We set up the analysis in Section 2 and provide a brief background on the data sets in Section 3. In Section 4.1, we compare results from the sub-canopy fires and in Section 4.2, we highlight the attributes that distinguish the grassland fire from the sub-canopy surface fires, thereby synthesizing the key insights from each of the three scenarios considered here. As with most data collected from prescribed burns, the current data were also occasionally influenced by wind variability. We take the opportunity to comment on the effect such variability may have on the fire while simultaneously attempting to summarize the general features that broadly characterize each scenario.We take this opportunity to comment on the choice of 1 hour moving averages in the analysis of the sub-canopy surface fires. Moving averages are computed in this work, instead of block averages, in order to preserve the resolution of the data and the smoothness of the signal. The selection of a 1 hour time period for computing the moving averages can be explained as follows. It has been documented in the literature that relatively long time periods are reasonable in computing eddy fluxes. To quote from Chapter 2 of the Handbook of Micrometeorology : ‘‘Eddy fluxes need to be formed over a sufficiently long time that any motions that contribute to the transport can be sampled adequately.
In practice,drainage collection pot this has meant that eddy fluxes have been calculated over time periods up to an hour in duration, sufficient for several of the largest planetary–boundary–layer scale to be sampled by the measuring system.’’ While this argument was made in the context of turbulence in the absence of a fire, it provides a reasonable estimate for the choice of the time period of averaging in the present study. Furthermore, for the backing fires of 2011 and 2012 discussed in this study, it was found that while the time required for the fire lines to pass the tower locations was of the order of minutes, the influence of the turbulence induced by the fire lines at the towers lasted for approximately an hour . One-hour moving averages generate a sufficiently smooth slowly-varying component for the velocity and temperature signals, which can be treated as the larger-scale atmospheric flow upon which the fire-induced fluctuations ‘‘ride’’. However, it is later seen from the MKE budget terms that the fire noticeably affects the mean flow even if we use a time window as conservative as 1 hour to separate the effects of the fire from the ambient flow. This, incidentally, gives us the advantage of being able to track the ‘‘new’’ mean state during FFP. Note that, in contrast to the present study, previous studies have computed block averages on the scale of a minute. Furthermore, they computed perturbations during the FFP period based on pre-FFP block means. That approach, in effect, treated all the coherent high- and low frequency velocity and temperature variations during the FFP period as fire-induced turbulence, thereby presuming no change in the ambient atmospheric conditions during the FFP period. Moreover, the over bar terms can be very sensitive to the averaging scheme and time window.
We have, therefore, included some information on the possible differences arising from using different averaging time windows in the form of a sensitivity analysis in the attached Supplementary Information. Only 1 hour of experimental data are available for the grassland fire. This makes it difficult to compute the slowly-varying parts of the velocity components and temperature as done for the sub-canopy surface fires, which is in turn used to compute the turbulent fluctuations. Another approach to compute the fluctuations would be to remove the 1 hour block mean obtained from the entire time series. However, as seen in Table 1, the spread-rate and intensity of the grassland fire is one to two orders of magnitude higher than the spread-rates and intensities of the sub-canopy surface fires. Moreover, the fire-induced turbulence intensity was found to be four to five times greater than the ambient atmospheric turbulence . Removing the 1 hour block mean to obtain the fire-induced turbulent fluctuations would have resulted in their underestimation during FFP. The relatively high intensity and spread-rate of the grassland fire suggests that the time scale of the influence of the fire at the measuring tower is comparatively much shorter , so that fire-induced changes can be considered entirely as turbulence without attributing any change to the mean state of the atmosphere. As mentioned above, tower-based data from four experimental fires are used in this study. We refer to the 2011 and 2012 backing surface fires beneath the canopy in the New Jersey Pinelands National Reserve as NJ2011 and NJ2012, the 2019 heading surface fire beneath the canopy in the NJPNR as NJ2019, and the heading surface fire from the FireFlux experiment in the grasslands of Texas as TX2006.
The 2019 burn unit is located at the Silas Little Experimental Forest within NJPNR, New Lisbon, New Jersey. For the sub-canopy burns, averaged half-hourly ambient wind velocity data in the stream wise and cross-stream directions are taken from nearby AmeriFlux towers: the Cedar Bridge Tower for NJ2011 and the Silas Little Experimental Forest Tower for NJ2012 and NJ2019. These data are provided by Clark and Heilman et al. ; they are shown in Fig. 1 for reference and comparison with the 1 hour moving means of the measured horizontal velocity components. For NJ2011, NJ2012, and TX2006, experimental burn data obtained from a single meteorological tower are studied in each case. For NJ2019, data obtained from two meteorological towers, i.e. the West Tower and a Control Tower, are studied for the entirety of this paper. The Control Tower is located outside the burn unit, 185 m away from its northern edge . This can also be seen in Fig. 1 of ref. Heilman et al. , which describes the burn unit and the location of the measurement towers. Measurement heights for each burn are shown in Table 1. It must be noted that the fuel consumption during the forest burns differed from each other. While NJ2011 occurred in a pine-dominated region, NJ2012 and NJ2019 occurred in an oak-dominated region. Although fuel loading of the forest floor was similar in both cases, consumption was relatively low in the oak-dominated burns. This is typical of hardwood-dominated forests: fuel consumption is usually less than that in forests with more pine trees and saplings. For a complete description of the burn experiments, including but not limited to detailed illustrations of the burn plots, their ignition lines, and the respective directions of fire line propagation relative to the ambient winds,10 liter pot we refer the reader to the works of Clements et al. , and Heilman et al. . Most of the important features of these four sets of data are summarized in Table 1. Note that although the pre-, during, and post-FFP periods indicated in Table 1 bear some similarity to the corresponding periods defined in the works mentioned above, they are not exactly the same. While the 1 hour moving means explored in the previous section are useful in studying the evolution of the slowly-varying mean, a first-order statistical analysis of the horizontal velocity components is required for insights into the frequency and direction of the strongest turbulent fluctuations a few minutes before, during, and after FFP. This is achieved with the help of a 2D histogram plotted on a wind compass, plotted using WindRose in MATLAB . This function groups velocity data from a time series into classes based on their magnitude, while preserving their direction relative to the positive stream wise direction. Note that zero degrees represents the positive stream wise direction on the wind rose in each scenario. Interesting observations can be made from the wind-rose statistics for the backing surface fires. As seen from Figs. 2–, a shift is seen at ℎ = 20 m in 2011 from high stream wise variability pre-FFP to high cross-stream variability during FFP and post FFP . This corroborates well with the increase in the magnitude of ��20 as seen in Fig. 1. In fact, the cross-stream winds show considerable variability in the range of 6 to 12 m/s during FFP as opposed to post-FFP suggesting that this effect is due to the presence of the flame. Comparison with the ambient cross-stream wind velocity, which shows a decreasing trend during and after FFP at the measurement tower, also emphasizes that the increased cross-stream variability is attributed to fire-induced entertainment.
We can also attribute this to the diversion of momentum into the cross-stream direction as the stream wise wind competes with the air entrained by the fire from the burnt region. Additionally, crown torching in the stand, especially within 15–20 m of the measurement tower, would have also induced strong horizontal velocity gradients near the canopytop, leading to fire-induced entertainment in the cross-stream direction near ℎ = 20 m. Note that the shift in the cross-stream direction happens late at ℎ = 3 m since it takes a while for the cross-stream wind near the canopy top to force the wind near the ground surface. At both ℎ = 3 m and 20 m in 2012, we see a noticeable shift from high variability in the cross-stream direction pre-FFP to high variability mostly in the stream wise direction during FFP and post FFP . This is attributed to the high ambient mean stream wise wind velocity as observed in Section 4.1.1. Furthermore, the relatively lower fire intensity in 2012 does not cause noticeable entertainment in the cross-stream direction during FFP. Contrast this with the FFP period in 2011 and 2 during which the cross-stream winds attain speeds in the range of 6 to 12 m/s. We now focus on the wind-rose statistics for the heading fire, i.e. NJ2019. At the West Tower, the pre-FFP, FFP, and post-FFP times are 1450 to 1520 LT, 1520 to 1550 LT, and 1550 to 1620 LT, respectively . An increase in stream wise variability is seen at the expense of cross-stream variability at ℎ = 20 m from pre-FFP to FFP times possibly because of entertainment of air from the upwind side of the fire at the West Tower. The considerable variability in the cross-stream direction post-FFP is, again, attributed to the ambient wind, which undergoes a southerly shift from the FFP time to post FFP time and has a strong cross-stream component. However, we expect the strong ambient cross-steam wind to also divert some of the stream wise eddies and momentum into the cross-stream direction. This would mean that while the cross-stream gust adds its own momentum to the canopy from aloft, it also potentially diverts some of the fire induced turbulence near the canopy top into the cross-stream direction. Moreover, the wind statistics near the surface seems to mimic that of the higher height suggesting a strong forcing by the ambient post-FFP cross-stream wind.Studies on turbulence during management-scale experimental burns conducted in differing conditions of wind and vegetation are relatively isolated. The integrated contextual framework provided in this work represents a major stride towards the synthesis of and comparison among such studies. Through the comparison, we have reinvestigated fire-induced turbulence dynamics with the backdrop of canopy turbulence or ABL turbulence from a fundamental standpoint. The effects of changes in ambient wind conditions on the measured data have been taken into account because of their linkage to fire behavior. While the inferences drawn here have also been informed by local meteorological conditions at the time of the burn experiments, we have successfully encapsulated the coherent patterns that broadly characterize fire-induced turbulent flow in the two environments primarily by examining turbulent fluxes and the TKE budget terms. We have also examined the terms of the MKE budget equation for all of the sub-canopy surface fire experiments. The sinking of air under gravitational acceleration or the rising of air due to buoyant updrafts accounts for most of the variability in the MKE tendency during and after fire-front passage. The ‘‘slowly-varying’’ mean flow responds noticeably to the presence of the fire and cannot be equated to the pre-FFP mean flow. In this study, major differences among three scenarios of surface fires have been highlighted: sub-canopy heading fires, sub-canopy backing fires, and heading grassland fires.