Effects of Fire Plumes From Pine Needles on Small-Scale Fading in Radiowave Propagation

In this article, radiowave fading caused by fire plumes due to the burning of heaps of pine needles is studied. Detailed characterization of the fire impact in terms of excess loss due to the yielded plumes at UHF frequencies, mainly in the L-band, when considering two different ignition sources, is also presented. The implemented small-scale fire scenario is composed of a 1 $\times 1$ m metallic grid covered with a 0.6 kg pine needle fuel bed that was placed right above a circular and linear gas burner to mimic an underbrush scenario. The propagation phenomena were evaluated through specific wideband measurements as the fire developed while intersecting the radio path at the antennas’ boresight. Results demonstrate changes in the propagation medium that is strongly influenced by the plume, which ultimately becomes (partly) ionized. Deep fades of up to 12 dB below median levels at specific frequencies during a short period of time, when the fire is more intense, were measured. We also obtained an overall median excess loss due to fire of up to 2 dB in the frequency range from 0.6 to 2 GHz in a relatively short path (below 1 m). A time-domain analysis allowed us to expand these results yielding fading values of the direct component of more than 3 dB in the presence of fire. Extending the findings presented in this article to a real-scale fire scenario, in which median excess loss of more than 20 dB and deeper nulls are ought to be expected, sets the rationale and motivation to pursue future work on the topic, with practical relevance to future mission-critical communication networks planning, such as, but not limited to, LTE Public-Safety and specific 5G use cases.


I. INTRODUCTION
R ADIOWAVE propagation in vegetation media and its experimental excess loss assessment under fire conditions is a rather important topic when communication reliability is at stake, particularly in remote rural areas when affected by severe wildfires. In events in which fire spreading gets out of control, emergency radio communications, which are usually employed to aid fire suppression operations, should provide the required resilience for effective disaster relief operations [1]. While the topic of radiowave propagation in vegetation has been extensively studied in the literature [2], [3], [4], [5], [6], the effect of fire on top of the attenuation caused by vegetation lacks more research. Therefore, the study of radiowave propagation in such scenarios is of uttermost importance in improving communication resilience for effective disaster relief operations to create an effective response and, ultimately, to keep people safe.
Radiowave attenuation in wildfires is closely related to the electron density present in the fire environment and to the effective collision frequency [7]. The alkali and alkaline earth metals (A-AEMs) present in vegetation, particularly the ones with low ionization energy, are transported into the combustion zone through air convections and are mainly responsible to generate electron populations, which are released during pyrolysis stage [8].
In recent years, only a few studies have been performed to understand the impact of flames on signal propagation. Schneider and Hofmann [9] have investigated the effects of absorption and dispersion phenomena of microwave signals while propagating through flames and the dependence of conductivity and optical constants on weakly ionized gas. A similar study was presented by Hata and Doi [10], at 40 GHz. Boan [8] presented measurement results of broadband radio propagation under fire. The results demonstrated that vegetation combustion has considerable effects on radiowave propagation, particularly at VHF and UHF frequency bands. Mphale and Heron [11] and Mphale et al. [12] measured microwave signal losses of 1.6-5.8 dB in a controlled pine litter fire. More recently, Li et al. [13] studied the influence This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ of flame and smoke on the attenuation of electromagnetic (EM) waves. It was demonstrated that EM wave loss in a fire environment is frequency-dependent since attenuation values of 5.43 and 0.26 dB, at the same smoke concentration, were measured at 360 and 400 MHz, respectively.
It is known that the flame itself has insignificant effects on wave attenuation; however, the generated smoke has a major role [13]. The smoke generated in wildfires may be defined as weakly ionized media. When an EM wave propagates through such a medium, electrons are accelerated by the electric field of the incident wave and acquire energy. Part of the absorbed energy is dissipated as heat, through collisions between neutral particles, ions, and molecules. This dissipation makes the incident wave amplitude to be attenuated along the distance [14].
The work presented herein is thought to be the first laboratory study on the effects of burning forest litter on radiowave propagation in a fully controlled environment when considering two different fire front spreadings, i.e., linear and circular ignition sources. The experimental setups were implemented in order to obtain a clear and homogeneous flame, which has a direct impact on the generation of the ionized gas during the combustion process. Signal attenuation caused by fire using two different burners was evaluated, for a sample of 0.6 kg pine needles heap exposed to fire. This work is also considered to be important in the validation of simulation models, such as the fire dynamic simulation (FDS). Once validated, the simulation tools will allow for the generalization of the fire scenarios for other configurations.
This article is organized as follows. Fire ignition and combustion process are discussed in Section II. Section III presents the scenario outset and the measuring equipment configuration. Result analysis are presented in Section IV, and conclusions are outlined in Section V.

II. FIRE INITIATION AND IONIZATION PROCESS
Although it is out of the scope of this article to address the actual physical (and chemical) phenomena, for completeness, it is important to give the context in which vegetation media exposed to fire may affect the propagation of radio signals. Combustion can be described as a specific group of chemical reactions where oxygen and fuel burn together at temperatures high enough to generate heat and combustion products [15]. There are three critical elements in order to guarantee combustion: fuel, oxygen, and heat. When associated, these parameters originate what is called the fire triangle, as presented in Fig. 1. In case any element is missing, the fire will cease [16].
This research field is closely related to the vegetation combustion process, which can be split into four distinctive stages: ignition, flaming, smoldering, and extinction. Ignition, which is conditioned by fuel characteristics and environmental factors, is the stage where fuel is ready to burst into flame. The glowing area generates small flames with a limited rate of heat production [17]. In the flaming stage, the combustion process itself begins. The presence of high temperatures has a major influence on biomass drying, which is followed by a pyrolytic step. At a temperature of about 400 K, when pyrolysis starts, cracking of the fuel molecules occurs [17].   [19] During this process, char, tar, and gaseous state compounds are generated. In the smoldering stage, the pyrolysis rate slows down, fewer flammable compounds are produced, and the ash content increases [17]. At extinction, the burning process ceases completely.
As introduced in Section I, signal attenuation may occur in wildfire scenarios due to the presence of free electrons released during vegetation combustion. There are two main mechanisms responsible for producing ions in flames: chemiionization and thermal ionization [18]. While chemi-ionization is associated with the chemical creation of charged particles in hydrocarbon reactions, thermal ionization involves low ionization energy elements and high temperatures for ion generation.
In a fire environment, when a considerable amount of effective ionizable fuel is available, thermal ionization is the predominant mechanism responsible for creating an electron population. The fuel, which should contain A-AEM with low ionization energy, has electrons attached to atoms that will be thermally excited to a state of free electrons [19]. The dominant chemical compounds in vegetation are lignin, hemicellulose, cellulose [19], and inorganic matter components. Among all the inorganic components, potassium (K), calcium (Ca), and magnesium (Mg) are crucial to determine signal attenuation in wildfires [19]. The ionization energy for these three elements is presented in Table I.
When considering a weakly ionized media as foreseen in the cold plasma model (CPM) [20], as the one generated during vegetation combustion, the propagation constant for such media is determined by the following equation [11]: where the real part of γ corresponds to the attenuation constant and its imaginary part to the phase constant. In this equation, ω is the signal angular frequency, µ 0 and ε 0 are the vacuum permeability and permittivity, respectively, and ε r is the relative permittivity of the plasma. This complex Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.
relative permittivity can be found considering parameters such as plasma frequency (ω p ) and the effective collision frequency (ν e f f ), as indicated in the following equation: where the plasma frequency is given by the following equation [21]: with N e being the electron density of the plasma, e the electron charge, and m e the electron mass. When considering the electron-neutral collisions as the dominant interactions, the effective collision frequency ν e f f in (3) can be written as follows [11]: where N m is the density of air molecules, a is the radius of air element molecules, and T is the absolute temperature. While the radius of the particles present in combustion, a, is always the same independently of the medium, since the particles being burned are always the same, the density of air molecules and the absolute temperature of the combustion depend on the type of material being burned [11]. In general, if the dielectric permittivity of a medium has a nonzero imaginary part, it will cause attenuation on the signal propagating through it, and the real part is different from unity, then it will also have an impact on the propagation speed [22].

III. EXPERIMENTAL SETUP
In order to analyze how fire environments may affect radio communications, two small-scale fire scenarios using linear and circular burners were tested. Figs. 2 and 3 depict the circular and linear burners, respectively, as well as the adopted measurement geometry and setup. The burners used for the measurements are made of steel tubes with diameters of 3 cm and with small circular holes with diameters of 2 mm separated by 10 mm between centers. The circular burner has a diameter of 0.51 m, while the linear burner has a length of 1.06 m. In both scenarios, the burners distribute propane gas that creates flames once ignited. However, as already mentioned, the influence of fire on the radio path depends on the gases that are released by the combustion of vegetation-type fuel. Thus, a metallic grid covered with 0.6 kg of pine needles was evenly distributed on top of the burners (22 cm above the table) for the purposes of creating gases that may interact with the radiowave propagation in the antenna's boresight. Pine needles were chosen as fuel for this initial set of experiments to simulate the forest (floor) litter since pine needles are known to be a common fuel present during fire seasons that enable the rapid propagation of fire. Besides, by restricting the forest litter used for the measurements to only one kind, it is easier to assess the fuel characteristics (such as density and humidity). Future work should address other constituents of trees such as twigs, branches, and trunks although their fire spread is much slower than those of grassland and leaves. For completeness of scenario characterization, the room was at a temperature of 23 • C, and the fuel and air humidity were 41% and 14.5%, respectively.
For these measurements, two antipodal Vivaldi antennas (BAVA), as shown in Fig. 4, were used. The Vivaldi antenna design was chosen due to its wide frequency bandwidth to cover the frequency range of interest, particularly for land mobile communications, which includes mission-critical communications, in addition to its high gain. In addition, the ease of production and low manufacturing costs make this antenna ideal for measurements in the proximity of the fire. Despite appropriate safety measures being in place during the entire measurement campaign, the exposure of the antennas to high temperatures in the vicinity of the fire may cause structural damage to the antenna over time. Thus, the rationale for using low-cost and easy-to-fabricate antennas can be easily replaced. These antennas were designed and optimized in a full-wave EM solver CST MWS, leading to the final design yielding dimensions of 290 × 288 mm 2 . The antennas were manufactured using printed circuit technology on an FR4 substrate. Impedance measurements of the antennas showed an operating frequency band from 600 to 8000 MHz [23].
The transmitter and receiver antennas were placed at a distance of 2.36 m from each other with their boresight at 33 cm above the fuel bed to allow sufficient concentration of fire-ionized gas. Both antennas were connected to a Rhode and Schwartz ZVM vector network analyzer (VNA) to retrieve the  transmission coefficient (S 21 ). At the receiver side, there was a low-noise amplifier between the antenna and the VNA to increase the signal-to-noise ratio and obtain a better dynamic range for the measurement system. The VNA was configured to frequency sweep mode, using a frequency bandwidth of 7.4 GHz, from 0.6 to 8 GHz. An averaging factor of 30 samples was employed at each frequency step in order to minimize the fast fading effects, hence yielding a sweep time of 13 s, each with 445 frequency points.
The collected measurement data may be affected by multipath due to reflections from both the metallic grid and table holding the burner, as shown in Fig. 5(a). Such reflections can affect the received signal level either constructively or destructively. As a means to quantify the effects of these reflections, a first measurement was taken considering only the metallic grid without the pine needle fuel bed. By applying an inverse fast Fourier transform (IFFT) to the complex frequency response retrieved by the VNA, the impulse response results, normalized to the first arrival, were obtained. These results are depicted in Fig. 5(b) for the circular burner scenario, where the peaks for both the reflection on the grid and the table are highlighted. These yielded relative delays (to the direct path) of 0.27 and 0.81 ns, corresponding to distances of 8.1 and 24.3 cm, respectively. This is in good agreement with the setup geometry and its physical path. Since the spatial resolution of this measurement was 4 cm, obtained from a frequency bandwidth of 7.4 GHz, it may be assumed that the peaks presented in Fig. 5(b) correspond to the multipath components reflected from the grid and on the table. This assessment validates the propagation phenomena in the measurement scenario.

IV. RESULT ANALYSIS A. Circular Burner
For the purpose of evaluating the excess loss caused by the presence of fire, the setup depicted in Fig. 2 was assembled (with grid, fuel bed, and so on), and measurements were taken with the burner turned off, turned on, and later with the presence of fire. These experiments allowed proper evaluation of the excess loss caused by the fire presence, which is expected to be both frequency selective and dispersion (timevarying). For the measurements without fire, referenced as no fire (NF), three frequency sweeps were considered to check the measurement repeatability under stationary channel conditions. This way, it is ensured that any time-varying effect is mainly due to the fire propagation phenomena. In the measurements with fire, ten frequency sweeps were logged, corresponding to the 130 s taken by the fire to consume the entire pine needle bed. This was the time considered for the measurement, which is the actual duration required to consume all the fuel placed over the metallic grid. If less time were to be considered, it would not be possible to analyze the combustion of whole fuel; if more time was to be considered, there would have been samples in which no fuel was being burned.
The results from these measurements are presented in Fig. 6, as well as some illustrative pictures of the scenario along with their respective thermal profiles. For this analysis, only the frequency band between 0.6 and 2 GHz was considered, as the emergency systems operate at the lower frequencies of the measured frequency band. As observed in Fig. 6(b)-(d), the results of the measurement with fire were split into three different datasets, corresponding to the initial stage of the fire (three frequency sweeps), the middle stage (three frequency sweeps), and the final stage (four frequency sweeps). In the initial stage, the fire was burning in the central part of the grid; in the middle stage, the fire front diverged from the center to the periphery; and in the final stage, the fire was only burning in the periphery of the grid and was already subsiding. To further simplify explanations and analysis on each individual dataset, these were represented as S1 (i.e., stage 1), S2, and S3, respectively. Fig. 6(a) depicts the results of the received power obtained for the scenario with the fuel bed. These results were normalized using as reference the results obtained without the pine needles as follows: where S fire 21 ( f ) and S reference 21 ( f ) represent the measurements with fire and reference, respectively, and S norm 21 ( f ) is the resulting normalized S 21 parameter. By using this case for normalization purposes, the unwanted multipath contributions from the propagation environment, including the reflections on the table used to deploy the fuel heaps, are minimized. Thus, the fire effects are isolated, and the propagation path through the fire plumes may be studied. As depicted in Fig. 6(d), the transmission coefficient approaches the reference values [see Fig. 6(a)], given that, at this stage, the fuel has already been consumed by the flames, and therefore, the radio channel becomes almost stationary. Moreover, these results prove the stability of the VNA. As such, it can be concluded that all the differences between sweeps in S1, S2, and S3 are due to the presence of fire. In S1, mostly concentrated in the line-of-sight region, there are attenuations as high as 12 dB and, thus, confirming the time-variability induced by the fire. We also verified higher temperatures in this stage of the fire, especially where the fire and fire plumes intercept the radio link, reaching temperatures between 500 • C and 550 • C. As time passes and the fuel is consumed, the attenuation values obtained diminish, as observed in S2 and S3. The temperatures also tend to diminish with time. In S2, the attenuation reaches a maximum of 5 dB. Finally, in S3, the four sweeps are very similar, and the normalized received signal level is concentrated around 0 dB, meaning that it is comparable to that of the reference measurement (i.e., without fire and pine needles).
The measurement results are also presented in terms of the cumulative distribution function (CDF) when considering 130 s of observation. The CDF results of frequency-selective fading phenomena for varying frequency bands, in the range of 600 MHz to 2 GHz, are depicted in Fig. 7(a). In the latter, it can be concluded that S1 presents higher values of attenuation, which shows that, when the radio link is completely obstructed by the fire, there is a considerable release of particles that introduce losses. As the fire spreads out and the fuel heap is consumed, the excess attenuation seems to decrease to zero, as in S3. Even though the fire from the gas burner is still present, the absence of pine needles yields a frequency response similar to reference one, i.e., without both fire and fuel. To further prove this point, a comparison of the scenario with and without fire (and both cases without the pine needles) is shown in Fig. 8, where the data of three sweeps retrieved by the VNA (not normalized) are presented.
We verify that the curves are very similar, meaning that the propane gas released by the burner has a residual influence on the received signal level, further proving that the alterations shown in Fig. 6 are due to the A-AEM particles released by the combustion of the fuel heap and not by the burner fire. Note that, as there was an amplifier in the receiver, some frequencies present S 21 higher than 0 dB. Moreover, the frequency-selective fading was analyzed at 100 MHz subbands. The results are presented in Table II,  in terms of the median of recorded fading values across the specific subband of each stage. As an example, Fig. 7(b) shows the CDF obtained for the 1.1-1.2 GHz band. In this figure, the effects of a fire during the combustion are even more pronounced, where the progression from S1 to S3 is visible, as the median of S1 for this band is approximately −5 dB for S1 and starts to approach 0 dB in S3. For S1, the median of attenuation obtained for the whole bandwidth analyzed was 1.9 dB. Moreover, the highest values of attenuation were between 0.9 and 1.6 GHz, achieving median values as high as 5.9 dB of attenuation. In S2, the attenuation was 0.8 dB, which is lower than for S1, in agreement with the results discussed before. However, for this case, the frequency band that achieved the highest attenuation was from 0.9 to 1.1 GHz. Finally, in S3, the normalized received power was close to 0 dB, being always higher than −1 dB as the measurements with fire are similar to the ones without fire due to the consumption of fuel. Furthermore, we emphasize that the attenuation in S3 tends to be lower than in NF. This is due to the absence of fuel over the grid in S3 that introduces some signal attenuation in NF.
To further complete the analysis, the time-domain representation of the previously presented results was also calculated. By applying an inverse Fourier transform (6) to the data in the frequency domain, the response in the time domain can be Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.  The results for all the measurement sweeps were summarized in Table III, where the trend of attenuation is visible to approach the reference as the fire of the pine needles extinguishes. Note that all the values of this table are normalized to the reference measurement so that the direct component (at antenna boresight) was set to 0 dB. This way, the results presented in the table represent the variation of attenuation of the direct component peak compared to the reference measurement.
Higher attenuations were obtained in the initial stage of the fire (Sweep 2 of S1), with an attenuation of higher than 3 dB compared to the reference. As the fire progresses, the attenuation decreases to 1 and 0.3 dB, at S2 and S3 stages, respectively. These results confirm the ones obtained in the frequency domain, where it was shown that higher attenuations were found in the initial stage of the fire when the fire front containing the gases from the combustion of the fuel was wider. Moreover, the case without fire presented higher attenuation values than those of the final stage of the fire. This is due to the fact that, by introducing the pine needles above the grid, the clearance of the first Fresnel zone decreased, leading to more signal attenuation. However, as the fire progresses, the pine needles burn away, eventually yielding Comparison of measurements with theoretical modeling for the circular burner scenario.
to ashes, and subsequently, the signal strength becomes closer to the reference level.
This attenuation of the direct component follows the same trend of the frequency response, where higher signal variation occurs for the initial fire stages and, thus, suggests the appearance of higher concentrations of A-AEM.
In order to further validate these results and link the measurements performed with the modeling shown in Section II, (1)-(4) were computed for different values of the effective collision frequency and electron density. Equation (1) provides the propagation constant; hence, (7) must be used to estimate the total signal attenuation [19] att = ℜ(γ ) × 20 log 10 (e) × fire dist (7) where att is the total attenuation in dB, ℜ(γ ) is the real part of the propagation constant, and fire dist is the distance that the propagating wave travels through the fire, in meters. A similar experiment presented in [12] concluded that, during the combustion of pine litter at temperatures of up to 807 • C, the electron density and the effective collision frequency reached values ranging from 0.51 to 1.35 × 10 16 m −3 and from 3.43 to 5.97 × 10 10 s −1 , respectively.
Using the values found in [12], the propagation model (7) was then used to compute the estimated signal loss during the experimental fires conducted herein, and the results provided by this model were then compared with the sweep that presented the highest attenuation, i.e., sweep 2, for different values of N e and v eff . For this particular sweep, a fire column width of 0.5 m can be inferred, from the photographs and thermal images. The results of this analysis are depicted in Fig. 9. To this extent, the model (7) was computed using the set of values extracted from [12] that yielded a more pronounced signal attenuation (Ne = 1.35 × 10 16 ; v eff = 3.43 × 10 10 ) and with the set of values that produced the lower attenuation values (N e = 5 × 10 15 ; v eff = 5.97 × 10 10 ). These curves are represented in Fig. 9 in purple and green, respectively.
Moreover, model (7) was computed for several values of N e and v e f f aiming at finding the set of values that minimize the root mean square error (RMSE) between both the simulated and measured data. The obtained set of values was N e = 7.58 × 10 15 and v e f f = 5.08 × 10 10 , which is represented in Fig. 9 by the red line. As expected, the values of N e and v e f f obtained from this analysis are within the range of values obtained in the experimental study presented in [12] since similar fuel was used.
Finally, during these experiments, unexpectedly increased values of attenuation were registered at frequencies ranging from 0.9 to 1.2 GHz. This frequency-dependent behavior is not foreseen by the CPM model and might be related to other propagation phenomena. Hence, model (7) was also computed to minimize the error between measured and simulated curves, excluding the values between 0.9 and 1.2 GHz, which yielded values of N e = 5 × 10 15 and v e f f = 6.06 × 10 10 , represented by the yellow line in Fig. 9. Using these values, different values of N e and v e f f were obtained, yielding values slightly outside the range found in [12]. However, this difference can be related to the fact that higher temperatures will yield higher values of signal attenuation, and the maximum temperatures registered in [12] were higher than those obtained during the experiments presented herein.

B. Linear Burner
The results obtained for the linear burner (see Fig. 3) are presented in this section.
A similar designation for the sweeps performed in this case was adopted: S1 refers to sweeps 1-3; S2 refers to sweeps 4-6; and, finally, S3 refers to sweeps 7-9. Fig. 10 shows the normalized received power for the same three stages considered before, as well as the case without fire. Photographs and thermal images of each stage are also presented for clarification.
From Fig. 10(a), it can be inferred that the channel remains stationary and the VNA stable, as all of the three sweeps considered are very similar and do not present great disparities. By analyzing the measurement results in the presence of fire, it can be seen that S1 [see Fig. 10(b)] presents higher differences between the different frequency sweeps, especially when compared to S2 and S3 [see Fig. 10(c) and (d)], achieving excess attenuations as high as 13 dB. However, contrary to the results for the circular burner scenario, the attenuation variation between frequency sweeps stabilizes during S2. This happened because of the different types of fire spreading away from the burner compared to the circular burner. During S1, the fire starts as a curtain along the burner but then spreads to the edges of the grid in S2, increasing the depth of the fire column that intercepts the propagation path. However, in S3, there are already two fire fronts but with less intensity than the starting fire leading to less attenuation as the measurements converge to the reference (around 0 dB). Nonetheless, the effects of the fire in the transmitted signal are clearly visible, especially at the initial stage of the fire, where differences between sweeps up to 10 dB were obtained.
For this scenario, a statistical analysis similar to the one previously presented for the circular burner was also performed. These results are also depicted in Fig. 11(a) and (b), for the entire frequency band considered and for the subband of 1.1 to 1.2 GHz, respectively. Similar to the circular burner scenario, this frequency band was chosen for this case as it presents greater values of attenuation and variation, as seen in Fig. 10(b). However, contrary to what was observed in the circular burner scenario, a significant difference between the four different stages in terms of CDF for the whole frequency band was not found. This is due to the smaller path traveled by the signal across the fire, leading to smaller dispersion of the results. Moreover, for the subband represented in Fig. 11(b), it was found that the attenuation values of S1 are distributed between 12 and 1 dB. In S3, even though some values below −5 dB were measured, the vast majority of the attenuation obtained (90%) was between 5 and 0 dB. The same trend of the values of attenuation approaching the NF case as the fire progresses can still be visible.
The summary of the statistical approach for the linear burner scenario is presented in Table IV. From these results, the highest values of attenuation were again observed between 0.9 and 1.6 GHz for different fire stages. Remarkably, for the majority of the cases presented, the median of attenuation in S2 is higher than the one in S1 and S3. Such behavior is thought to be due to the fire starting to divide into two fronts due to the geometry of the setup around this time, leading to a thicker fire front. This is clearly shown in Fig. 10(f) and (j) where only one fire front is visible in S1, while, in the middle stage, a thicker fire front can be seen in Fig. 10(g) and (k). In terms of the final stage (S3), the results obtained are close to 0 dB, as almost all of the fuel is burned, meaning that the measured values are approaching the reference. Even though there are three fire fronts still visible, the central one does not have fuel, meaning that it will not cause additional attenuation, as discussed in the circular. The two peripheral fire fronts had smaller intensity as the fuel was almost entirely consumed leading to a smaller impact on the radio path.
In addition, the time-domain analysis was also performed for the linear burner scenario. Table V shows the normalized  attenuation of the direct component compared to the reference   TABLE IV  MEDIAN OF THE MEASURED NORMALIZED RECEIVED POWER  FOR THE LINEAR BURNER SCENARIO FOR DIFFERENT  FREQUENCY BANDS (IN dB)  Contrary to what happened in the circular burner, higher attenuation is obtained during both the initial and middle stages of the fire, with maximums of 1.87 and 2 dB, respectively, where the fire front starts to widen and break into two fronts. Yielded attenuations were smaller than the ones for the circular burner (where the maximum was higher than 3 dB) as the amount of fuel being consumed in each time instant was smaller, and the fire front was narrower. These results also agree with the ones previously presented in the statistical analysis, where higher median attenuations were obtained in the first two stages of the fire. Also, in agreement with both statistical analysis and results for the circular burner, the attenuation converges to the reference case as the fire extinguishers.
These results show that the wider the fire front that contains the A-AEM, the greater the impact the fire plume will have on the radiowave propagation.
Similar to the case of the circular burner, a comparison between the measured results and the theoretical curve was also implemented for the linear burner when considering the sweep with higher attenuation (sweep 4), as shown in Fig. 12. Note that, for this case, the width of the fire front is not as thick as in the previous case. From the analysis of the photographs and thermal images, one could infer a width of 0.25 m.
The same interval of values for the electron density and effective collision frequency was used, as the fuel being burned was the same. The values for these parameters that yielded smaller error were N e = 8.09 × 10 15 and v e f f = 4.94 × 10 10 . When the anomalous region is not considered (in this case from 1.1 to 1.6 GHz), better results were obtained for N e = 5.17 × 10 15 and v e f f = 5.38 × 10 10 . The maximum and minimum attenuations were obtained for N e = 1.35 × 10 16 and v e f f = 3.43 × 10 10 and N e = 5 × 10 15 and v e f f = 5.97 × 10 10 , respectively.
It can be observed that, for both cases, most of the measured attenuation values are between the theoretical curves. However, some phenomena may be affecting the results that are not considered in the CPM, such as the attenuation due to the ashes being released during combustion, as the CPM only considers the effects of plasma on the propagation. Moreover, as the combustion leads to a rapid change of propagating media, during a sweep of the VNA, the electron density and effective collision frequency may not be constant.

V. CONCLUSION
An experimental study of two small-scale fire scenarios was presented, for circular and linear propane gas burners, where a fuel bed of 0.6 kg of pine needles was distributed over a 1 m 2 area. This is thought to be the first study in a completely controlled (bench) environment on a small scale that enabled us to ascertain whether the fire has a real impact on radiowave propagation using two different ignition sources and fire spreadings. These results will provide important input data for the modeling of fire scenarios using different simulation tools, which, once validated, will allow for the generalization of other scenarios and configurations.
In these measurements, temperatures of up to 900 • C were registered during the 130 s combustion of the pine needles, yielding maximum attenuations of 13 dB for sweeps at the start of the fire. Measured median fading values of around 2 dB were recorded at the beginning of the fire for the circular burner scenario when a frequency band from 0.6 to 2 GHz was considered. For the scenario with the linear burner, the attenuation values presented a median of around 1 dB at the start and middle stages of the fire.
The results obtained from the measurements show that the circular burner yields higher values of attenuation than the linear burner, namely, in the first stage of the fire, since it produces a wider fire column. However, in both scenarios analyzed, there appears to be a frequency band from 1 to 1.3 GHz in which radiowave fading is higher. Aside from this, there does not seem to be other tendencies with the increase or decrease in frequency for the considered frequency band.
The results were also analyzed in the time domain, where a good agreement with the attenuations calculated in the statistical approach was achieved.
It was also demonstrated experimentally that fire without vegetation (only from the release of propane gas from the burner) had no effect on the transmitted signal, proving that it is the presence of the A-AEMs in vegetation that affects communications. These results also prove that the larger the A-AEM quantity in a vegetation volume, the greater the excess signal loss will be.
Finally, the measured results were compared to the theoretical curves obtained from the CPM, where typical values for electron density and effective collision frequency were computed. These results showed that, for the majority of the band, the attenuation observed during the measurements was in accordance with the theoretical modeling.
The results presented herein show that the combustion of 0.6 kg of pine needles distributed over 1 m 2 introduces significant attenuation and fading in radio communications, and these values are expected to increase with the fuel quantity, which can compromise the performance of emergency radio communication systems during wildfires. Further work will address an experimental validation at larger scale fires for varying vegetation species and dimensions. He is currently a Professor of telecommunications with the Department of Science and Technology, Escola Naval, Instituto Universitário Militar, Alfeite, Almada, Portugal, and a Researcher with the Centro de Investigaçõo Naval, Almada, and the Instituto de Telecomunicações, Lisbon. He has coauthored more than 40 technical papers in peer-reviewed international journals and conference proceedings. His current interests include microwave imaging, antennas for biomedical applications, 3-D-printed antennas, radiowave propagation, and machine learning applied to microwaves. He joined IST in 1980, where he is currently a Full Professor of microwaves, radio wave propagation, and antennas with the Department of Electrical and Computer Engineering. He is a Senior Researcher with the Instituto de Telecomunicações, Leiria, Portugal, where he is a member of the Board of Directors. He has coauthored one book, three book chapters, more than 250 technical papers in peer-reviewed international journals and conference proceedings, and seven patents in the areas of antennas and radio wave propagation modeling. His current research interests include antennas for millimeter-wave applications in 5G and satellite communications, RFID and ultrawideband (UWB) antennas for the Internet of Things (IoT), metamaterials, and medical microwave imaging.
Dr. Fernandes was a Guest Editor of the Special Issue on "Antennas and propagation at mm-and sub-mmwaves" of the IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION in April 2013. He started teaching at IST as an Assistant Lecturer in 1965 and became a Full Professor in 1979. From 1966 to 1986, he acted as a part-time consulting engineer in the fields of computers and telecommunications. First, as the Chairperson of the IST Computer Centre from 1983 to 1985 and later as the Chairperson of the National Foundation for Scientific Computing from 1986 to 1989, he was responsible for major improvements in the computer facilities in many Portuguese universities and for the initial deployment of the Portuguese scientific data network (RCCN). From 1989 to 1992, he was the President of the Portuguese National Science and Technology Research Council (JNICT). During that period, JNICT became the most important funding agency in Portugal, and its annual budget increased from U.S. $30 to $150 million. Since 1993, he has been the Chairperson of the Board of Directors of the Instituto de Telecomunicações, Leiria, Portugal, a private not-for-profit research and development (R&D) association of six universities, one polytechnic, and two companies (Portugal Telecom Inovaçõo and Nokia Solutions and Networks), Lisbon. The manpower at the Instituto de Telecomunicações includes more than 350 Ph.D. holding researchers and supports the activities of about 230 Ph.D. and 200 M.Sc. students. From 1986 to 2013, he gained extensive experience in the European Union Framework Programme activities. As a consulting engineer, he was also involved with computer simulation of industrial processes, namely, applied to the handling and storage of grain. In 1974, he has taught telecommunication systems both at graduate and postgraduate levels, and supervised about ten master' and doctorate students. He was twice distinguished by the students as the best professor in electrical engineering. He is currently a Professor Emeritus with IST. He is the author or a coauthor of 30 communications to scientific meetings, 16 journal articles, and five textbooks. His main research subjects have been the calculation and measurement of radiation patterns of microwave aperture antennas, the effects of propagation in the performance of analog and digital radio links, redundancy reduction in the coding of still and moving images, and, more recently, dielectric horn antennas.
Dr. Salema is a Counselor Member of the Portuguese Institute of Engineers (Ordem dos Engenheiros), a fellow and the President of the Lisbon Academy of Sciences, and a fellow (and former President) of the Portuguese Academy of Engineering. He has served as a member of the Selection Committee of the Institute of Engineering and Technology (IET) A. F. Harvey Research Prize from 2015 to 2020. In 1974, he was awarded the Marconi Prize by the Institute of Electrical and Electronic Engineers (IEE).
Rafael F. S. Caldeirinha (Senior Member, IEEE) was born in Leiria, Portugal, in 1974. He received the B.Eng. degree (Hons.) in electronic and communication engineering and the Ph.D. degree from the University of Glamorgan, Pontypridd, U.K., in 1997 and 2001, respectively, and the Habilitation (Agregaçõo) degree from the University of Aveiro, Aveiro, Portugal, in 2020.
He has been a Senior Researcher and the Head of the Antennas & Propagation (A&P-Lr) Research Group, Instituto de Telecomunicações, Leiria, since 2010, and a Coordinator Professor of mobile communications with the Polytechnic of Leiria, Leiria, since 2001. He has authored or coauthored more than 190 papers in conferences and international journals, and four contributions to the ITU-R Study Group, which formed the basis of the ITU-R P. 833-5 (2005) recommendation. His research interests include studies of radiowave propagation through vegetation media, including wildfires, radio channel sounding and modeling, and frequency-selective surfaces, for applications at microwave and millimeter-wave frequencies. Prof