the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Jul 2025
Dynamical Linkages Between Planetary Boundary Layer Schemes and Wildfire Spread Processes
Abstract. Wildfires can significantly enhance surface sensible heat and modify the state of the near-surface atmosphere, becoming key factors in triggering turbulence and restructuring the boundary layer. This study uses high-resolution simulations with WRF-Fire, combined with hourly observational data from six meteorological stations (five national and one emergency stations) during the Jinyun Mountain wildfire in Chongqing, China, to systematically evaluate the performance of five planetary boundary layer (PBL) schemes (MYJ, MYNN2, MYNN3, BouLac, UW) in simulating temperature, wind speed, and turbulence intensity. Results show that all schemes can reproduce the diurnal trends of temperature and wind speed but exhibit significant differences in amplitude response and simulation errors. The MYNN3 scheme captures the spatiotemporal variations of turbulence intensity and wind speed at different heights more accurately, thereby better representing the 2-meter temperature and 10-meter wind speed response and reduces the model’s cold bias in high-temperature simulations. The BouLac and MYNN2 schemes also show some response to thermal disturbances at certain sites but perform poorly under strong perturbations and exhibit large fluctuations. The MYJ and UW schemes show overall weaker turbulence and fail to capture local circulation variations effectively. Turbulent energy budget analysis of MYNN3 indicates a buoyancy-dominated turbulence generation mechanism, with vertical transport promoting upper-level disturbances. This reveals MYNN3’s greater sensitivity to wildfire thermal perturbations and more complete feedback processes, providing a scientific basis for selecting PBL schemes in mountainous wildfire simulations.
- Preprint
(5979 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
CEC1: 'Comment on egusphere-2025-3072', Astrid Kerkweg, 30 Jul 2025
-
AC1: 'Reply on CEC1', Yongli Wang, 18 Jan 2026
Thanks for your kindly suggestion, The new title is "Dynamical Linkages Between Planetary Boundary Layer Schemes and Wildfire Spread Processes: a case study using WRF-Fire version 4.6"
Citation: https://doi.org/10.5194/egusphere-2025-3072-AC1
-
AC1: 'Reply on CEC1', Yongli Wang, 18 Jan 2026
-
RC1: 'Comment on egusphere-2025-3072', Anonymous Referee #1, 16 Nov 2025
This manuscript evaluates five PBL schemes within the WRF-Fire coupled atmosphere-fire model using simulations of a wildfire case in a mountainous region in China. The study evaluates the simulations using near-surface variables (temperature at 2 m and wind at 10 m), and vertical turbulence structures (TKE and PBL height). The manuscript is well written, the figures are high quality, and there's scientific value in the analysis. However, the manuscript doesn't seem to include a model development component, so it may be best to transfer to another journal.
Issues to be addressed:
It is not entirely clear what is the role of including the fire component in these simulations, given that fire spread was not compared or discussed. It seems the manuscript's primary goal is to evaluate the PBL schemes under “stressed” conditions, which happens to be fire heat. The manuscript does not compare the effect of the different PBL in the resulting fire spread, and does not quantify the magnitude of the fire perturbation to the PBL (e.g. TKE in coupled simulation minus TKE in uncoupled simulation). A revised analysis including results of the simulated fire would be a substantial contribution to the existing coupled fire modeling literature.
Comparing the PBL schemes and their representation of the processes during a fire provides helpful guidance on the characteristics of these schemes under non-typical conditions. However, fire-induced processes occur at the microscale, where turbulent eddies propagate in the 3-D space, which are not represented by a PBL scheme. Most, if not all, of these schemes are 1-D. There are no fluxes outside of the vertical column over each grid point. Hence, it's expected that all PBL schemes will show a poor representation of the atmospheric state in the vicinity of the fire, especially at a spatial resolution that can be resolved by LES. That's also the case in regions of complex topography.
The grid spacing of 300 is arguably too fine for using a PBL scheme, especially with fire-induced perturbations. It would be more reasonable to simulate the innermost domain using LES, with the PBL scheme in d02 providing the boundary conditions to d03, which is the typical configuration used with WRF-Fire. It's also important to consider the vertical resolution: with WRF's default configuration used in this study (i.e. CONUS), the 1st model level represents a layer starting at the surface to approx. 50 m AGL. This is an important aspect of the model configuration considering the fire heat is at the 1st few meters above the surface. A vertical layer of 50 m cannot properly represent the actual intensity of the heat and the magnitude of the perturbation it produces.
The current manuscript does not discuss these caveats, which are important aspects to be considered when simulating the atmospheric processes induced by the fire and driving the fire spread. From the perspective of someone interested in simulations that represent the fire-atmosphere coupled processes and their impact on the chaotic nature of fire spread, the model configuration used in the manuscript would be hard to justify. However, LES simulations (including LES nests within WRF) are computationally expensive, and perhaps there are other use cases where microscale processes may not be so important. Since this manuscript is not necessarily interested in addressing the typical challenges associated with WRF-Fire, I suggest the authors strengthen the motivation and potential application of this study, along with caveats in the model configuration.
The authors' motivation for using a PBL parameterization in a 300 m domain, along with a 50-m vertical layer as the 1st model level must be clearly stated. I strongly recommend they include a simulation using one of WRF's LES options in this analysis, which can then be configured with the 1st model level closer to the ground. This will provide a better understanding of the effects of using a PBL scheme instead of LES, and whether this is a critical aspect to be accounted for in these simulations. It would be desirable and insightful (but not critical) to also include one of the 3-D PBL options, which can be configured following the other PBL schemes.
Their comparison with observations is based on 2-m temperature and 10-m winds. However, the simulations' temperature and winds represent a 0-50 m deep layer. Although WRF's atmosphere and land surface are coupled, the fire heat is only coupled to the atmosphere. This means the soil temperature can only increase from the increased air temperature, not from the fire directly. This leads to more inertia in the heat transfer and underrepresented heat gradients. Considering that T2 is a diagnostic variable derived from T, and T represents a 50-m deep layer, I'd expect the fire feedback to be substantially diluted. The same applies to 10-m winds. These variables are interpolated to 2 and 10 m, they are not resolved by the model at these scales. For this reason, I suggest including a control simulation with fire feedback turned off (or just no fire at all). This will show the magnitude of the fire perturbations affecting the PBL schemes. They may be quite minor. In addition, this would allow the authors to compare how each scheme transports the fire heat, i.e. by looking at the temperature or energy difference between the control and the corresponding fire simulation for each PBL scheme.
Lastly, the CONUS namelist does not include the fire configuration. Please include the fire configuration, including ignition location and any other entry under “&fire” that is different from the model default.
*It's worth noting that the official release of the WRF model only allows a meso-LES configuration with the YSU scheme. Yet, it is possible to make it work with other schemes with minimal code modification: the variables in the registry associated with the package pbl=0 must include the same variables in the associated pbl scheme.
Citation: https://doi.org/10.5194/egusphere-2025-3072-RC1 -
AC2: 'Reply on RC1', Yongli Wang, 18 Jan 2026
This manuscript evaluates five PBL schemes within the WRF-Fire coupled atmosphere-fire model using simulations of a wildfire case in a mountainous region in China. The study evaluates the simulations using near-surface variables (temperature at 2 m and wind at 10 m), and vertical turbulence structures (TKE and PBL height). The manuscript is well written, the figures are high quality, and there's scientific value in the analysis. However, the manuscript doesn't seem to include a model development component, so it may be best to transfer to another journal.
We thank the reviewer for the suggestion. The primary objective of this study is to evaluate the applicability of the coupled atmosphere-fire model in complex forested terrain. This evaluation relies heavily on hourly data from automatic weather stations located in the immediate vicinity of the wildfire, including one emergency mobile station (covering two positions) and five automatic observation stations (comprising two upwind stations, two downwind stations, and one high-altitude station). Notably, the emergency mobile station operated within a 3 km range; the nearest automatic station (Chaoyang) is less than 3 km in linear distance from the emergency mobile station, and the farthest station (Batang) is less than 9 km away. Consequently, these observational data provide a robust basis for evaluating the forest fire simulation results. The value of this study lies in the application of rare observational data acquired near the wildfire region and the assessment of how heat released by the fire impacts local circulation, providing significant scientific value for the future predictive application of coupled models.
Issues to be addressed:
- It is not entirely clear what is the role of including the fire component in these simulations, given that fire spread was not compared or discussed. It seems the manuscript's primary goal is to evaluate the PBL schemes under “stressed” conditions, which happens to be fire heat. The manuscript does not compare the effect of the different PBL in the resulting fire spread, and does not quantify the magnitude of the fire perturbation to the PBL (e.g. TKE in coupled simulation minus TKE in uncoupled simulation). A revised analysis including results of the simulated fire would be a substantial contribution to the existing coupled fire modeling literature.
We thank the reviewer for pointing out the important issue regarding the role of the fire module and the depth of the fire–boundary layer coupling analysis. We have supplemented and clarified this issue as follows:
First, the purpose of this study is to evaluate the applicability and usability of the coupled atmosphere-fire model WRF-Fire v4.6 in complex terrain regions. Currently, evaluations of this coupled model in complex mountainous areas are scarce. We utilize the strong, spatially heterogeneous thermal forcing generated by a real forest fire event as the physical background to test the differences in performance of different PBL schemes under typical, strongly perturbed conditions. Compared to idealized heating or no-fire scenarios, WRF-Fire provides a more realistic spatiotemporal distribution of heat release, making the response differences of various PBL schemes under the same external fire forcing comparable and practically significant.
Regarding the issue that fire behavior is not directly shown in the original manuscript, the revised manuscript has added the burned area results simulated by WRF-Fire and performed a quantitative comparison with the MODIS burned area product (Figure 3). This analysis shows that the spatial scale of the simulated fire is consistent with observations, thereby verifying the rationality of the fire–atmosphere coupled simulation at the regional scale and avoiding the treatment of the fire module merely as an "idealized heat source." The simulated burned area varied considerably (47.38%–92.82%) across the five boundary layer schemes tested (MYJ, MYNN2, MYNN3, BouLac, and UW). The MYNN3 scheme demonstrated superior performance, providing the most accurate estimate.
Regarding quantifying the degree of fire perturbation on the PBL (such as the difference between fire-on and fire-off simulations), we agree that this analysis is physically meaningful. In fact, the authors have systematically carried out fire-on/fire-off control experiments in previous research targeting fires in the forest-grassland transition zone (see: https://www.mdpi.com/2571-6255/6/11/443), evaluating the impact of atmosphere-fire interactions on wildfire behavior under both coupled and uncoupled experimental settings. The focus of the current study is specifically to compare the differences in response modes of various PBL schemes under identical fire thermal forcing conditions, rather than quantifying the absolute intensity of the fire perturbation.
In summary, by introducing a real forest fire process, combining near-fire station observational data, and utilizing MODIS burned area to evaluate simulation results, this study clearly demonstrates the physical characteristics of performance differences among PBL schemes under fire–atmosphere coupled conditions. This work not only distinguishes itself from traditional PBL comparison studies under no-fire backgrounds but also provides new empirical evidence for the applicability assessment of coupled fire–atmosphere models through the rare observed hourly data under complex underlying surfaces and strong thermal perturbation conditions.
- Comparing the PBL schemes and their representation of the processes during a fire provides helpful guidance on the characteristics of these schemes under non-typical conditions. However, fire-induced processes occur at the microscale, where turbulent eddies propagate in the 3-D space, which are not represented by a PBL scheme. Most, if not all, of these schemes are 1-D. There are no fluxes outside of the vertical column over each grid point. Hence, it's expected that all PBL schemes will show a poor representation of the atmospheric state in the vicinity of the fire, especially at a spatial resolution that can be resolved by LES. That's also the case in regions of complex topography.
We thank the reviewer for this critical comment regarding fire-induced microscale processes and the applicability of PBL parameterizations. We agree that fire-induced turbulence essentially possesses distinct 3D characteristics and occurs within the microscale range; such processes are not the design target of traditional PBL schemes. Therefore, near the fire burning region especially under complex terrain conditions, PBL schemes have inherent limitations in characterizing the local atmospheric state.
It should be noted that although the PBL schemes used here are theoretically 1D vertical parameterizations, we enabled diff_opt=2 and km_opt=4 in the model configuration. This allows turbulent diffusion in both horizontal and vertical directions to participate in the numerical evolution. This setting enhances the model's ability to dissipate and adjust sub-grid scale perturbations to some extent, ensuring that fire-induced thermal perturbations are not entirely confined to a single-column structure but can propagate limitedly at the grid scale.
According to the suggestion, by using the default model configuration with 45 vertical layers and three nested domains (7500 m, 1500 m, and 300 m; Rothermel fire grid: 30 m), we had tried time steps of 45 s, 15 s, and 10 s, finding that only the 10 s time step allowed the simulation to complete the full 145-hour duration; we subsequently attempted a three-domain nested simulation using WRF-LES for the innermost domain (increasing vertical layers to 65 to meet the minimum requirement of 60 layers for WRF-LES) with the same horizontal resolutions and tested time steps of 10 s, 5 s, 2 s, and 1 s, but the results showed that the WRF-LES configuration is unstable at every time step for this 145-hour event, and since the instability in the coupled atmosphere-fire model cannot be resolved in the short term, the PBL scheme remains the currently feasible approach despite its limitations. Furthermore, at a 2 s time step, the ratio of computation time to simulation time exceeded 5:1 (5 minutes of runtime for 1 minute of simulation), which is unacceptable for operational fire spread forecasting; in future work, the authors plan to conduct WRF-LES simulations in flat terrain and smaller area to analyze microscale fire-induced processes and the impact of 3D turbulence propagation on fire spread.
Based on the reasons above, this study focuses on comparing the relative response characteristics of multiple PBL schemes under identical fire-induced thermal forcing conditions, provided that stable operation is ensured. This choice does not imply that PBL schemes can fully describe fire-induced microscale turbulence; rather, it is a compromise solution that is feasible and reproducible under current resolution and computational conditions. We have explicitly pointed out this limitation in the discussion section of the revised manuscript and emphasized that the conclusions should be understood as a comparison of PBL parameterization behaviors, not a direct resolution of the fire ground's microscale turbulent structure.
- The grid spacing of 300 is arguably too fine for using a PBL scheme, especially with fire-induced perturbations. It would be more reasonable to simulate the innermost domain using LES, with the PBL scheme in d02 providing the boundary conditions to d03, which is the typical configuration used with WRF-Fire. It's also important to consider the vertical resolution: with WRF's default configuration used in this study (i.e. CONUS), the 1st model level represents a layer starting at the surface to approx. 50 m AGL. This is an important aspect of the model configuration considering the fire heat is at the 1st few meters above the surface. A vertical layer of 50 m cannot properly represent the actual intensity of the heat and the magnitude of the perturbation it produces.
We thank the reviewer for the suggestion. As mentioned previously, for configurations where d01 and d02 use the boundary layer scheme and d03 uses the LES scheme, the number of vertical layers must be set to 60 or more to ensure proper simulation. In our experiments, we employed vertical layer counts of 75 and 65, with the first atmospheric layer set within 10 meters of the ground; however, the coupled model could not run normally. When the vertical layer count was set to 60 with d03 using a PBL scheme, the simulation was also unstable. Currently, the atmosphere-fire coupled model does not support restart simulations, meaning we also cannot resume from a breakpoint.
The height of the fire front is closely related to the height of the fuel and the thickness of the first atmospheric layer. In this study, forest trees account for over 98% of the fuel. In WRF-Fire, the Rothermel forest fire spread model utilizes the synthetic wind method proposed by Albini et al., which synthesizes wind and slope effects into an "effective wind." This effective wind is then substituted into the Rothermel formula using a wind speed height of 6.096 meters. This approach maintains the form of the original formula while allowing application in complex terrain and non-heading wind situations. Considering the actual circumstances of forest fire occurrence, a reasonable layer thickness would be 10–20 meters. However, due to the excessive heat released by the fire, the calculation of the coupled model becomes unstable. Therefore, the coupled model cannot reproduce this forest fire spread process when the model layer height is set too low.
- The current manuscript does not discuss these caveats, which are important aspects to be considered when simulating the atmospheric processes induced by the fire and driving the fire spread. From the perspective of someone interested in simulations that represent the fire-atmosphere coupled processes and their impact on the chaotic nature of fire spread, the model configuration used in the manuscript would be hard to justify. However, LES simulations (including LES nests within WRF) are computationally expensive, and perhaps there are other use cases where microscale processes may not be so important. Since this manuscript is not necessarily interested in addressing the typical challenges associated with WRF-Fire, I suggest the authorsstrengthen the motivation and potential application of this study, along with caveats in the model configuration.
We thank the reviewer for the in-depth comments regarding the innermost grid resolution, the applicability of PBL versus LES, and the vertical resolution issues. We agree that this issue is one of the key technical challenges when simulating fire–atmosphere coupling processes. In response to the previous and above comments, our response is as follows:
First, we acknowledge the reviewer's judgment, the 300 m horizontal resolution lies within the "gray zone" (terra incognita) between traditional PBL parameterization and LES. In the presence of strong fire-induced thermal perturbations, employing PBL schemes inevitably has physical limitations. At the same time, the first layer thickness of approximately 50 m is indeed difficult to resolve the strong heat release and vertical gradients occurring within the few meters near the ground where the fire front exists. These issues have been explicitly discussed as important limitations of the model configuration in the revised manuscript.
Regarding the choice of methodology, we attempted to use a nested LES for the innermost domain following the typical WRF-Fire configuration (with the PBL scheme of d02 providing boundary conditions for d03). However, under the combined effects of complex mountainous terrain, non-stationary fire heat release, and the continuous multi-day simulation period involved in this study, the LES experiments faced significant difficulties in terms of numerical stability and computational cost. Actual experiments showed that multiple sets of LES simulations struggled to run stably for more than about one day, and the computational cost made it unfeasible to conduct a systematic comparison of multiple PBL/LES schemes under current computing conditions. Therefore, LES is not selected as the primary simulation framework for this paper.
Based on the practical constraints, this paper adopts an innermost resolution of approximately 300 m combined with PBL parameterization schemes, aiming to achieve a compromise configuration that is numerically stable, computationally reproducible, and capable of covering the complete fire process. It needs to be emphasized that this study does not attempt to resolve fire-induced microscale turbulent structures or directly characterize the chaotic nature of fire spread but rather focuses on the relative response differences of different PBL schemes regarding near-surface wind fields, turbulence intensity, and boundary layer structures under the same fire heat forcing background. Under this research objective, PBL schemes still provide a comparable diagnostic framework.
Furthermore, although the first layer thickness is about 50 m, the 2-m temperature and 10-m wind speed are not simple averages of this layer but are diagnosed by the surface layer scheme based on stability and surface fluxes. Therefore, the response differences of different PBL–surface layer combinations to observed variables under fire conditions can still reflect the handling characteristics of parameterization schemes towards fire-induced perturbations. Relevant physical assumptions and their limitations have been explained in the revised manuscript.
Finally, we have strengthened the elaboration of the research motivation and potential application scenarios in the revised manuscript, explicitly pointing out that the results of this paper are mainly applicable to the following scenario: evaluating and comparing the behavior of PBL parameterization schemes in strong, heterogeneous thermal perturbation environments under conditions where computational resources are limited and long-duration nested LES cannot be conducted. This paper does not aim to replace high-resolution LES or microscale fire behavior models, but to provide a reference for understanding and improving the applicability boundaries of PBL schemes in coupled models.
The advised sentences are added:
In complex terrain regions, wildfire behavior is significantly modulated by terrain-induced circulations, slope winds, and fire–atmosphere interactions. To strike a balance between physical consistency and computational feasibility, this study employs the WRF-Fire coupled model with an innermost horizontal resolution of 300 m for the atmospheric model and 30 m for the fire behavior model. The 30 m fire grid reasonably characterizes fuel heterogeneity, slope and aspect effects, and fireline geometry in complex terrain. Meanwhile, the 300 m resolution atmospheric model can resolve mesoscale terrain circulations that critically impact fire behavior, while supporting multi-day continuous simulations and multiple sensitivity experiments. At this scale, meeting the grid resolution, vertical stratification, and time step requirements for LES is difficult; furthermore, coupling with strong fire heat fluxes often leads to numerical instability and unacceptable computational costs, making it practically unfeasible. Given that the objective of this study is not to resolve fire-induced microscale turbulent structures, but to evaluate the response characteristics of near-surface meteorological variables and different boundary layer parameterization schemes under typical wildfire backgrounds, this study adopts PBL parameterization schemes at 300 m resolution to obtain robust, reproducible, and computationally manageable simulation results. It should be noted that this configuration cannot finely represent the microscale turbulence and thermal plume structures near the fire front, which constitutes the primary limitation of the current study regarding scale and model configuration.
- The authors' motivation for using a PBL parameterization in a 300 m domain, along with a 50-m vertical layer as the 1st model level must be clearly stated. I strongly recommend they includea simulation using one of WRF's LES optionsin this analysis, which can then be configured with the 1st model level closer to the ground. This will provide a better understanding of the effects of using a PBL scheme instead of LES, and whether this is a critical aspect to be accounted for in these simulations. It would be desirable and insightful (but not critical) to also include one of the 3-D PBL options, which can be configured following the other PBL schemes.
We thank the reviewer for this critical suggestion. We have explicitly supplemented the motivation for adopting PBL schemes in the revised manuscript. Although the horizontal resolution of the innermost atmospheric grid reaches 300 m, this scale is still insufficient to meet the basic premise of LES, which is to explicitly resolve the energy-containing turbulent eddies. At this resolution, most of the turbulent kinetic energy remains within the sub-grid scale range; thus, employing LES is not physically self-consistent and makes it difficult to obtain stable, reproducible multi-day simulation results numerically. In contrast, PBL parameterization schemes remain a more robust and widely applied choice within this scale range, especially suitable for the research objective of conducting multi-scheme comparisons and sensitivity experiments in complex terrain regions.
We also agree with the reviewer regarding the importance of vertical resolution and have explicitly pointed out in the revised manuscript that the default WRF vertical stratification (with the first layer at approximately 50 m) indeed cannot finely represent the strong thermal gradients and turbulent structures within the few meters near the fire front. However, this study focuses on the integrated response of near-surface meteorological elements (such as 2-m temperature and 10-m wind speed) under fire conditions, and the relative performance differences of different PBL schemes to fire-induced perturbations, rather than the microscale structure of the fire plume itself. Within the framework of the current multi-day continuous simulation and multiple experiments, increasing the near-surface vertical resolution would significantly increase computational costs and introduce numerical instability. Therefore, this study adopted a vertical stratification setting that represents a compromise between feasibility and physical consistency.
We acknowledge that LES is of great value for understanding fire–atmosphere interactions and have tested multiple LES configurations in our preliminary work, including sensitivity experiments with reduced first-layer heights. However, under complex terrain conditions, LES coupled with the interactive fire model exhibits poor numerical stability; the simulation duration typically struggles to exceed 24 hours, and the computational cost increases significantly, making it difficult to support the multi-day simulations and systematic multi-scheme comparisons required for this paper. Consequently, LES results are not included as the main analytical content of this article.
It should be noted that the vast majority of PBL schemes in WRF (including MYJ, MYNN, BouLac, and UW) are inherently based on the one-dimensional vertical flux assumption; their differences mainly lie in turbulence closure forms, mixing lengths, and stability functions. In this study, by enabling 3D turbulence diffusion options (diff_opt = 2, km_opt = 4), we allowed sub-grid turbulent diffusion in both horizontal and vertical directions, thereby mitigating the limitations of the pure 1D assumption to a certain extent. Due to limitations in computational resources and the research focus, we did not further introduce experimental 3D PBL configurations, but we agree that this direction holds potential value and have acknowledged it as a possible extension for future work in the revised manuscript.
Under the conditions of 300 m horizontal resolution and 50 m first-layer height, the amplitude of fire-induced microscale turbulence and strong thermal plumes might be underestimated, which could affect the absolute intensity of fire–atmosphere interactions. However, since all experiments employ the same dynamic framework and vertical stratification, this bias is consistent across different PBL schemes and thus does not affect the conclusions regarding the relative performance and sensitivity of the different PBL schemes presented in this paper.
- Their comparison with observations is based on 2-m temperature and 10-m winds. However, the simulations' temperature and winds represent a 0-50 m deep layer. Although WRF's atmosphere and land surface are coupled, the fire heat is only coupled to the atmosphere. This means the soil temperature can only increase from the increased air temperature, not from the fire directly. This leads to more inertia in the heat transfer and underrepresented heat gradients. Considering that T2 is a diagnostic variable derived from T, and T represents a 50-m deep layer, I'd expect the fire feedback to be substantially diluted. The same applies to 10-m winds. These variables are interpolated to 2 and 10 m, they are not resolved by the model at these scales. For this reason, I suggestincluding a control simulation with fire feedback turned off (or just no fire at all). This will show the magnitude of the fire perturbations affecting the PBL schemes. They may be quite minor. In addition, this would allow the authors to compare how each scheme transports the fire heat, i.e. by looking at the temperature or energy difference between the control and the corresponding fire simulation for each PBL scheme.
We thank the reviewer for the in-depth discussion regarding the diagnostic mechanism of near-surface variables and the fire–land–atmosphere coupling issue. Our understanding and response to this issue are as follows.
First, in WRF, 2-m temperature (T2) and 10-m wind speed (U10) are diagnosed by the surface layer parameterization scheme based on the atmospheric state of the first layer, surface fluxes, and Monin–Obukhov similarity theory. Although the thickness of the first layer is indeed approximately 50 m, the calculation processes for T2 and U10 explicitly introduce stability corrections, friction velocity, and scaling variables. Therefore, these near-surface variables are not explicitly resolved at their nominal heights, and that fire-induced perturbations may be diluted when expressed through diagnostic quantities.
Second, we agree with the important fact pointed out by the reviewer: in WRF-Fire, fire heat is only coupled to the atmosphere and does not directly act on the soil. Consequently, changes in soil temperature mainly originate from the adjustment of sensible heat flux after the air is heated by the fire, rather than direct fire–soil heat conduction. This mechanism indeed leads to greater thermal inertia in land–atmosphere heat exchange and may underestimate the near-surface thermal gradient.
However, despite these limitations, T2 and U10 remain the only variables directly comparable to routine surface observations and are therefore commonly used in previous WRF-Fire evaluation studies. In our study, these variables are not interpreted as exact point-scale representations, but rather as indicators of the integrated near-surface atmospheric response to fire-induced forcing under different PBL parameterizations.
Regarding the suggestion to add a "fire-off" (closed fire feedback) or no-fire control experiment, we consider this suggestion physically reasonable. However, given that the authors have conducted similar research previously, it is not adopted within the framework of this study, primarily based on the objective of this study.
Finally, regarding the LES issue, we agree that adopting LES and setting the first layer height lower is theoretically more beneficial for resolving fire-induced turbulent structures, which is also one of the main advantages of LES or ultra-high-resolution simulations. However, in the simulation of complex mountain terrain and long-duration fire processes involved in this study, nested LES faces significant challenges in terms of numerical stability and computational cost. Therefore, we adopted a first layer thickness of approximately 50 m as a compromise solution, allowing the simulation to cover the complete fire process and perform multi-scheme comparisons while ensuring numerical stability.
- Lastly, the CONUS namelist does not include the fire configuration. Pleaseinclude the fire configuration, including ignition location and any other entry under “&fire” that is different from the model default.
We thank the reviewer for the suggestion. The namelist settings used in the simulation is as follows:
&domains
sr_x = 0, 0, 10,
sr_y = 0, 0, 10,
&fire
ifire = 0, 0, 2,
fire_num_ignitions = 0, 0, 1,
fire_ignition_start_lon1 = 0, 0, 106.3428,
fire_ignition_start_lat1 = 0, 0, 29.7705,
fire_ignition_end_lon1 = 0, 0, 106.3428,
fire_ignition_end_lat1 = 0, 0, 29.7705,
fire_ignition_radius1 = 0, 0, 35,
fire_ignition_start_time1 = 0, 0, 52200,
fire_ignition_end_time1 = 0, 0, 52204,
fire_print_msg = 0, 0, 0,
fire_print_file = 0, 0, 0,
fmoist_run = .false., .false., .true.
fmoist_interp = .false., .false., .true.,
fmoist_only = .false., .fasle., .false.,
fmoist_freq = 0, 0, 0,
fmoist_dt = 10, 10, 10,
fire_fmc_read = 0, 0, 0,
fire_boundary_guard = -1, -1, -1,
fire_fuel_left_method = 1, 1, 1,
fire_fuel_left_irl = 2, 2, 2,
fire_fuel_left_jrl = 2, 2, 2,
fire_atm_feedback = 1., 1., 1.,
fire_grows_only = 1, 1, 1,
fire_viscosity = 0.4, 0.4, 0.4,
fire_upwinding = 3, 3, 3,
fire_lfn_ext_up = 1.0, 1.0, 1.0,
fire_test_steps = 0, 0, 0,
fire_topo_from_atm = 1, 1, 0,
*It's worth noting that the official release of the WRF model only allows a meso-LES configuration with the YSU scheme. Yet, it is possible to make it work with other schemes with minimal code modification: the variables in the registry associated with the package pbl=0 must include the same variables in the associated pbl scheme.
We appreciate the reviewer's suggestion and will strengthen our work in this direction in future research.
Citation: https://doi.org/10.5194/egusphere-2025-3072-AC2
-
AC2: 'Reply on RC1', Yongli Wang, 18 Jan 2026
-
RC2: 'Comment on egusphere-2025-3072', Anonymous Referee #2, 04 Dec 2025
The authors tried to address the dynamical linkage between wildfire processes and the modeling scheme in WRF. However, the paper has no discussion about wildfire ignition and spread. Also, WRF-Fire has modeling and practical limitations for wildland urban interface. Authors need to address how the information from these simulations can be used for decision making in the build environment. The title reference to wildfire is misleading and needs to be changed. When the wildfire component is removed, the paper reads similar to other papers on comparing PBL schemes at a specific site. To differentiate the current work from literature, the authors need to identify unique contribution from the current work.
Other Comments:
1. Authors need to justify on the choice of the PBL scheme. Why were the five schemes chosen from the plethora of PBL schemes in WRF. Also why were the more recent Hybrid PBL-LES schemes not considered?
2. Any reason why the same surface scheme was used for all the PBLs? MYNN has its own surface scheme too.
3. WRF is ideally run with 1:3 refinement to minimize dispersion. A higher ratio is used mostly when downscaling to LES to avoid the Terra Incognito. Any specific reason on the use of 1:5 ratio in this study?
4. 300 m is a gray area for the application of a. PBL scheme. Any specific reason why was the grid not further refined and PBL replaced with LES?
5. 300 m resolution is too coarse for wildfire studies especially since the wind speed accuracy over complex terrain is the most vital factor for studying wildfire ignition and propagation.
6. How many vertical layers were within the ABL? The 2 m and 10 m are diagnostic variables and the accuracy over complex terrain depends on the height of the first vertical cells.
7. What land usage data and terrain height data was used to initialize the simulations?
8. The conclusion that the MYNN3 was the best is not clearly seen from Figure 5 and 6, especially since no error bars from observations have been included.
Citation: https://doi.org/10.5194/egusphere-2025-3072-RC2 -
AC3: 'Reply on RC2', Yongli Wang, 18 Jan 2026
The authors tried to address the dynamical linkage between wildfire processes and the modeling scheme in WRF. However, the paper has no discussion about wildfire ignition and spread. Also, WRF-Fire has modeling and practical limitations for wildland urban interface. Authors need to address how the information from these simulations can be used for decision making in the build environment. The title reference to wildfire is misleading and needs to be changed. When the wildfire component is removed, the paper reads similar to other papers on comparing PBL schemes at a specific site. To differentiate the current work from literature, the authors need to identify unique contribution from the current work.
We thank the reviewer for the constructive comments. In the revised manuscript, we have further clarified the research objectives, the core role of observational data, and the unique contributions of this paper relative to existing work. Specific explanations are as follows:
Although this wildfire occurred near a city, while over 98% of the burned area consisted of forest vegetation. The fire area is located within a scenic area covered by continuous forests and did not involve the burning of buildings or typical Wildland-Urban Interface (WUI) fire behavior processes. Therefore, this study remains focused on the issue of atmospheric response under wildfire conditions, rather than urban or structural fire simulation.
The focus of this study is not to finely characterize fire ignition mechanisms or provide engineering-grade predictions of fire spread, but rather to utilize the strong thermal perturbations generated by a real wildfire event as a "stress test" to examine the performance differences of different boundary layer schemes under extreme near-surface conditions. To avoid remaining solely at a theoretical or idealized comparison level, this paper conducts evaluations based on station observational data near the fire area (distances ranging from 1 to 9 km).
Specifically, this study employed hourly data from 5 fixed automatic weather stations and 1 emergency mobile observation station surrounding the wildfire to systematically compare the 2-m temperature, 10-m wind speed, and related atmospheric characteristics simulated by WRF-Fire. This type of observational data possesses irreplaceable importance: the observation locations were directly affected by the fire field, capable of truthfully reflecting the near-surface response under conditions of fire-induced heating and enhanced turbulence; the distribution of multiple stations (upwind, fire edge, and downwind) allows the evaluation to go beyond a single point and reflect spatial differences under complex terrain conditions; the emergency mobile station provided supplementary information at close range to the fire, enhancing the constraints on model results. It is precisely based on these observational data that this study can distinguish the response characteristics of different PBL schemes regarding near-surface wind fields, turbulence intensity, and boundary layer structures in a fire environment.
Furthermore, a comparative analysis between the WRF-Fire simulated burned area and the MODIS burned area product (Figure 3) has been added to the revised manuscript to verify the rationality of the fire–atmosphere coupled simulation, thereby avoiding the simplification of the research into a "fire-module-removed PBL comparison experiment". The five boundary layer schemes (MYJ, MYNN2, MYNN3, BouLac, and UW) produced burned area estimates ranging from 47.38% to 92.82%, the MYNN3 scheme had the best result.
In summary, the unique contribution of this study lies in the systematic evaluation of the performance of multiple PBL schemes under strong fire-induced thermal forcing conditions, within the context of a real wildfire in complex terrain, utilizing near-field station observational data. The research results provide valuable references for model development regarding the applicability and limitations of PBL parameterization schemes under non-typical, forced conditions, which is highly aligned with GMD's focus on model behavior and applicability boundaries.
Other Comments:
- Authors need to justify on the choice of the PBL scheme. Why were the five schemes chosen from the plethora of PBL schemes in WRF. Also why were the more recent Hybrid PBL-LES schemes not considered?
We thank the reviewer for this important question. In this study, the MYJ, MYNN2.5, MYNN3, BouLac, and UW boundary layer schemes are selected for their consistency with the same surface layer, and also because they are widely used in wildfire–atmosphere coupling research.
The objective of this study is not to be exhaustive of all PBL schemes, but to compare the schemes that are most representative, exhibit good stability, and hold practical application value in regional-scale wildfire simulations.
In this study, a planetary boundary layer (PBL) parameterization scheme (bl_pbl_physics ≠ 0) is retained, while three-dimensional turbulence diffusion is activated (diff_opt = 2) together with a turbulent kinetic energy–based subgrid-scale closure (km_opt = 4). This configuration allows explicit three-dimensional turbulent transport while maintaining conventional PBL parameterization, enabling partial resolution of larger turbulent motions and parameterization of smaller-scale turbulence.
As the PBL parameterization remains active, this setup does not represent a true large-eddy simulation (LES). Nevertheless, compared to traditional PBL configurations that rely solely on one-dimensional vertical mixing, the inclusion of three-dimensional turbulence diffusion and TKE-based closure places the simulation within the so-called gray zone between PBL and LES. Therefore, this configuration can be reasonably classified as a hybrid PBL–LES approach, characterized by the coexistence of parameterized boundary-layer processes and partially resolved turbulent motions, with potential implications for fire-induced flow structures and associated uncertainties due to possible double counting of turbulence.
- Any reason why the same surface scheme was used for all the PBLs? MYNN has its own surface scheme too.
In WRF, the surface layer scheme acts as a critical interface between the land surface model and the boundary layer scheme. It does not directly predict atmospheric state variables, but diagnoses momentum, heat, and moisture fluxes based on Monin–Obukhov similarity theory, utilizing the lowest-level atmospheric variables and surface characteristics. The surface layer scheme provides the boundary layer scheme with friction velocity, surface fluxes, and stability parameters; these quantities serve as the lower boundary conditions for the boundary layer turbulence equations, directly regulating turbulence production and vertical mixing intensity.
Different surface layer schemes employ distinct variants in calculating stability functions and roughness lengths, which affect the computation of turbulent exchange. The purpose of using a single, unified surface layer scheme across all experiments is to maximize the isolation of the effects of the boundary layer schemes themselves. If different PBL schemes are paired with different surface layer schemes, it would introduce additional coupling effects, making it difficult to distinguish whether the differences in results originate from the PBL scheme or the flux calculations.
Although the MYNN scheme comes with a dedicated surface layer option, a unified surface layer setting helps ensure that all schemes are compared under identical surface forcing conditions.
The new sentence has been added in the manuscript:
Therefore, the sensitivity of wildfire–atmosphere coupling to PBL schemes partly originates from how surface-layer fluxes are ingested and redistributed within the boundary layer, rather than from differences in the surface layer formulation itself.
- WRF is ideally run with 1:3 refinement to minimize dispersion. A higher ratio is used mostly when downscaling to LES to avoid the Terra Incognito. Any specific reason on the use of 1:5 ratio in this study?
The adoption of a 1:5 nesting ratio in this study (with an innermost atmospheric model resolution of 300 m and a fire spread model resolution of 30 m, the time step is 0.2 sencond) represents a trade-off between computational efficiency, applicability in the gray zone, and regional representativeness.
Fire–atmosphere coupling is highly non-linear, even a time step difference of 0.2 s can alter the turbulent structure and the local circulation around the fire front, leading to differences in wind speed and temperature feedback. Adopting a 1:5 ratio not only reduces the number of nesting levels but also allows the time step of the innermost domain to be minimized.
Although a 1:3 nesting ratio helps reduce numerical dispersion, a 1:5 ratio is also common in existing WRF-Fire research when downscaling from synoptic scales to convection-resolving or fire scales. The outer grids primarily provide large-scale background circulation constraints, while the innermost 300 m grid focuses on the relative differences between different PBL schemes. Since all schemes utilize the same model configuration, this setup is suitable for comparing simulation differences arising from different physical schemes, ensuring the scientific validity and practicality of the conclusions.
- 300 m is a gray area for the application of a. PBL scheme. Any specific reason why was the grid not further refined and PBL replaced with LES?
We agree that the horizontal resolution of 300 m lies within the "gray zone" (or terra incognita) between traditional PBL parameterization schemes and explicit LES. Under ideal conditions, further refining the grid and adopting LES would allow for a more direct resolution of turbulent structures. However, in the context of regional-scale, multi-day continuous coupled wildfire simulations, LES still faces a series of practical and unresolved challenges; therefore, it was not selected as the primary configuration for this paper.
First, the demand for computational resources by LES increases non-linearly. Refining the horizontal resolution from 300 m to 100 m or even 50 m requires not only a significant reduction in the time step but also a synchronous refinement of vertical stratification (particularly requiring a layer thickness of <10 m in the surface layer). This typically increases the total computational cost by more than an order of magnitude. Given the complex terrain and the multi-day wildfire process investigated in this study, such a configuration is difficult to support systematic sensitivity experiments under currently available standard computational resources.
Second, the numerical stability of LES under strong wildfire–atmosphere coupling conditions remains an unresolved problem. The high spatiotemporal heterogeneity of fire-induced heat fluxes significantly amplifies the buoyancy production term in the surface layer, rendering LES highly sensitive to sub-grid scale (SGS) schemes, filter scales, and dissipation parameters. We had tried the time step 1 second in the d01 by using the LES configuration in d03, the simulation could not be well produced even in one simulation day.
Third, the compatibility of LES with existing operational wildfire forecasting frameworks remains limited. Currently, most regional-scale wildfire forecasting systems (including operational WRF-Fire applications) still rely on PBL parameterization as the core configuration. The objective of this study is precisely to evaluate the performance differences and applicability boundaries of different PBL schemes under strong fire-induced thermal perturbations within this realistic and widely used model configuration, rather than to explore the optimal simulation scheme under idealized conditions.
Therefore, although the 300 m resolution lies within the transition zone for PBL schemes, it still possesses clear practical significance for regional-scale wildfire simulations. The results of this paper can provide a benchmark reference for future research utilizing higher resolutions or hybrid PBL–LES frameworks.
- 300 m resolution is too coarse for wildfire studies especially since the wind speed accuracy over complex terrain is the most vital factor for studying wildfire ignition and propagation.
Thanks for the suggestion, although a horizontal resolution of 300 m cannot resolve fire-line scale turbulence structures and local wind field details, especially under complex terrain conditions, which is insufficient to finely characterize the ignition process and the transient dynamics of the fire front. However, it is important to emphasize that the objective of this study is not to finely reproduce the fire spread process itself, but to focus on the following two issues of greater regional-scale significance:
First, this paper focuses on the modulation effect of fire-induced heat fluxes on the overall boundary layer structure. Although the internal structure of the fire front cannot well be resolved at 30 m resolution (fire model resolution), the total heat released by the fire and its impacts on near-surface stability, turbulence intensity, and wind speed profiles can still be effectively characterized. Simulation results indicate that at this resolution, the model possesses a good capability to reproduce the burned area and the overall evolution of the fire front, which provides a reasonable basis for analyzing the differences in the fire–boundary layer coupling response among different PBL schemes.
Second, the importance of wind speed accuracy in wildfire research is not limited to the fire-line scale. In regional-scale simulations, wind fields within the surface layer and boundary layer are equally crucial for predicting fire spread direction, expansion trends, and potential risk areas. By comparing different PBL schemes, this paper reveals their different response mechanisms regarding wind speed, turbulence, and stability under strong thermal forcing. These differences have direct significance for understanding and improving future wildfire forecasts.
Finally, from an application perspective, the 300 m resolution represents a realistic trade-off between computational cost and physical representation capability. At the current stage, relying entirely on LES for regional-scale, complex terrain, multi-day wildfire simulations still face significant computational and technical barriers. In comparison, systematically evaluating the applicability of PBL schemes at this resolution helps to clarify the strengths and limitations of different schemes. These schemes selection in future higher-resolution or hybrid frameworks could provide physical and engineering references for the gradual transition to finer simulations.
Therefore, although the 300 m resolution cannot well resolve fire-line scale processes, its scientific value and practical significance in regional-scale wildfire simulation and boundary layer response research remain clear.
- How many vertical layers were within the ABL? The 2 m and 10 m are diagnostic variables and the accuracy over complex terrain depends on the height of the first vertical cells.
Under the default 45-layer configuration, the daytime planetary boundary layer typically contains approximately 4–6 model layers (depending on the specific PBL height). In WRF, the surface layer scheme applies Monin–Obukhov similarity theory between the surface and the first atmospheric model level to diagnostic 2-m temperature/humidity and 10-m winds, regardless of whether the first model level is located below or above 10 or 50 meters. Rather, it relies more on the combined interaction of the lowest model level height, surface roughness characteristics, and the validity of the surface layer similarity theory assumptions.
- What land usage data and terrain height data was used to initialize the simulations?
For the initialization of the atmospheric model in this study, the land use data adopted the default land use classification scheme of the WPS system to maintain consistency in surface parameterization across all experiments.
Given the sensitivity of complex terrain and wildfire processes to high-resolution terrain and fuel information, this study made targeted replacements for terrain height and fuel data related to fire behavior. Specifically, the terrain height data utilized 30-meter the Shuttle Radar Topography Mission (SRTM) Digital Elevation Database (DEM) data to more accurately characterize slope, aspect, and local topographic relief features within the study area. The fuel type data required by the fire behavior model employed the FROM-GLC 30 m global land cover dataset from Tsinghua University, which is remapped according to the fuel classification requirements of WRF-Fire.
This configuration improves the representation accuracy of the spatial distribution of terrain and fuels without altering the atmospheric physical parameterization settings, thereby facilitating a better characterization of the fire spread process under complex terrain conditions and its perturbation to the near-surface atmosphere.
- The conclusion that the MYNN3 was the best is not clearly seen from Figure 5 and 6, especially since no error bars from observations have been included.
We thank the reviewer for highlighting this important issue. We agree that relying solely on the individual statistical metrics presented in Figures 5 and 6 makes it difficult to definitively conclude that MYNN3 is "optimal" across all meteorological variables and aspects. Consequently, we have modified the relevant terminology in the revised manuscript and supplemented the analysis with additional evidence to support the assessment of MYNN3's comprehensive performance.
First, it should be clarified that the definition of "optimal" in this study is not based solely on a single meteorological variable or statistical metric, but rather on a comprehensive consideration of the consistency in simulating near-surface thermal states, wind field structures, and fire behavior. In the revised manuscript, we have introduced a comparative analysis between the simulated burned area and the MODIS satellite-retrieved burned area product. The results indicate that among all PBL schemes, MYNN3 shows the closest agreement with the satellite retrievals regarding the spatial distribution and overall magnitude of the burned area. From the perspective of fire behavior response, this result provides independent evidence for the comprehensive applicability of MYNN3 in the context of fire–atmosphere coupling, extending beyond reliance solely on station statistics of near-surface meteorological elements.
Second, regarding the observation comparison, MYNN3 exhibits the smallest mean bias for 2-m temperature, suggesting a relatively more reasonable characterization of near-surface thermal conditions. It is important to note that both 2-m temperature and 10-m wind speed are variables diagnosed via surface layer similarity theory. Their errors are associated not only with the PBL scheme but are also collectively influenced by the surface layer scheme, terrain representativeness errors, and the spatial heterogeneity of fire-induced perturbations. Under conditions of complex terrain and strong heterogeneous heat sources, the sensitivity of different variables to model errors is inherently inconsistent.
Regarding 10-m wind speed, MYNN3 exhibits relatively larger mean bias and RMSE; this finding has been truthfully retained and discussed in the revised manuscript. It remains difficult to unequivocally attribute this bias to a single mechanism; it is likely related to the combined action of the following factors:
（1）10-m wind speed is highly sensitive to sub-grid terrain effects and local channeling effects, which are difficult to fully resolve at 300 m resolution;
（2）The impact of fire-induced thermal perturbations on the wind field possesses significant spatiotemporal heterogeneity, making wind speed errors at the station scale more prone to amplification;
（3）Since 10-m wind speed is a variable diagnosed via surface layer similarity theory rather than explicitly resolved by the PBL scheme, different PBL schemes indirectly influence friction velocity and stability parameters by altering the wind speed profile, turbulence mixing intensity, and thermal stability in the first model layer. Under strong fire-induced perturbations and complex terrain, this indirect influence may be amplified, thereby increasing the uncertainty of the 10-m wind diagnosis.
Finally, regarding the issue of observational uncertainty (error bars), we agree that this is an important aspect, but with practical limitations. Systematic errors and representativeness errors of automatic weather stations are difficult to accurately quantify in complex terrain and fire environments; therefore, we did not perform a unified estimation of error bars in this paper.
Citation: https://doi.org/10.5194/egusphere-2025-3072-AC3
-
AC3: 'Reply on RC2', Yongli Wang, 18 Jan 2026
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,922 | 152 | 42 | 2,116 | 39 | 41 |
- HTML: 1,922
- PDF: 152
- XML: 42
- Total: 2,116
- BibTeX: 39
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Dear authors,
in my role as Executive editor of GMD, I would like to bring to your attention our Editorial version 1.2: https://www.geosci-model-dev.net/12/2215/2019/
This highlights some requirements of papers published in GMD, which is also available on the GMD website in the ‘Manuscript Types’ section: http://www.geoscientific-model-development.net/submission/manuscript_types.html
In particular, please note that for your paper, the following requirements have not been met in the Discussions paper:
As you use WRF-Fire for your analysis, please add something like “a case study using WRF-FIRE version x.y” to the title of your manuscript upon submission of the revised version to GMD.
Best regards, Astrid Kerkweg