Assessing wildland fire suppression effectiveness with infrared imaging on experimental fires

Twenty-two experimental burns were conducted between 5 and 12 September 2019 at the Canadian Forest Service’s Rose Experimental Burn Station, near Thessalon, Ontario, Canada. A burn platform of 7.34 × 4.88 m was located at the base of a 30-m scaffold tower where IR imagery cameras and one visual camera were mounted 15 m above the platform and angled (~21°) to capture the full platform within their field of view (see Table 1 for IR camera specifications). The base of the burn platform consisted of 12 ‘fire-proof’ Marinite boards arranged into three rows of four panels (2.34 × 1.21 m). This experimental burn set-up and IR imagery methodology for measuring fire behaviour has been used in previous studies at this location (e.g. Johnston et al. 2017, 2018). Fig. 1 is an annotated photograph of the burn platform and tower.

Fig. 1.

Annotated photograph of the experimental burn platform. The fuel bed is 7.34 by 4.88 m and comprises 12 Marinite boards. Fuel moisture samples were taken from each outside edge panel (i.e. all panels excluding panels 6 and 7) and fuel consumption measurements (i.e. post-fire ash weight) were taken for each panel immediately after the burn. Primary ignition occurred (when the general wind direction was upwind from the platform) on panels 9–12, whereas secondary ignition (when the general wind direction was downwind from the platform) occurred on panels 1–4.

Prior to the burns, a bed of cured fine fuels was evenly distributed on each panel of the burn platform. Two different fine fuel types were used: dried longleaf pine needles (Pinus palustris), and cut and cured Timothy grass (Phleum pretense). Although stored in a location that was mostly sheltered from rainfall, storage was not environmentally controlled and moisture content in both fuel types varied with ambient conditions on the day of the burn.

All fuel was sorted and dispersed to remove coarse woody material to ensure the fuel consisted of only fine fuels. Two different fuel loadings were used: a moderate fuel loading consisting of a wet mass of 1.0 kg/m² of fuel and a heavy fuel loading consisting of a wet mass of 1.5 kg/m² of fuel. The moderate fuel load of 1.0 kg/m² of pine needles has been used in previous experiments at this site (e.g. Johnston et al. 2017, 2018) and sustainably carried spreading fires through the entire fuel bed. The heavy fuel load was chosen to represent 50% more fuel, to potentially generate higher-intensity fires. All of the 12 sections of the fuel bed (Fig. 1) were loaded with the same amount of fuel to ensure homogeneity in the distribution of the fuel across the burn pad. Ten fuel bed moisture samples (~20–30 g per sample; this sample weight is ~1% of the fuel on each panel in the moderate fuel load grouping and it is therefore reasonable to assume that it did not impact the fire intensity calculation while still providing a robust moisture sample) were taken 5 min prior to each burn, one from each of the outer sections of the fuel bed (all sections excluding nos 6 and 7 in Fig. 1). Samples were weighed and dried in the lab at 95°C for 24 h and an average moisture content (gravimetric moisture by dry weight) for each fuel bed was calculated. Fuel bed depth was also measured on the outer sections of the fuel bed prior to the burn and used to calculate an overall average depth and fuel bed bulk density.

The experimental suppression treatment applied to each fire was water delivered from soft backpack pumps with a jet stream nozzle. Two different water application tactics were used: direct and indirect suppression. During direct suppression, water was precisely targeted using a forceful stream directly into the flaming combustion zone of the fire front. For indirect suppression, water was applied with a dispersed stream on the fuel immediately ahead of the advancing fire front and sprayed to attempt to cover a wider coverage area, targeting a suppression width of 30 cm. Although typically ground suppression in Ontario does not involve indirect suppression using backpack pumps, this wetting of the fuels ahead of the flame front is sometimes used by aerial suppression and in sprinkler applications. Indirect suppression provided an opportunity to mimic the wetting of fuels of aerial suppression while also providing paired contrasts with direct combustion zone suppression.

There were two different dose levels for suppression: a low suppression effort amount consisting of ‘10 pumps’ or approximately 1000 mL of water delivered in total, and a high suppression effort amount consisting of ‘16 pumps’ or approximately 1600 mL of water. Water amounts were delivered uniformly across the width of the fuel bed.

We ignited each burn using a standard forestry drip torch across the width of the fuel bed along the upwind side of the platform (typically panels 9–12 in Fig. 1). When wind flow was in the opposite direction, the ignition line spanned panels 1–4 to ensure the fire was spreading with the wind and exhibiting head fire spread; this occurred for six of the burns (B15, B16, B17, B18, B19, B21 in Table 2). Two suppression treatment passes were made during each burn. The first suppression pass was delivered once the flame front had spread a third of the distance down the burn pad (2.34 m). The fire was left to re-establish and to then spread another third down the burn pad before the second suppression pass. The suppression tactic used for the first and second suppression passes alternated between indirect and direct suppression (i.e. if the first suppression pass for a burn was direct suppression, then the first suppression pass for the next burn would be indirect suppression), with each suppression tactic being used once for each burn. The suppression dose level (i.e. low and high water) was held constant for the two suppression passes on a specific burn but varied between the two dose levels from burn to burn. Note that in some cases, suppression eliminated the re-establishment of sustainable flaming spread and the fire had to be re-ignited (the decision to re-ignite was made if the fire was exhibiting no flaming or active spread after 5 min since the suppression treatment and it was clear that glowing combustion was decreasing, and the fire was approaching extinguishment). In these cases, the establishment of fire spread was consistent with the first observation before the second suppression pass occurred. Once each burn was complete, the residual ash was collected immediately and weighed by burn panel to provide a measure of the fuel consumption. Note that the application of water required individuals to walk across the burn pad, which may have compacted fuels; however; we feel that the variability in wind and its subsequent effect on fire behaviour would have masked the effects of any small, compressed pockets of fuel on the burn pad. Table 2 shows details for each experimental burn, including fuel type, fuel load, fuel depth, suppression amount, moisture content of fuels and weather. Weather observations (i.e. temperature, relative humidity, wind speed and maximum wind speed) were recorded by a fixed station located 100 m from the burn site with an anemometer height of 10 m. The weather observations collected are from the closest hour to when the fire was ignited.

Each burn was observed using two IR cameras, one mid-wave infrared (MWIR) and one long-wave infrared (LWIR) and a visual imagery video camera, all mounted 15 m above the platform. The MWIR camera was used to capture spatial and temporally detailed thermal imagery used to estimate fire intensity and rate of spread, which allowed us to estimate the change in fire behaviour as a result of suppression actions. The methods used are consistent with those described in Johnston et al. (2017, 2018). The LWIR camera was used for ambient thermal imaging and to capture the timing and coverage area of water application. The visual imagery was used to reference fire behaviour attributes from the IR camera.

This analysis was designed to examine the impact of fire suppression treatments on both rate of spread and fireline intensity by using time series data for each row of the burn pad and examining the period immediately before and after suppression.

Pre-suppression fire behaviour estimates were obtained using the 10-s interval prior to the time when suppression began, and post-suppression fire behaviour estimates were obtained using the 10-s interval after the application of suppression. For direct suppression, pre-suppression estimates were calculated using the median observations during the 10-s interval prior to the time suppression began, and post-suppression estimates were calculated using observations during the 10-s interval after the application of suppression across the 1.22 m panel. For indirect suppression, the time suppression took place was not the time when the flame front reached the area where water had been applied, as the water was applied slightly ahead of the flame front. We used the LWIR imagery to estimate the time the flame front reached the treated area and the time when the flame front passed beyond the treated area. Therefore, the pre-suppression time interval ended when the flame front reached the start of the treated area and began 10 s prior to that point. The 10-s post-suppression time interval started after the entire flame front had reached the treated area. In some cases, after suppression was applied, the fire moved so slowly that it stayed within the same pixel for the entire 10-s duration. To estimate rate of spread for these missing time periods when the fire exhibited no active spread, we used the next observation of fire spread.

To calculate a ‘fire behaviour difference’ value, the median post-suppression value was subtracted from the median pre-suppression value for both the rate of spread (referred to as ROS_diff) and fireline intensity (referred to as FI_diff). This method of estimating ROS_diff and FI_diff was repeated for rows 2 and 3 in each of the burns (i.e. 44 ROS_diff values and 44 FI_diff values).

To quantify the impact of the four treatments on fire behaviour, two different four-way factorial ANOVA comparisons were made, one using ROS_diff as the metric of comparison and the other using FI_diff for each burn. When treatment groups or interactions between treatments showed strong evidence of a difference between groups (P < 0.05), Tukey’s Honest Significant Difference post-hoc tests were used to determine which specific treatment group means were different and to identify the direction of any such differences.

The influence of the pre-suppression fire behaviour on the difference in fire behaviour was also examined in terms of the extent to which this relationship varied with suppression type and the amount of water applied. After graphical exploration of these data, two different linear regression models were developed: one for indirect suppression and one for direct suppression. In both models, the independent variables were the 10-s average pre-suppression fire behaviour estimate, the amount of water applied (i.e. low or high) and an interaction term between the two, with the response variable being either ROS_diff or FI_diff.

On average, suppression reduced the rate of spread by 0.5 m/min (s.d. 0.7), or ~60%, and fireline intensity by 50 kW/m (s.d. 159), or 50%. For pre- and post-suppression rate of spread and fireline intensity observations, refer to Supplementary Appendix S2. As fireline intensity is a product of estimated rate of spread and FRED, there is a very strong correlation between rate of spread and fireline intensity. We therefore present only the results for the ROS_diff analysis here (for FI_diff results, see Supplementary Appendix S3).

The four-way ANOVA comparison of ROS_diff showed no evidence of differences in fire behaviour across treatment groups except for fuel type, where only weak evidence was observed (F = 7.00, P = 0.1). Tukey’s post hoc test results suggest that for the main effect of fuel type, grass fuels generally exhibited a greater change in fire behaviour than pine fuels (P = 0.16). These results are consistent with Fig. 4, showing that the average ROS_diff for grass fuels is higher than for pine fuels, a finding likely due to the fact that grass fuels generally supported faster spread rates (Table 4), and therefore had more potential for ROS to decrease. Similarly to the ROS_diff findings, the four-way ANOVA comparison showed no evidence of differences in FI_diff across treatment groups except for fuel type (F = 5.2, P = 0.03).

Fig. 4.

Boxplots of the difference in spread rate before and after suppression, ROS_diff (y-axis; units are metres per minute), for each of the four treatment groups examined: (a) suppression; (b) water amount; (c) fuel type; (d) fuel load. The grey dashed line (y = 0 m/min) indicates no change is observed pre and post suppression treatment. In some cases, the post-suppression rate of spread was higher than the pre-suppression rate of spread (as indicated by a negative ROS_diff).

The analysis of the relationship between the pre-suppression rate of spread and the average ROS_diff showed a positive linear relationship between the two variables (Fig. 5), with strong evidence (P < 0.001) of this increasing relationship for both indirect and direct suppression (Table 5). For fires with greater spread rates, the ROS_diff is greater than fires with slower spread rates, which would imply that both treatments are reducing spread rate to the minimum level that is measurable with the IR camera and arrival time methodology we used. However, this relationship does not seem to change between indirect and direct suppression, and there seems to be little influence of the amount of water applied on ROS_diff (as indicated by the absence of evidence of an interaction between water amount and pre-suppression rate of spread in Table 5).

Fig. 5.

Observations and regression lines of the average rate of spread before suppression compared with the average ROS_diff for both (a) direct, and (b) indirect suppression. This figure shows four regressions: one for low and high water for each suppression type. The solid horizontal black line at y = 0 represents the theoretical relationship that we would see if suppression did not have any influence on fire behaviour (i.e. ROS_diff = 0). The dashed y = x line represents the theoretical relationship that we would see if suppression always reduced rate of spread to 0, indicating that any amount of water applied was fully effective at eliminating spread. Models were built using the following number of observations: direct suppression, low water (n = 24); direct suppression, high water (n = 20); indirect suppression, low water (n = 24); indirect suppression, high water (n = 20).

As the area treated varied between the two suppression techniques, the amount of water per unit area (as determined using estimates of coverage area from the LWIR imagery) applied to the fuel differed; the depth of water delivered was equivalent to approximately 1.4 and 2.3 mm for the low and high water direct suppression treatments respectively, and 0.7 and 1.1 mm for the low and high water treatments for indirect suppression. In these experiments, we applied more water than the amount suggested in fire suppression models to extinguish fires with a flame front fire intensity of 1000 kW/m (e.g. Loane and Gould (1986) and McFayden et al. (2023b) suggest ~0.5 mm of water is required to extinguish a flame front with a fire intensity of 1000 kW/m for pine fuels). Although the water depths applied did reduce spread rate to nearly zero, the fires often continued to burn and were not completely extinguished. The discrepancy between our results and those from the literature may stem from differences in how water was applied, as models (e.g. Loane and Gould 1986, McFayden et al. 2023b) usually assume water is uniformly distributed across the entire combustion zone, reducing the energy levels in all portions of the combustion zone and eliminating potential reheating to ignition temperatures. In our experiment, water was applied to the actively flaming portion of the combustion zone at the leading edge of the flame front and, as a result, the glowing combustion of the litter bed remained active. The partially consumed fuel in these areas continued to release energy, reheating the unburned fuels nearby.

Direct suppression almost always reduced spread rate to the minimum level that could be estimated by this methodology (as demonstrated in Fig. 5); however, this was not the case for the indirect suppression treatment. This different result is likely due to the difference in effective water depth (per unit treatment area) delivered in each suppression method. Indirect suppression distributed the experimental treatment water volume over approximately twice the area of our direct suppression treatment and therefore the effective water depth delivered by indirect suppression was approximately half that of the direct suppression treatment (e.g. 0.7 mm water depth compared with 1.4 mm water depth for the low water treatment). The two suppression treatments also physically impacted fuel bed structure differently and this may have led to some observed differences. The forceful stream of the direct suppression tactic resulted in fuels in the combustion zone being moved around and broken up, thereby disrupting the ability of the flame front to continue passing through the fuel bed, whereas indirect suppression was applied with a more dispersed stream and did not disrupt the continuity of the fuel bed.

This analysis was designed to examine the impact of suppression treatments on the duration of fireline holding time by using time series data and examining the period immediately before suppression with the length of time required for a fire to resume active spread.

The duration of suppression-induced fireline holding time was determined by examining the length of time, in seconds, for which the rate of spread was reduced and showed less variability after suppression. Typically, the post-suppression spread rate was reduced close to 0 m/s and these spread rate observations exhibited virtually no variability. The resumption of more vigorous flaming spread was indicated by not only an increase in rate of spread, but also an obvious increase in variability of the spread rate observations in the IR-based time series. In cases where the rate of spread did not decrease after suppression, the duration of hold was determined to be zero.

A four-way factorial ANOVA comparison of suppression hold time in seconds for each of the four treatment factors was made to assess whether there was evidence of an influence of the four treatment factors on duration of suppression-induced hold. When treatment groups or interactions between treatments showed strong evidence of a difference between groups (P < 0.05), Tukey’s Honest Significant Difference post-hoc tests were performed to determine which specific treatment group means were different and the direction of this difference.

On average, suppression held the fireline for 2 min (average 123 s; s.d. 17 s). In the cases where suppression eliminated the re-establishment of sustainable flaming spread (n = 17), the duration of the suppression-induced hold was set equal to infinity. There were also nine cases where the suppression-induced hold was 0 s. The amount of water was the only treatment factor that exhibited some evidence of a main effect (F = 6.43, P = 0.1), with interactions (P < 0.05) between suppression type and water amount, suppression type and fuel type, and suppression type, water amount and fuel type. Tukey’s post hoc test revealed strong evidence that the high water treatment resulted in a greater duration of hold, with an average hold time 64 s longer (P < 0.05) than the low water treatment condition. This interaction between suppression type and water amount was further examined, showing very strong evidence of a difference for indirect suppression and water amount; the high water treatment resulted in a 120-s increase in duration of hold compared with low water for indirect suppression only (P < 0.01). This finding is likely due to the increased fuel moisture content as a result of the high water condition. This increase in fireline holding time as a function of the water treatment amount was not observed for the direct suppression treatments, likely because the water was being applied directly into the combustion zone, and any impact of the effects of the greater water amount used here might be hidden by the reheating possible from the remaining active combustion zone.

The comparison between average fire behaviour before suppression and the resulting duration of suppression-induced hold was then examined to see if spread rate and intensity influenced fireline hold time. In general, as expected, as the average rate of spread (Fig. 6) or fireline intensity before suppression increased, the ability of suppression to hold the fireline decreased. The interaction of suppression type and water amount is evident for indirect suppression, with low and high water treatments showing separation in the hold durations in both rate of spread (Fig. 6) and fireline intensity plots (Supplementary Appendix S3).

Fig. 6.

Relationship between the average fire behaviour before suppression and the duration of suppression-inducted fireline hold: (a) the average rate of spread before suppression and the duration of suppression-induced hold for direct suppression; (b) the average rate of spread before suppression and the duration of suppression-induced hold for indirect suppression. The points at the top represent observations with a duration of hold set to infinity.

This analysis was designed to examine the impact of water application on the energy released from a fire, as measured by a decrease in FRED.

For the analysis of FRED, the entire burn pad, as opposed to isolating the inner 2.44 m, was used to account for fuel consumption across the entire fuel bed. First, the pixels on the burn pad that received suppression treatment were extracted by isolating the pixels that had fire arrival times between the suppression start and end times. Consequently, for each burn, each pixel in the fuel bed was classified as being one of three types; pixels that had (1) no suppression, (2) direct suppression, or (3) indirect suppression. FRED values were summarised for each of these groupings. The mean FRED values for each of these groups for each burn were compared using visualisations of the distributions and one-way ANOVA comparisons of the means to determine if water did affect the total energy released from the fire, and to estimate the strength of evidence of any difference between direct versus indirect suppression. Interactions between water amount, fuel type and fuel load were examined for each of the suppression categories.

Because FRED integrates radiative energy released over the full duration of the burn, which includes both the flaming and extended period of smouldering combustion, the smouldering or glowing combustion contributed to this total energy release. The energy from this extended period of fuel burn-out could mask any signal of suppression being applied to influence flaming combustion. We therefore examined an energy release measure that represents energy released during only the flaming portion of the fire, which we refer to as ‘flaming FRED’ (F-FRED). This metric integrates radiative energy measured from each pixel starting when the FRPD at that pixel exceeds a value of 0, indicating the cell is now likely flaming, for a pre-defined period of time. Flaming combustion times of 45 s (slightly longer than the time observed by Kidnie and Wotton (2015) for grass fuels, and 60 s (consistent with Taylor et al. 2004) for pine fuels were used.

In the no-suppression areas, the low fuel load pine fuel group (1.0 kg/m²) had an average FRED of ~1700 kJ/m² (s.d. 295), the high fuel load pine fuel group (1.5 kg/m²) had an average FRED of ~3000 kJ/m² (s.d. 123), and the grass fuel group (1.0 kg/m²) had an average FRED of ~1700 kJ/m² (s.d. 113). Although a reduction in FRED was anticipated in cells that received suppression treatment, statistical evidence supporting this was not strong (F = 2.29, P = 0.1), with FRED reductions of 246 kJ/m² for direct suppression and 117 kJ/m² for indirect suppression. The primary factor contributing to differences in FRED was fuel load (F = 86.35, P < 0.001). On average, the high fuel load released nearly 80% (1265 kJ/m²) more energy than the low fuel load. Assuming both fuel loadings burned with identical consumption efficiency, we would expect to observe an ~50% greater energy release from the higher fuel loading.

The absence of clear differences in energy release between treated and untreated cells is likely due to FRED integrating energy release throughout the burn, including the smouldering combustion phase. An analysis of F-FRED (which focuses specifically on the flaming combustion phase) indicated that F-FRED represents ~50% of the total radiative energy of FRED (Fig. 7a). This proportion remained consistent across various cells on the burn pad (Fig. 7a). Examples of a typical FRPD time series observed from untreated (Fig. 7b) and treated cells (Fig. 7c, d) show a clear decrease in FRPD with suppression.

Fig. 7.

(a) Cumulative FRED i(kW/m²) through time for a sub-sample (n = 10) of pixels in a burn that did not receive suppression. The red shaded area highlights the first 60 s after the pixel was ignited (i.e.. FRP > 0). (b) Time series of FRPD (kW/m²) for a single pixel that did not receive suppression. (c) Time series for FRPD (kW/m²) for a single pixel that received direct suppression. (d) Time series for FRPD (kW/m²) for a single pixel that received indirect suppression. All figures are from B1.

Comparisons between untreated areas of the burn pad and cells that received suppression demonstrated strong evidence that the cells that received suppression had less radiative energy release, as demonstrated by reductions in F-FRED, than the untreated areas (F = 10.91, P < 0.001). Direct suppression resulted in an average reduction of 258 kJ/m², and indirect suppression led to a reduction of 189 kJ/m² compared with energy observed from the untreated portions of the fuel bed. However, there is no evidence that suggests there was a difference in flaming energy between the suppression treatment types (P = 0.45), or the amount of water delivered to the treatment areas (F = 2.16, P = 0.15). The high fuel load grouping had a 34% higher F-FRED, a difference of 233 kJ/m² compared with low fuel load experiments, below the expected 50% increase due to increased fuel loading.

Water inhibits the heating of forest fuels to combustion temperature by absorbing energy from the combustion reaction to heat and vaporise it. That energy, ~2600 kJ/kg at 100°C, accounts for the latent heat of vaporisation of water and the energy required to heat the water in the suppression zone from ambient (assumed to be ~20°C) up to 100°C. If we assume the entire mass of water added to the fuel during suppression acts as an energy sink during combustion, the reduction in net energy release expected compared with untreated portions of the fuel bed can be estimated. Once scaled down by the radiative fraction of 0.15 (Johnston et al. 2017), the impact of 1 L of water in the combustion zone on the observed radiative energy is ~390 kJ/m². When accounting for the water coverage area of the suppression treatments, this results in a theoretical energy reduction of 550 kJ/m² for the low water and 900 kJ/m² for the high water direct suppression treatments (Fig. 8a), and 270 kJ/m² for the low water and 430 kJ/m² for the high water indirect suppression treatments (Fig. 8b). These observed values are lower than the theoretically expected energy loss (Fig. 8), although the reduction in F-FRED alone suggests the IR imagery captures the physical effects of water application in a flaming combustion zone.

Fig. 8.

Boxplots of the observed values for the reduction in F-FRED for (a) direct suppression, and (b) indirect suppression compared with the untreated cells. The red dashed lines represent the theoretical energy reduction given the amount of water added in the treatment area. For direct suppression, the low water condition has a theoretical energy reduction of 550 kJ/m² and the high water has a reduction of 900 kJ/m². For indirect suppression, the low water condition has a theoretical energy reduction of 270 kJ/m² and the high water has a reduction of 430 kJ/m².

The comparison of reductions in radiative energy release measured by F-FRED revealed that suppression areas had 30% less energy measured compared with unsuppressed areas, whereas the same comparison with FRED only showed a ~15% reduction in measured energy. Although the energy measured from F-FRED in the suppression areas is slightly lower than we would expect given the heat of vaporisation of water (Fig. 8), this outcome is not surprising because there is a possibility of water loss in both suppression treatments. The direct suppression stream likely penetrates through the fuel complex to the surface of the Marinite boards, which would reduce the ability of the water to inhibit combustion energy release. Similarly, for the indirect suppression treatment, it is likely that some of the water sprayed onto the fuel bed evaporated as the flame front advanced towards the suppression area.