing flame adhesion and improving convective heat transfer efficiency [54]. When the wind speed increases to 2 m s⁻¹, the relationship between the two becomes competitive, flame adhesion weakens, heat concentration decreases, and the fire spread rate slows [55,56]. The unified perspective of “buoyancy–momentum balance” explains why wind–slope coupling exhibits a shift between cooperation and competition across different wind speed ranges, providing a comparable physical basis for understanding the nonlinearity of coupling effects.

The relative relationship between slope and wind direction further alters the velocity field structure, thereby reshaping the geometry of the fire line and the differences in lateral propagation. Unlike the inverted V-shaped fire line geometry commonly seen in uphill experiments [31,57,58,59,60], the fire line in this study extends downward as a whole and is distributed in a slightly asymmetrical manner. Furthermore, the spread between the central and lateral regions shows notable differences, suggesting that there is spatial differentiation in heat transfer paths and combustion intensity at different locations. Because this study did not modify the model, the results mainly reflect the distribution of the velocity field during downhill propagation rather than empirically adjusting the wind–slope superposition parameters [26]. Without relying on empirical corrections, this study provides evidence of fire line morphology and lateral non-uniform propagation under downslope downwind conditions, offering a new reference for distinguishing uphill and downhill propagation mechanisms and improving the modeling assumptions of coupling terms.

4.1.2. Decoupling of CO₂ Emissions from the ROS and Temperature

Our simulation results indicate that, in steep downslope terrain, cumulative CO₂ emissions may be decoupled from ROS and near-surface temperature. This decoupling implies that emissions inference based on “fire intensity as a proxy” is unreliable in this case. Under a 35° slope and windless conditions, although the spread rate only slightly increases and the peak temperature even decreases slightly, the cumulative CO₂ emissions reach 717.5 kg, the maximum among all simulated scenarios. This result differs from the empirical assumption based on experiments on flat or uphill terrain that higher fire intensity (usually measured by the ROS) corresponds to higher carbon emissions [61]. However, this finding aligns with the understanding of combustion mechanisms: total carbon emissions are more directly controlled by fuel combustion efficiency rather than by instantaneous propagation intensity [62]. Therefore, in steep downhill scenarios, ROS or temperature is difficult to use as a stable proxy for cumulative emissions.

From a mechanistic perspective, this decoupling can be explained by more thorough mixing and more complete oxidation processes, thus making “fuel consumption × combustion efficiency” a more suitable emission characterization framework. The buoyancy-induced vortex structure and enhanced turbulence in steep downslope areas may promote the mixing of air, combustible gas, and solid fuel. Therefore, even if ROS and temperature do not increase synchronously, high cumulative CO₂ emissions may still be generated. Similar evidence that emissions are primarily controlled by fuel and oxidation efficiency can also be found in peat fire studies, which indicate that emission levels are mainly determined by fuel carbon content and oxidation efficiency [63]. Meanwhile, analysis of the 2023 Canadian wildfires also indicates that the record carbon emissions were attributable to the cumulative effect of continued burning (including smouldering/residual burning) under drought conditions rather than the high-intensity flaming front [12]. Furthermore, studies on pine needle layers have shown that increased fuel loading often leads to increased total heat release, although the heat intensity per unit area does not increase accordingly [64]. This result suggests that current emission inventory models, which generally rely on linear estimations of fire intensity [65] or burned area [66,67,68], may be biased under conditions of slope or significant fuel stratification. In summary, in the context of hillside fires, emission estimations should shift from being based solely on propagation intensity or burned area to a process-based framework that explicitly characterizes fuel consumption and combustion efficiency.

4.2. Management Implications and Practical Applications

Based on the U-shaped spread relationship of the simulation results, the terrain has two sensitive zones for the acceleration of fires. This pattern suggests that differentiated management should be implemented for gentle and steep slope areas in the risk assessment and early warning system. Optimizing terrain classification thresholds and strengthening monitoring within key slope ranges can improve the accuracy of risk identification and the efficiency of emergency response. Meanwhile, emission assessment models also need further improvement. Relying solely on the rate of spread (ROS) is insufficient to reflect the differences in combustion efficiency under mountainous conditions. Incorporating topographical factors into the inventory model can improve the accuracy of regional carbon emission accounting and climate feedback simulation.

Under downwind and downslope conditions, gentle slopes (approximately 10–25°) are more likely to maintain a stable and uniform combustion process, making them advantageous sections for achieving fuel reduction and safety control during prescribed burning. When the slope and wind speed increase, the concentration of fire energy increases, the response of the fire line accelerates, the safe window for operation shrinks, and the difficulty of control increases significantly. This trend suggests that combustion schemes should strike a balance between fire efficiency and safety boundaries, and the appropriate ignition window should be determined by rationally selecting slope sections and wind speed limits, thereby achieving complete combustion and operational stability under safety constraints.

However, management challenges are not limited to operational control. The study results show that even low- to moderate-intensity downhill burning can produce relatively high levels of carbon dioxide emissions, forcing regulators to weigh fuel reduction against carbon emission targets [13,69]. Furthermore, the decoupling of emissions from fire line temperatures further suggests that the ecological impact may be greater than expected. Even if the fire progresses slowly, when the combustion efficiency is high, the accumulation of underground heat can still cause overheating of deep soil, damaging the root system and microbial community of temperate pine forests [70]. Therefore, constructing a comprehensive decision-making framework that accounts for fire behavior characteristics, emission effects, and ecological feedback is a key approach to achieving both the safety and ecological sustainability of planned incineration.

4.3. Study Limitations and Future Directions

At the model level, although this study has high fidelity, it is still limited by idealized assumptions, such as a homogeneous fuel layer and regularized canopy structure. In actual Pinus yunnanensis forests, surface and near-surface fuels often display pronounced spatial heterogeneity (local accumulation, gaps, and moisture variability) and vertical stratification (needle litter, duff, and shrub branches). This heterogeneity can modify near-surface ventilation and convective radiative feedback and alter the propensity for flame attachment, which in turn may shift the balance between buoyancy-driven spread and wind or slope-driven spread.

In the future, efforts should be made to promote the coupling of high-fidelity models with mesoscale atmospheric models (such as WRF-Fire) and integrate heterogeneous fuel data [60]. The role of fuel characteristics in fire control has not been fully realized. The fuel moisture content (FMC) and fuel bed thickness are both critical threshold parameters that determine the persistence and spread rate of fires [71]. Current models lack characterization of these fuel-thermal coupling processes, which may underestimate the fire potential in steep slope areas. Furthermore, this study only considers CO₂ and CO emissions and does not cover key components, such as CH₄ and aerosols, which limits the comprehensive assessment of atmospheric chemistry and environmental effects.

Finding a balance between computationally expensive CFD models and oversimplified operational models remains a core challenge in fire science. Recent advances in medium-complexity models (such as QUIC-Fire, [72]) offer promising pathways to address this problem, although their effectiveness depends on systematic validation with high-quality experimental data, and data acquisition itself remains difficult [73,74]. Future research urgently needs interdisciplinary collaboration to combine coupled atmospheric–fire simulation with prescribed burning at the field scale. Comprehensive studies in representative ecosystems, such as Asian mountain forests, can be used to calibrate and refine the physical processes in the model and reveal the applicability of the downslope fire dynamics identified in this paper across different climate and topographical contexts. Additionally, high-resolution remote sensing technologies such as LiDAR can be introduced to acquire three-dimensional fuel structure, canopy parameters, and micro topographic information [75]. This information can then be compared with or assimilated from CFD flow field and heat release rate results to improve the reliability and spatial consistency of model input parameterization. These efforts will provide a scientific basis for developing integrated management strategies and promote the synergistic improvement of disaster risk reduction, emission control and ecological resilience.

5. Conclusions

This study used physics-based fire dynamics simulations to assess the coupled effects of slope and wind speed on fire behavior and gas emissions during prescribed fires involving surface combustibles in Yunnan pine forests. The results indicate that under mountainous conditions, commonly used propagation or intensity indicators (such as ROS and near-surface temperature) cannot stably represent cumulative CO₂ emissions, and the intensity–emission relationship may deviate in specific scenarios. Therefore, simplifying emissions to a linear function of intensity or burn area introduces uncertainty, while explaining emission differences in terms of the fuel consumption process and combustion efficiency produces more consistent results.

At the carbon inventory model level, the above understanding has direct significance for the accounting of emissions from the controlled combustion of surface fuels. For coniferous forest areas with significant topographic relief, emission estimations should shift from the approach of “area-driven+fixed emission factor/intensity proxy” to a more advanced approach of parameterizing process quantities that can characterize “litter fuel consumption × combustion efficiency (including open flame and residual combustion stage structure)” to reduce the systematic bias caused by the slope–wind field coupling and improve scenario extrapolation capabilities. Future research should focus on translating the interaction between slope and wind speed into stratified schemes or correction terms that can be used in inventory models (e.g., adjusting the proportion of litter consumption, residual combustion percentage, and emission components according to the slope range and wind speed level) and combining this with field observations to constrain fuel consumption, combustion duration, and residual combustion contribution, thereby supporting comparable accounting and uncertainty assessments of regulated combustion emissions in mountainous areas.