Characterizing the Role of Moisture and Smoke on the 2021 Santa Coloma de Queralt Pyroconvective Event Using WRF‐Fire

Smoke from wildfires or burning biomass directly affects air quality and weather through modulating cloud microphysics and radiation. A simple wildfire emission coupling of black carbon (BC) and organic carbon (OC) with microphysics was implemented using the Weather Research and Forecasting model's fire module. A set of large‐eddy simulations inspired by unique surface and upper atmospheric observations from the 2021 Santa Coloma de Queralt Fire (Spain) were conducted to investigate the influence of background conditions and interactions between atmospheric and fire processes such as fire smoke, ambient moisture, and latent heat release on the formation and evolution of pyroconvective clouds. While the microphysical impact of BC and OC emissions on the dynamics of fire behavior is minimal on short time scales (<6 hr), their presence increased the cloud water content and decreased the rain rates in our case study. In our case study, atmospheric moisture played an important role in the formation and development of pyroconvective clouds, which in turn enhanced the surface winds (8%) and fire spread rate (25%). The influence of fuel moisture on the pyroconvective cloud formation is smaller when compared with the atmospheric moisture content. A better representation of cloud processes can improve the mesoscale forecasts, which is important for better fire behavior modeling.

The interactions of biomass burning aerosols and cloud microphysics are complex. Indirect representation of the aerosol mass in models using satellite data relies on assumptions regarding fire intensity, chemical composition, and vertical structure of the planetary boundary layer that leads to large uncertainty in the estimated total daily emission by a factor of about 20-50 between different models (Ye et al., 2021). Of the biomass burning aerosols, organic carbon (OC) and black carbon (BC) are the most important for the model physics as they can both act as cloud condensation nuclei (CCN) as they age, and the latter has strongly absorbing radiative properties (Matsui et al., 2018;Stier et al., 2006). The size distribution and hygroscopicity of the emissions are deterministic variables for cloud microphysics that are initially highly dynamic and vary widely as the aerosols age. In climate models, the aging process (here, conversion from hydrophobic to hydrophilic aerosols) is usually represented by a simple e-folding conversion with a turnover time (τ) of about 1.2-2.5 days (e.g., Colarco et al., 2010). However, intercomparison of observation and microphysics-based aging model simulation shows that τ can be as small as 0.6-2 hr and 1-3 hr near the source of emissions during summer and wintertime, respectively (He et al., 2016).
Although modeling all the complexities of fire smoke emissions can be a daunting challenge, a simple model for fire aerosols can improve the skill of mesoscale forecasts and their applications by taking into account their radiative and microphysical effects. Recent numerical experiments with the Weather Research and Forecasting (WRF) model show that adjustment of the aerosol characteristics based on OC and BC data from NASA Goddard Earth Observing System version 5 reanalysis (GEOS-5;Rienecker et al., 2008) in the Thompson-Eidhammer aerosol-aware microphysics (Thompson & Eidhammer, 2014, hereafter TE14) scheme lead to improved simulation of key variables such as cloud cover, air temperature, and incident radiation at the surface when compared with satellite and ground observations (Conrick et al., 2021). Another successful example of incorporating the radiative impact of OC and BC in WRF-Solar for the record-breaking 2020 wildfire season was presented by Juliano et al. (2022). WRF-Solar is an augmentation of WRF for forecasting solar power irradiance that includes radiation, clouds, and aerosol interactions. This modeling study used the GEOS-5 aerosol forcing of OC and BC to show a significant reduction in solar energy production by about 50% due to the wildfire smoke, which agrees with multiple surface observation sources. The combination of increased reliance on solar power and the frequent wildfires in these regions is a cause of concern without proper risk mitigation. Kochanski et al. (2019) used the direct approach of coupling the emissions in a similar model (WRF-SFIRE-CHEM) in the mountain valleys across Northern California during the 19-20 August 2015 wildfires. The smoke radiative feedback resulted in 3°C surface cooling, which caused an inversion-like condition and trapped the smoke in the area. However, this study did not include the microphysical impact of the wildfire smoke. A recent study (Y. Zhang et al., 2022) utilized WRF-Chem and a spectral-bin microphysics scheme to investigate the impact of regional and remote wildfire emissions on precipitation rates and hail in the western and central United States. The study found that heavy precipitation rates and hail increased by 38% and 34% respectively due to the combined effects of smoke and heat released. The study also found that local fires had a smaller impact than remote fires.
Here, we present a simple direct coupling of the fire emissions and microphysics in WRF-Fire (Coen et al., 2013), a widely used wildland fire-behavior physics module. The model calculates and tracks the hydrophobic and hydrophilic OC and BC based on the fire behavior and allows the hydrophilic (termed "water-friendly" by TE14) aerosols to act as CCN in the TE14 aerosol-aware microphysics scheme. We applied the new emission model to investigate the formation and development of pyroconvective clouds. Pyroconvective clouds form when a wildfire smoke plume reaches its condensation saturation level (American Meteorological Society, 2022). Depending on their vertical development, these clouds are referred to as pyrocumulus (pyroCu) or pyrocumulonimbus (pyroCb). Pyroconvective clouds lead to changes in the fire rate of spread caused by vigorous updraft/downdraft speeds (Rodriguez et al., 2020) generally increasing the fire intensity. They also increase the chance of long-range spotting due to the strong winds (e.g., Thurston et al., 2017).
The fire emissions in WRF-Chem and HRRR-smoke models are calculated using the fire radiative power measurements and satellite-based AOD. Satellite measurements are limited because of their overpass time and spatial resolution. Therefore, the aerosols are emitted in the model by assuming a typical diurnal cycle, and the vertical distribution of the plume is parameterized (e.g., Kumar et al., 2022). In order to avoid these assumptions, the EGHDAMI ET AL.

10.1029/2022MS003288
3 of 17 emissions can be calculated with a fire behavior model to calculate emissions based on the amount of fuel burnt. This consideration is critical to the simulation of convective scale phenomena such as pyroconvective clouds that require higher temporal and spatial resolution to be resolved accurately. A modeling approach that focuses only on the number of aerosols used by the microphysics processes as input does not significantly increase computational time and can ultimately benefit operational forecasts.
Prediction of pyroconvective clouds is challenging because several environmental factors such as meteorological conditions, fire intensity, fuel characteristics, and resulting emissions of heat, moisture, and aerosols interplay to influence their life cycle. The aerosols emitted by fires can act as CCN and ice nuclei (IN), and therefore change the microphysical and dynamical evolution of the clouds. Here, we do not take into account the direct smoke and radiation interactions and instead leave such investigation for future work. In addition to the impact of smoke, the sources of moisture in the formation of the pyroconvective clouds are not fully understood. There are various hypotheses regarding the relative importance of fire-released moisture on the ambient moisture sources in the initiation of pyroconvective clouds. Potter (2005) hypothesized that the moisture released from the fire is a major contributor to the pyrocumulus formation using historical radiosonde data. This theory implies a cloud base lower than the environmental lifted condensation level (LCL). On the other hand, Luderer et al. (2006Luderer et al. ( , 2009) used a cloud resolving plume model to show that the combination of heat and moisture released from the fire results in a higher cloud base than the LCL, in agreement with Lidar observations by Lareau and Clements (2016). A higher cloud base level of (∼1 km above the LCL) is also independently confirmed by Kablick et al. (2018) based on CloudSat profiler in another case of a pyroCb. Finally, it is worth noting that while the sensible heat generated by the fire is crucial for the pyrocumulus formation as demonstrated in previously studies (e.g., Lareau & Clements, 2017;Tory et al., 2018;Y. Zhang et al., 2019), it is not the focus of this work.
Ground-based observation of pyroconvective clouds is scarce and usually not possible due to inherently dangerous and spontaneous field conditions. Therefore, numerical models play a crucial role in advancing our understanding and enabling their characterization. Based on a series of idealized simulations with a second-order turbulence parameterization, Reutter et al. (2014) hypothesized that the increase in the aerosols due to smoke will result in smaller cloud droplet sizes and therefore, delayed onset of precipitation (see also Kablick et al., 2018). However, a more comprehensive study of pyrocumulus cloud formation considered different updrafts and aerosol loadings revealing that the interactions between fire behavior, turbulence, and cloud condensation and, therefore, precipitation are very complicated and do not vary monotonically (Chang et al., 2015).
Using a set of large-eddy simulations (LESs), we investigated the role of atmospheric moisture, fuel moisture, latent heat release, and fire emissions on pyroconvective cloud initiation and evolution. We relied on the unique observations from firefighters' perspective including the sounding profile, plume monitoring, fuel property, and fire behavior on site to set up the model and interpret the results (Castellnou et al., 2022a(Castellnou et al., , 2022b. The prior studies (e.g., Chang et al., 2015;Reutter et al., 2014) relied on prescribed pyrogenic aerosols; whereas in our model, we directly calculated the aerosol forcing emitted by the fire. We focus on an idealized domain setup and short time scales to improve our understanding of fundamental processes, this being a first step for modeling real case scenarios. This research will contribute to a better understanding of the role of environmental factors in the initiation and evolution of pyroconvective clouds. Future applications can be extended to improve the representation of aerosol emissions by fire in weather and air quality models.
This manuscript is organized as follows. The description of our case study, emission model, and numerical experiments are given in Section 2. In Section 3, the results are presented and discussed. Conclusions and future directions are given in Section 4.

The Case of Santa Coloma de Queralt Fire
The Santa Coloma de Queralt (SCQ) Fire in Catalonia, Spain started on Saturday afternoon, 24 July 2021, at about 14:00Z, in the province of Tarragona (Figure 1a, Castellnou et al., 2022a). The fire was initially driven by northwesterly winds toward the southeast. The interplay of the westerly synoptic-scale wind and a southwesterly sea breeze eventually drove the fire to the northeast. The fire experienced two critical periods. The first was on Saturday afternoon ( Figure 1b The fire that was initially pushed in the west-east direction by westerly dry winds was held on a valley bottom only after the arrival of the sea breeze to the area. The interaction of both winds as they converged kept the fire from spreading eastward and allowed the tactical deployment of the firefighters to stop the uncontrollable upslope spread of the fire (see Figure S1 in Supporting Information S1). By Sunday afternoon, the firefighters had almost extinguished the fire before the second critical period where deep pyrocumulus clouds were spotted The firefighters had to temporarily evacuate the area due to hazardous conditions caused by erratic wind behavior. The fire burned about 1,657 ha of land, and 168 residents were evacuated before it finally subsided on Sunday evening (22:00Z). The fuel characteristics were similar over the area, and the dead fuel moisture content was between 6% and 7.7% during the event. We hypothesize that the difference in plume development between 24 and 25 July was strongly influenced by atmospheric conditions, such as the sea breeze, which penetrated deeper into the mainland on the second day. This change in atmospheric moisture availability, upper-atmosphere stability conditions, and shear may have contributed to the observed differences.
The fire responders in the area released a radiosonde in the pyrocumulus updrafts on 25 July at about 20:19Z. Figure 2 shows temperature and dewpoint profiles from ERA5 reanalysis (solid lines) and observation (dotted lines). As can be seen in the observation profile, the dewpoint and temperature are very close above 700 hPa , indicating an environment favorable for the pyrocumulus cloud formation with a cloud base about 700 hPa and a cloud top above 400 hPa . The observation profile also stops at about 400 hPa because the radiosonde was affected by strong downdrafts. Because the radiosonde is released in the pyrocumulus updrafts near the fire, the data are influenced by the fire's sensible and latent heat release and converging winds above the fire. Therefore, we chose the ERA5 data to set up the model sounding, which may better represent the ambient air. However, we increased the relative humidity by 20% (dotted-dashed line) in the sounding profile because ERA5 was too dry to represent the local conditions and did not produce a pyrocumulus cloud. (Maps of surface winds and relative humidity from Catalonia weather service can be found in Figures S2 and S3 in Supporting Information S1). We expect large gradients in the moisture field because the fire is located between the dry and moist convergence zone of westerly dry and moist southwesterly winds. Furthermore, the local circulation is impacted by the complex topography which intensifies the mesoscale circulation. These details are not well represented in a model with coarse resolution.

Aerosol-Aware Microphysics and Fire Emission Model
The goal of the implemented model is to calculate fire emissions and pass them to other physics modules, particularly the microphysics in this work. We used the TE14 scheme which is a microphysics parameterization with explicit diagnosis of CCN and IN activation by tracking and predicting the number of available aerosols rather than a prescribed estimate. The aerosols used for diagnosis of CCN are the number of water-friendly aerosols (hydrophilic) regardless of their chemical compositions, which in the TE14 model assumed to be a combination of sulfates, sea salts, and organic matter. The initial profile of water-friendly aerosols, which here we refer to as background aerosols, is estimated based on continental climatological values (based on Goddard Chemistry Aerosol Radiation and Transport, GOCART; see TE14) starting from O(3 × 10 8 kg −1 ) near the surface and decaying to O(5 × 10 7 kg −1 ) above the boundary layer. The fire emissions that are water-friendly therefore must be added to these aerosols for CCN activation.
From the fire emissions, we only consider the OC and BC, which are the most influential aerosols on radiation and microphysics. With respect to shortwave radiation, OC aerosols are strongly scattering, while BC aerosols are strongly absorbing. In the fire behavior module, the emissions mass of aerosols species based on the burned fuel are estimated from the following equation (Wiedinmyer et al., 2011): where the emission mass for species is calculated based on the burnt area and fuel loading . The emission factor efi depends on the land-cover type. We assume temperate forest land cover biomass burned, corresponding to emission factors of 9.2 and 0.56 g kg −1 (Wiedinmyer et al., 2011) for OC and BC, respectively. From the mass concentration, we calculated the number of aerosols assuming a bimodal lognormal distribution in an equilibrium The green dotted-dashed line shows the ERA5 dew point with an increased relative humidity of 20% (the convective available potential energy will be ∼2,500 J Kg −1 ).
6 of 17 state with 75%-25% particles in accumulation and Aitken modes. The density and radius parameters were chosen following Chin et al. (2002) for their compatibility and consistency with the WRF chemistry module and TE14. We assumed all the emissions were initially hydrophobic and would become hydrophilic with hygroscopy of as they age in both Aitken and accumulation modes for simplicity. The aging process is modeled by a simple e-folding conversion with turnover time ( ): where here BC1 and BC2 show the hydrophobic and hydrophilic BC number concentration and and are dry and wet depositions. Similar equations were used for OC. Here, we ignored the dry deposition because it is negligible at short timescales. The hydrophilic aerosols are removed only due to nucleation or wet scavenging (TE14).
The emissions were added to the background water-friendly aerosols in TE14 microphysics when the activation of CCN was calculated. The TE14 used a bulk model for CCN activation in which the bulk hygroscopicity for the background water friendly aerosols is 0.4. Higher values of means a higher tendency for the particles to absorb water. We used the values of 0.1 and 0.2 for both species once they have aged following examples in the literature (Andrea & Rosenfeld et al., 2008;Mikhailov et al., 2015;Reutter et al., 2014;Twohy et al., 2021) for model-sensitivity analysis. The mixed background and smoke hygroscopicity were defined as the weighted average of the background and fire emitted species for simplicity.
The code modifications and calculation of the fire emissions are contained within the fire module code and given to the microphysics scheme as an input as shown in the schematic Figure 3. In summary, the model development includes the following: (a) Introducing new variables for hydrophilic and hydrophobic OC and BC, (b) Calculating the fire emissions based on emission factors provided and the fuel burnt as Equation 1, (c) Calculating the number of aerosols (the code is adopted from WRF-Chem) from the mass with distribution parameters provided in Chin et al. (2002), (d) Adding the aerosol variables from the fire behavior modules as an input to microphysics, (e) Aging and conversion of the hydrophobic to hydrophilic aerosols, (f) Adding the hydrophilic aerosols to the water friendly aerosols for cloud nucleation, (g) Modeling wet scavenging for fire aerosols similar to the water friendly aerosols.

Numerical Experiments
We used the WRF Model (WRF-ARW; Skamarock et al., 2019) version 4.3 in LES mode for simulating the explicit turbulent interactions of pyrocumulus clouds and the environment in an idealized domain. The surface and upper thermodynamic conditions are constrained by the observations. The numerical domain included two centered one-way nested domains with 360 × 360 dimensions with 150 and 50-m grid spacing. The outer domain had periodic boundary conditions. This setup allows homogeneous turbulent boundary conditions for the inner domain without the atmospheric feedback from the fire returning into the domain. Both domains had 91 vertical levels with 23 layers in the first kilometer with flat terrain. The model top is set to 15 km . With this horizontal resolution, the most important convective-scale motions are resolved explicitly; therefore, no planetary boundary layer or cumulus parameterization was needed. The 3D TKE-based SGS scheme of Deardorff (1980) is used to parameterize the subgrid-scale turbulence for the LES runs. Physics options included TE14 microphysics parameterization and the revised MM5 surface layer scheme (Jimenez et al., 2012). No radiation scheme is used. We chose 0.5-m surface-roughness length, a median value for typical forest and shrubland. We chose the 6-m wind height (fire_wind_height) for the flame spread following Rossa and Fernandes (2018). The distribution of the fuel (González-Olabarria et al., 2019) within 1 km of the location of the fire shows that the dominant fuel types are very heavy shrub load and high-load conifer litter with shrub understory based on Scott and  Table 1. We prescribed a constant heat flux (15 W m −2 ) on the surface representing the late afternoon conditions, and let the model run for 2 hr to spin up before igniting the fire. For all simulations, the fire was ignited over a line 1.5 km long, centered in the south-north direction and 2 km east from where the domain starts. This length scale was chosen following the observed fire isochrones at the time of the pyrocumulus formation during the SCQ event (Figure 1a). We trigger turbulence at initialization by assigning small randomly distributed potential temperature perturbations to the lowest four model grid cells. All the simulations ran for a total time of 6 hr including the spin up.
For our control run (CTL), we used the temperature and moisture sounding profiles at 25 July 2021 20:00Z from ERA5, and to achieve a profile closer to the observation sounding, we increased the relative humidity by a multiplicative of 20% in the entire column by increasing specific humidity only (Figure 2). A dry atmosphere case (CTLD) without the increase in relative humidity was conducted to examine the role of ambient moisture. We prescribed a constant wind profile of 5 m s −1 during the simulation (in the positive X direction), representing the observed surface winds to drive the fire but isolating the impact of vertical shear. Note that the default background aerosols (i.e., no smoke aerosols) were used in the TE14 microphysics parameterization in the control runs. The fuel moisture was set to (6%) for all these runs. The SMK and SMKD configurations are the same as the CTL and CTLD configurations except they also include the BC and OC fire emissions.
To investigate the impact of fuel moisture, we conducted runs of FM4 and FM8 similar to SMK, but the fuel moisture is modified to 4% and 8%, respectively. In addition, we included a simulation where the fuel moisture is kept the same (6%), but the latent heat release of the fire was set to zero (NOLH). Table 2 summarizes the model simulations performed in this study.
Finally, to investigate the sensitivity of our results to OC and BC aging turnover time and hygroscopicity, we conducted a series of simulations KXTY similar to SMK simulation, but where X/10 is the hygroscopicity and varies between 0.1 and 0.2 and aging turnover time is Y and varies between 1, 2, 4, and 8 hr (Table 3). Ensemble simulations with 10 members were produced by shifting the inner domain relative to the outer domain which is initialized with random temperature perturbations by 1 km for each simulation for all cases in Tables 2 and 3. The ensemble averages are used in our analysis to minimize the influence of turbulence variability in the analysis.

Results and Discussion
In this section, we investigate the role of atmospheric moisture, fuel moisture, latent heat release, and fire emissions on the pyroconvective cloud formation and evolution by analyzing in detail the evolution and structure of the dynamical and microphysical properties of the clouds. We complete the sensitivity analysis by studying how the aerosol hygroscopicity and aging turnover time impact the formation of the pyrocumuli.

The Role of Atmospheric Moisture and Fire Emissions on Fire Dynamic
The pyroconvective clouds form as a direct result of the fire and their growth depends on the evolution of the fire front. Figure 4 shows the map of fire heat flux and 10-m s −1 winds averaged in the lowest 1-km for the SMK simulation. The line of fire spread propagates without significant narrowing, providing a consistent heat source from the fire. This behavior is typically seen for timber and understory fuel type (e.g., Coen et al., 2013), unlike the homogenous fuel such as grass, which results in narrowing of the fire line as the fire propagates faster in the center of the line than the edges. Nevertheless, the fire heat release is less concentrated and loses intensity (from 28.6 to 22.9 kW m −2 from 2:30 to 5:30 hr) over the duration of the simulation. Based on Byram equations, the fire heat flux was estimated by the fire fighters to a maximum value of 280 kW m −2 (personal communication), which is comparable with  the maximum values that are shown in this figure (∼140-340 kW m −2 ). The wind pattern in all the figures shows areas of enhanced converging wind near the head of the fire due to convective updrafts. The convergence can be seen where the blue arrows on both sides of the fire are tilted toward the center. In the upstream, there are areas of divergence due to convective downdrafts. Notably, at 3:30 hr (Figure 4b), the direction of the flow is reversed against the background wind ahead of the fire (X ∼ 6 km ). These changes of wind direction are direct results of fire and are not captured in weather prediction models that are not coupled with a fire behavior model. The unpredictable changes in the wind direction can create dangerous conditions for the operational responders in the area. Figure 5 shows the differences in the 10-m wind speeds between the SMK and SMKD ensemble runs. Although the fire spread rate is slightly slower for dry atmospheric conditions, it is still possible to note the enhanced wind speed pattern near the fire front, resulting in faster fire spread. The differences between the CTL and CTLD are similar and hence not shown here, but the quantitative analysis is presented with the time-series of domain averaged variables next. Figure 6 shows the time series of the burnt area and 10-m wind speed. The wind in the vicinity of the fire is important for the fire spread and fire responders. The surface winds are averaged for a 2 × 2 km 2 box that is centered over the fire front (see the black square box in Figure 4a). There is no substantial difference between the SMK and CTL timeseries. However, the burnt area is about 25% (coefficient of variation CV < 0.1) larger overall than the dry cases (SMKD, CTLD). The faster fire spread was caused by the enhanced (about 8% overall compared to the dry cases with CV < 0.1) updrafts at the fire fronts. There is no significant difference in the near-surface dynamics due to the impact of emissions. Although the cases presented here show small differences between SMK and CTL simulation, longer simulation time and inclusion of radiative influence of aerosols may enhance these differences. Figure 7 compares the vertically integrated cloud-water path time series averaged over the domain and rain rate averaged over the areas where it rained. The time series show intermittent formation and dissipation of the pyroconvective clouds for SMK and CTL cases. This pulsing behavior is similar to severe-weather-producing weakly forced thunderstorms (e.g., Miller & Mote, 2017) that are produced in environments with high convective energy available (see Figure 2) and low vertical wind shear. Initially, SMK and CTL cases are very similar because both simulations start with a clean atmosphere. As smoke aerosols age from hydrophobic to hydrophilic with time and are available to act as CCN, the SMK case shows higher amounts of cloud water of up to 17% (CV ∼ 0.6) and 9% (CV ∼ 0.5) for the final one and 3 hr, respectively. Given the high variability in the results, to achieve better statistical accuracy and fully explore the effects of fire aerosols on cloud microphysical properties across the phase space, further case studies and a larger ensemble are required. Note that the blue lines near the horizontal axis are cloud water paths for SMKD and CTLD cases showing that no cloud is produced in the dry atmospheric conditions. The modest rain rates or virga are possible during the pyrocumulus cloud formation. Here, we see only a small difference in the precipitation amount due to the aerosols. In the three final hours of the simulation the precipitation rates have decreased about 14% (CV ∼ 0.5) for the SMK compared to the CTL case. These findings agree with Reutter et al. (2014) that the hydrometeor loading will increase with the smoke. Figure 8 shows the profiles of 1-and 99-percentile vertical wind, rain, and cloud water mixing ratio and smoke (domain averaged sum of hydrophobic and hydrophilic BC and OC). This figure shows convective activity in CTL and SMK cases in which the updrafts are enhanced (Figure 8a), and rain and cloud water are present in the domain (Figures 8b and 8c) relative to the dry cases. In fact, in CTLD and SMKD, the cloud water mixing ratio is negligible. Therefore, the enhanced updrafts are because of the cloud formation. As the cloud forms, the latent heat release from condensation makes the air warmer and more unstable, which rises faster than when the cloud does not form. Moist convection makes the plume rise and heat transfer more efficient. The faster updrafts result in increased surface winds due to conservation of momentum and ultimately leading to faster fire spread rates (see Figure 6). We notice that the rainwater mixing ratio is decreased. In contrast, the cloud water and  (Figure 8e) increased due to the smoke. A possible explanation (see Kablick et al., 2018;Reutter et al., 2014) for reduced rainwater is the higher number of cloud nuclei results in smaller droplets with a slower sedimentation rate when the moisture is limited. Finally, Figure 8d shows the profile of domain averaged smoke. The profiles demonstrate that enhanced mixing due to pyroconvective activity results in deeper penetration of the smoke into the troposphere, whereas for the SMKD case, the smoke remains within the limits of the boundary layer (∼2 km ). It should be noted the values for smoke are domain averaged, and within the core of the smoke plume, the smoke concentrations vary between 10 9 and 10 14 kg −1 . The profile of variability seen here is consistent in the timeseries (Figures 6 and 7) and the vertical profiles, where the standard deviations for hydrometeors are much larger than the surface winds or the burnt area. We highlighted the wind vectors with red and blue colors where the updrafts/downdrafts at 500 m are stronger than ±0.25 m s −1 , respectively. The convergence zones (areas where −∇ ⋅ is higher than one standard deviation from the mean; here 0.001 m −1 ) is marked with black contour lines. The black square box in (a) indicates the 2 × 2 km 2 area centered over the head of the fire and is used for calculating the timeseries in Figure 6.

The Influence of Atmospheric Moisture and Fire Emissions on Pyroconvective Cloud Formation
EGHDAMI ET AL.

10.1029/2022MS003288
10 of 17 Figure 9 shows the cross-section of water-vapor mixing ratio, winds, cloud and rain mixing ratio, and equivalent potential temperature for 2:25 hr after the simulation starts for the ensemble quantities. We chose this time because it corresponds to the peak convective activity before the first pyroconvective cloud dissipates. The cross-section is centered at the domain in the X-direction (as shown in Figure 5 with the black dashed line). For the dry cases SMKD (CTLD, not shown because it looks similar to SMKD), examining the wind fields ( Figure 9a) and equivalent potential temperatures (Figure 9d) shows that the convection due to the fire is very shallow and confined to the first 2 km near the surface only. However, for the control cases of CTL and SMK, both convective columns are very similar, with a cloud top of about 8 km (∼350 hPa ) based on the model results. The model cloud top agrees with the sounding observation (Figure 2), which shows a cloud top of at least 400 hPa . The differences between CTL and SMK are small but can be seen better in the time series (Figure 6). These figures clearly show a much stronger enhancement in the vertical wind in the upper levels than at the surface, which increases the safety risk for firefighting aircraft. The ensemble means presented are a smoothed velocity field. The domain maximum horizontal wind speed, downdrafts, and updrafts speed for the SMK case are about 16.5, −8.5, and 25 m s −1 , respectively. A movie of the same cross-sections but only for a single ensemble member is provided in Movie S1 to show the extreme winds. A direct comparison to observations requires the interpolation of the model to the radiosonde trajectory. Instead, we compared the temperature and dew point temperatures at three locations ( = 7, 7.5, and 8 km ) in the center of the domain ( = 9 km ) where the cloud forms (see Figure S4 in Supporting Information S1). The profiles are close to each other, indicating the presence of a pyrocumulus cloud between 700 and 400 hPa .
For an easier comparison of the wind fields between the SMKD, CTL, and SMK ensembles, we calculated the probability distribution of the vertical and horizontal winds below 300 hPa . We plotted the differences between the probability distributions in Figure 10. There is a positive and significant increase in the frequency toward the extreme values of the wind distributions for SMK compared to SMKD in vertical ( Figure 10a) and horizontal wind (Figure 10b). The positive increase in the tails can be seen between 4-8 m s −1 and 4-20 m s −1 for vertical and horizontal winds. There is no noticeable difference between SMK and CTL wind speed distributions indicating the smoke is less influential on the winds.

The Role of Surface Fuel Moisture Content and Latent Heat Release
A relevant finding in our study is that the surface fuel moisture content and latent heat release are less influential on the pyroconvective cloud initiation than the atmospheric column stability. The surface condition, especially the fuel moisture, can change the fire behavior near the surface. In our case study, we show the impact of the fuel moisture is smaller on the atmospheric profile moisture. Figure 11 shows the identical cross-sections of the water vapor, cloud mixing ratio, winds, and equivalent potential temperature for the SMK, FM4, FM8, and NOLH simulations at the same time. For both FM4 and FM8 (Figures 11b, 11c, 11f, and 11g) cases with relatively dry and wet fuel moisture content, we find that the pyroconvective clouds are forming in all the experiments. The cloud and rain mixing ratios have increased in FM4 compared to the SMK simulation (∼0.5 g kg −1 ), while in contrast FM8 the cloud and rain content decreased. Note that the drier fuel causes faster burns and more intense fire heat release.
In NOLH simulation, which fire latent heat release is omitted, there is less pyroconvective cloud content compared to SMK. Indeed, the cloud height for all the CTL, SMK, and NOLH is the same and about 8 km (∼350 hPa ). This reinforces the previously proposed hypothesis (e.g., Lareau & Clements, 2016) that although the fire heat release is essential for the cloud initiation, the moisture and heat are rapidly diluted above the surface. The following section will provide a more comprehensive summary of the burnt area and surface winds.

Sensitivity to the Hygroscopicity and Aging Turnover Time
The availability of more hydrophilic BC and OC for nucleation (i.e., a smaller turnover time) resulted in a higher cloud water path in all simulations ( Figure 12a). However, the rain rate was higher for a 4-hr turnover time compared to an 8-hr turnover time and then decreased for smaller turnover times (Figure 12b), suggesting that other factors, such as mois ture availability, may also contribute to rain production. For example, the absence of smoke in CTL simulation or smaller amounts of hydrophilic BC and OC in simulations with longer turnover time (K1T8 and K2T8) produced lower amounts of cloud water mixing ratio. In contrast, more available hydrophilic nuclei resulted in lower rain rates (K2T1). The microphysical behavior change is small (less than 20% increase or decrease) in the range of parameters we tested for hygroscopicity and turnover-time parameters and Figure 6. Time series of burnt area (black) and 10-m horizontal wind (blue) for SMK, SMKD, CTL, and CTLD cases. The surface winds are averaged for a 2 × 2 km 2 box that is centered over the fire front (see the black square box in Figure 4a). All the time series are averaged over the ensembles. The shading shows one standard deviation between the ensembles for the CTL case. All the time series are averaged over the ensembles. The shading shows one standard deviation between the ensembles for the CTL case. The values for dry cases are zero.
12 of 17 for short timescale simulations (<6 hr). Moreover, the fire behavior (burnt area and surface winds in Figures 12c  and 12d) is much more sensitive to available ambient moisture than smoke microphysics interactions. The lack of humidity in the CTLD and SMKD reduced the final burnt area by about 20%. This difference is about 10% in the near-surface wind. Although the surface wind speed (Figure 12d) and burnt area are directly related, this difference is because the wind is averaged and, therefore, smoothed out. Our results are consistent with the previous work on the hydrometeors loading and fire dynamic for pyroconvective clouds (Reutter et al., 2014). This small sensitivity of fire spread and dynamics to the smoke aging parameters in the model is promising given the high uncertainties related to the measurement and modeling of the aerosols. However, we believe that smoke may have a greater influence on cloud formation at longer timescales and in larger fire complexes, and its impacts should be studied using real case simulations. In our case, variables such as fire heat flux and moisture are more critical for reproducing a cloud structure similar to the observed pyrocumulus, compared to the impact of smoke.

Conclusions
We investigated the role of emitted carbonous aerosols, specific humidity profile, and fuel moisture content in the formation of pyrocumulus clouds. The motivation for this study is to demonstrate that a fire aerosol emission model integrated with the meteorological and air quality processes is needed to improve operational fire prediction systems. To this end, we linked the fire-produced BC and OC aerosols and water-friendly CCN in WRF-Fire to couple the fire smoke and microphysics modules. We have studied the impact of the fire emissions on the pyrocumulus cloud formation and development. Using a suite of LESs, we performed a systematic sensitivity analysis to determine the influence of atmospheric moisture, fuel moisture, latent heat release, and fire-emitted . Cross-sections of (a-c) water vapor mixing ratio, zonal and vertical winds (vectors) and (d-f) sum of cloud, rain, and frozen hydrometeor mixing ratio, and equivalent potential temperature (contour lines) at the center of the domain 2:25 hr after the start of the simulations for SMKD, CTL, and SMK cases. The cross-section is centered at the domain in the X-direction (as shown in Figure 5). All the fields are ensemble averages. Figure 10. Differences in probability distribution function of (a) vertical velocity and (b) horizontal wind speeds. The winds are from the 2 × 2 km 2 box that is centered over the fire front (see the black square box in Figure 4a) and below 300 hPa . The mean is removed from the horizontal wind field before the analysis.
14 of 17 aerosols on pyroconvective cloud initiation and evolution. The experiments presented are inspired and interpreted by comparison to the observations.
Our findings indicate that atmospheric moisture, which is directly related to available moist convective energy, has a relatively more important role in triggering the formation of deep pyroconvective clouds compared to fuel moisture. The convective cloud updrafts increase near-surface winds and consequently burnt area (by about 25%). The moisture released from the burning plays a less influential role in contrast to what is proposed by Potter (2005) and in the support of the observations by Lareau and Clements (2016). In fact, drier fuel led to more intense fire spread and heat flux resulting in higher wind intensities in our study. Our case study found that fire smoke has only a minor influence on pyroconvective dynamics, at least in the initial hours of the simulation. In our case study, particularly the final hour (5-6 hr), the cloud-water path increased while the precipitation rate decreased when compared with the scenario that included only background aerosols. In general, sensitivity analysis by using various hygroscopicity and aging turnover times showed that a higher number of aerosols available for nucleation results in increased cloud content and reduced precipitation, which agrees with previous studies (e.g., Reutter et al., 2014). However, the sensitivity to these parameters is very small in our case study (less than 20% variability in the microphysics), which is promising for modeling efforts since the uncertainties in modeling and measurements of fire emissions are very high. Because pyroconvective cloud formation is a complex phenomenon that depends on atmospheric thermodynamics and fire characteristics, the results presented here are case-dependent. It is worth noting that in the SCQ case, where the fire spread is downhill, which means the pyroconvective formation is unlikely due to slope-induced acceleration (Castellnou et al., 2022a(Castellnou et al., , 2022b. We have simulations (not shown) similar to SMKD and SMK (dry and moist conditions) where we increased and decreased the fire heat flux by 50%, respectively. In these simulations, the pyroconvective cloud did not form, which is expected according to previous studies and underscores the role of fire heat flux on the pyroconvective cloud formation. In simulations where the fire heat flux is constant, the formation of pyrocumulus clouds occurs in the presence of moist atmospheric conditions but not in dry conditions. Therefore, although high fire heat flux is a necessary condition for pyroconvective activity, it is not sufficient. Figure 11. Cross-sections of (a-c) water vapor mixing ratio, zonal and vertical winds (vectors) and (e-h) sum of cloud, rain, and frozen hydrometeor mixing ratio, and equivalent potential temperature (contour lines) at the center of the domain 2:25 hr after the start of the simulations for FM4, FM8, and NOLH cases. The cross-section is centered at the domain in the X-direction. All the fields are ensemble averages.

10.1029/2022MS003288
15 of 17 Our study used idealized simulation here only for a short time scale (<6 hr), as a first step toward the model development and understanding the fire emission impacts on fire weather. A comprehensive study of real scenarios where radiative impacts are included would be necessary to fully understand the impact of fire smoke on pyroconvective clouds and plume-driven fires. Another influential factor in the studied event is the effect of the terrain on the local wind circulation and fire spread. These aspects of the problem will be considered by running a real case WRF-Fire scenario with activating aerosol-radiation interactions, following the recent WRF-Solar developments that now include the impact of BC as outlined in Juliano et al. (2022) in future work. Finally, a comprehensive comparison with field measurements of aerosol concentrations and dynamical and microphysical properties of pyroconvective clouds is necessary to constrain the model parameters better.

Conflict of Interest
The authors declare no conflicts of interest relevant to this study. Figure 12. The ratio of (a) domain averaged cloud water path and (b) rain rate in the final 3 hr with respect to SMK simulation (SMK simulation is therefore 1.0). The ratio of (c) burnt area at the end of the simulation and (d) 10-m wind speed averaged in the 2 × 2 km 2 box centered near the fire front in the final 3 hr with respect to the SMK simulation. The sensitivity tests with variable turnover time and hygroscopicity are shown with the circles, and other simulations are shown with the numbers. The blue and red markers indicate hygroscopicity values of 0.2 and 0.1, respectively. The markers with a turnover time of 2 hr are slightly shifted horizontally to avoid overlapping. The CTLD and SMKD are not shown in (a) and (b) since the cloud does not form. Note that the CTL and CTLD runs do not include smoke emissions; therefore, the turnover time and hygroscopicity are not relevant to these runs.