Literature review

To better understand how thermocouples and their data are used in fire ecology, a systematic literature review began with a search on the Web of Science and Google Scholar. The systematic review began by searching the Web of Science with the following Boolean operators: TS=(thermocouple* AND fire) AND TI=(fire). The approximately 334 search results were assessed by title and abstract on the following criteria: (1) Primary literature (study reports, not reviews) reporting data collected from thermocouples (2) in the field (laboratory studies were excluded) and (3) in a wildland (not industrial or structural) context; 46 citations were downloaded and full text acquired, although a few more were excluded after reviewing full text for not meeting the second criterion. Each included title was forward-searched on Google Scholar using the “cited by” feature; results were restricted to those that included the term thermocouple using the “search within citing articles” feature. In the end, 105 unique studies were included in the review (Table S1) and assessed by criteria given in Table tab:review.

Thermocouple response trials

To determine how thermocouple temperature readings vary as their spatial and temporal relationship to a heat source changes, an experimental array was developed to test thermocouple heating and cooling responses to a Bunsen burner in a fume hood with gas access and ventilation (Fig. fig:array). Four K-type thermocouples, each with 0.25 mm diameter (30 gauge) wires enclosed with a 5x20 mm stainless steel sheath (Adafruit Industries, Brooklyn, NY USA), were placed in test tube clamps in each position (1 & 2) on two test tube stands with a Bunsen burner between them (Fig. fig:array). The thermocouples were connected to a Feather M0 Adalogger, an Arduino-based datalogger from Adafruit Industries, which combines an ATSAMD21G18 ARM Cortex M0 microprocessor with a microSD slot for removable data storage. Each thermocouple passed through an Adafruit thermocouple amplifier using the MAX31855 thermocouple-to-digital converter. The Feather sampled and logged each thermocouple at a frequency of approximately 1.5 Hz.

The experimental array used to test thermocouple heating and cooling responses to a Bunsen burner in a fume hood. A single K-type thermocouple (0.25 mm diameter wires with 5x20 mm stainless steel sheath) was held by test tube clamps at each of the Positions 1 & 2 on both test tube stands; all four thermocouples were connected to an Arduino-based datalogger (not shown, see Methods). (A) At the beginning, the burner was ignited and thermocouples at Positions 1 logged flame temperature and thermocouples at Positions 2 logged ambient air temperature. In separate trials, the flame burned for four different durations (5 s, 10 s, 15 s, and 30 s) before it was extinguished; each flame duration had five trials (n=5). (B) Once the flame was extinguished, the test tube clamps were rotated such that their orientations were reversed: Those thermocouples that were logging flame temperature while the flame burned were allowed to cool in air at ambient temperature, while the thermocouples that had been at ambient air temperature away from the flame now recorded air temperature above the extinguished burner. fig:array

Each trial proceeded as follows: dataloggers recorded ambient air temperatures for $<$ 1 min before the Bunsen burner was ignited for a fixed number of seconds (flame duration), at which point the flame was extinguished and each pair of thermocouples rotated 180^o around the test tube stand such that thermocouples in position 2 while the flame was burning switched to record air temperature over the extinguished burner (position 1, Fig. fig:array), and thermocouples heated by the burner (positions 1) were allowed to cool away from the burner (position 2) after the flame was extinguished. Once all thermocouples had returned to ambient air temperature, the trial was repeated for n=5 trials for each of four flame durations: 5 s, 10 s, 15 s, and 30 s. This set-up best simulates wildland applications in which thermocouples are placed in open air above the ground or in herbaceous/grass canopies. The relationship of these data to other applications, such as thermocouples placed in soil or against plant tissue, is considered in the Discussion.

Literature review

While the 105 papers reviewed here are not an exhaustive set of field-based wildland fire science using thermocouples, they do appear to be a representative sample, with studies testing a wide variety of fire, plant, and soil responses (Fig. fig:litrev) worldwide and from various ecosystems. This diversity produces the variety of thermocouple placements reported (Fig. fig:litrev). The majority (71%) of studies that reported thermocouple type used (18% of studies did not report this information) employed K-type thermocouples; six studies used type T, three used type J, and 2 used type E. No temporal trend in thermocouple typesi.e., older studies using now-obscure materialswas apparent: Type T use spanned 19652014, while type J use spanned 19692014. The two studies that used the ultra-thin type E thermocouples were published in 2010 & 2014, and monitored air temperature in smoke plumes.

Although thermocouple data were reported as several variables, the most common were maximum temperature, residence time, and degree $\cdot$ seconds (Fig. fig:litrev). Nearly of studies reported residence time (also referred to as duration of heating), which measures the length of time thermocouples registered temperatures above a threshold value. Among studies that employed such a metric, 60^oC was the most common threshold.

Results of a systematic literature review on the use of thermocouple data in 105 wildland fire science studies. Top left: Four broad categories of responses. Fire includes studies that measured fire behavior, fire/flame temperature, and heat flux. Plant responses were varied. Soil includes nutrient responses, soil properties, and temperature. Other refers to fuel consumption and particle & gas emissions. Top right: Variables used to report thermocouple data. Bottom left: Temperatures used in studies that employed thermal thresholdssuch as in the calculation of residence time and degree $\cdot$ seconds. Bottom right: Summary of where thermocouples were placed in the fire environment. “Above ground” refers to placements not specifically associated with plant parts or vegetation layers. fig:litrev

Thermocouple response trials

As anticipated, thermocouples exposed to the flame of a Bunsen burner (A, Position 1, Fig. fig:array) continued to register high temperatures after the flame had been extinguished and those thermocouples were immediately moved into air at ambient temperature (B, Position 2, Fig. fig:array; Fig. fig:burner). Likewise, thermocouples that had been pointing away from the flame and registered ambient air temperature (A, Position 2, Fig. fig:array) continued to register the temperature of ambient air when positioned over the extinguished burner (B, Position 1, Fig. fig:array; Fig. fig:burner). In all trials, thermocouples positioned above the flame reached their maximum temperature rapidly (47 s) and immediately began to register cooling once the flame was extinguished and the thermocouples were rotated into ambient air unexposed to the flame. However, it took 82100 s for these thermocouples to return to ambient temperature despite no further heat input. Thermocouples exposed to longer flame durations maintained maximum temperatures until the flame was extinguished, and cooled at the same decelerating rate. Meanwhile, in all trials, thermocouples opposite the burner registered ambient air temperature and showed no temperature increase once moved over the extinguished burner (Fig. fig:burner).

Temperature-time curves for four flame durations recorded by K-type thermocouples before (A) and after (B) a Bunsen burner flame was extinguished. The solid dark red line at time 00:00 indicates ignition, while the dotted black line indicates the burner was extinguished and the thermocouples rotated 180^o. A and B refer to top and bottom panels, respectively, in (Fig. fig:array). “Initial position” refers to where each thermocouple was at the beginning of the trial. fig:burner

The past and present of thermocouple data interpretation and application

The impact of these findings on published results has yet to be determined. Broadly stated, conclusions drawn from results in which statistical models testing residence time or degree $\cdot$ seconds as response variables led to a rejection of the null hypothesis, but models with maximum temperature as the response variable did not, likely merit reëvaluation. No obvious cases stand out from this review, although some papers do show divergent patterns between maximum temperature and residence time & degree $\cdot$ seconds, and larger effects between groups among the latter (e.g. Vermeire and Roth 2011; Kral et al. 2018). While a lack of obvious problems is good, a cynic might realise: (1) this review was primarily methodological, and results were not specifically assessed; (2) fire ecologists don’t always use fire measurements in their analyses even when experimental fires are fully instrumented (e.g. Russell et al. 2015); and (3) maximum temperature is frequently left out when ecologists do incorporate fire measurements in statistical tests (Strong, Ganguli, and Vermeire 2013; Russell et al. 2013)possibly a type of file drawer problem in which negative results are not reported when positive results are right at hand.

None of this is to say that thermocouple data are without value in fire ecology, and there is a simple solution for this problem: Truncate the temperature-time curve at the point the thermocouple’s signal becomes determined by the physical properties of the material rather than the heat source of interest. Fig. fig:concept presents two examples of this approach, one for fine fuels, and another for coarse fuels. Each are based on the obvious point that when the heat sourceflame front or smouldering combustionno longer increases or maintains the thermocouple’s reading through the input of additional heat, the thermocouple will begin to cool, and the reading will go down. While a lot of maths can apparently explain how and why the thermocouple cools as it does, at this point, those data are irrelevant to the fire ecologist; they are an artefact of the measuring device, not the phenomenon measured. The only difference between the approach to fine fuels and coarse fuels is that for fine fuels, the curve is truncated immediately after the maximum temperature is reached, as all combustible material is consumed, and the thermocouple immediately begins to cool. In the coarse fuel examplewhich could also be observed from deep flame bases in litter-dominated surface fuelsthe maximum temperature reading is sustained by extended heat input from continued combustion. Obviously, the truncation applies to the portion of the recorded time-temperature curve following peak heating and not simply to the first instance of temperature decline, such as might be caused by diurnal cooling if thermocouples have been deployed well ahead of the burn.

The logic of residence time (a.k.a., duration of heating) is so cogent and its metrics so classic that it can be difficult to come to terms with the fact that the tools of fire ecology fail to represent it well, but in fact, thermocouples mislead by inflating the expected effect. The present experiment offers a starting point to understand the artifice of cooling thermocouples such that this understanding can be translated to the wildland fire environment. It is reasonable that once a flame is extinguished, as in the transition from A to B in Fig. fig:array, the surrounding air should no longer be heated; this is confirmed by the introduction of a new thermocouple, the one switched from opposite the burner to above it after the flame was extinguished (Position 2 to 1, Fig. fig:array), which did not register any increase in temperature. This situation is identical to that of thermocouples in a wildland situation fixed at height intervals above the ground, or in the canopy of fine herbaceous fuels; in either case, once the flame front passes, the thermocouple is surrounded only by air and further heating is limited to radiation from whateverif anysmouldering combustion lingers in the area. Just like the thermocouples in the burner experiment in the fume hood, the temperature gradient between post-heated thermocouples and their surrounding medium in the field is at its sharpest, the condition that causes the greatest degree of error in thermocouple readings (Blevins and Pitts 1999). This is apparent in the initially sharp cooling curve of thermocouples just after maximum heating; as the difference between the thermocouple’s temperature and that of the surrounding air decreases, so does the rate of cooling (Fig. fig:burner). The same pattern is apparent in the rapid increases, sharp spikes, and slow declines characteristic of temperature-time profiles from thermocouples placed in herbaceous fuelbeds (Engle et al. 1989; Jacoby, Ansley, and Trevino 1992; Fidelis et al. 2010; Kral et al. 2015).

Conceptual models of temperature-time profiles generalized from Fig. fig:burner for fine and coarse fuels. As in Fig. fig:burner, the solid, dark red line indicates ignition and the dotted line indicates the point at which heat production ceases (either the flame front moves on, or combustion ceases to produce more energy than is absorbed by heat sinks in the fire environment). In both graphs, A represents the portion of the temperature-time curve that is biologically relevantheating as a result of exothermic reactions fueled by plant biomass in the wildland fire environmentwhile B represents a temperature-time relationship that is merely an artefact of thermocouple diffusivity, and not driven by actual energy in the wildland fire environment. Thus, fire ecologists should exclude the B portion of temperature-time profilesdepicted as the broken linefrom calculations of residence time and degree $\cdot$ seconds. In these models, cooling is plotted as a fourth-order polynomial fit to the B portions of Fig. fig:burner and the heating curve taken from head fire data in Rothermel (1972). fig:concept

The logic of residence time is even more compelling when thermocouples are placed in or against solid material like woody stems or soil. In soil heating studies, spikiness of temperature-time profiles declinesi.e., the curves smooth outas depth increases (DeBano, Rice, and Eugene 1979; Bradstock and Auld 1995; Busse, Shestak, and Hubbert 2013; Carrington 2010; Ramírez Trejo et al. 2010). Because the soil isn’t combusting such that thermocouples are left surrounded only by air, surely the temperatures recorded indicate retained heat, or smouldering combustion (e.g., Odion and Davis 2000)? The issue is that this is true, which exacerbates the problem: unlike thermocouples left surrounded by air, for which all of the post-heating curve can be attributed to the slow heat diffusivity of the thermocouple, the post-heating curve from thermocouples placed in soil is a function of the heat diffusivities of both the thermocouple and the soil, and maths are required to parse their relative contributions to determine how much heat is actually being retained by the soil (which in turn requires information on the heat diffusivity of both the thermocouple and the soil or vegetation media). Soil and wood are both fine insulators: are extended above-ambient thermocouple readings a function of retained heat by these insulators, or a function of these insulators slowing the diffusion of heat away from the thermocouple? Furthermore, soil moisture plays an essential role in heat transfer rates (DeBano 2000; DeBano, Rice, and Eugene 1979; Frandsen and Ryan 1986; Hartford and Frandsen 1992). Thus, unless soil moisture content is controlledor at least measured and accounted forvery little about temperature changes over time in soil can be directly attributed to properties of the fire. Raw temperature-time profiles provide no information to resolve these issues. The conservative approach for fire ecologists is to truncate the curve at the point the thermocouple’s reading begins to decline (e.g., Fig. fig:concept).

The future of thermocouple data

Fire ecologists today collect and interpret thermocouple data from wildland fires no differently than their predecessors did in the 1960s. The only substantial changes have been on the datalogger side: Trollope (1978) described using multiple “electronic temperature recorders”two for temperatures 0800^oC, another for 130670^oCgraduated in 20^oC increments, while Engle et al. (1989) used a magnetic tape recorder that only logged at 0.5 Hz. But while dataloggers are more portable and more powerful, the cost per thermocouple channel has not dropped to a point where many researchers can afford to put more than a handful in a plot and average across them. (Cost quotations weren’t a part of this review, but among commercial systems, ballpark datalogger costs range from approximately US$125275 for a one- or four-channel HOBO thermocouple data logger, respectively (Onset Computer Corporation, Bourne, MA), to at least US$1500 for a high-end, eight-channel logger (CR1000X, Campbell Scientific, Logan, UT). These do not include the cost of the thermocouples or cases to protect the electronics from heat exposure, which can reach US$500 per unit (Jacoby, Ansley, and Trevino 1992; Butler et al. 2010).)

Costs limit replication, which limits spatially-explicit data on fire behaviour. For example, when a number of thermocouples are deployed at fixed distances and the arrival times of flaming fronts recorded with synchronised timestamps, time-temperature data can be used to calculate rate of spread (Davies et al. 2009; Zopfi 2020). But such analysis requires (1) more than one thermocouple probe at a sample point (ideally three, to calculate rate of spread without needing to have an independent record of direction of spread, see (Simard, Deacon, and Adams 1982)) and (2) synchronised timestamps, which is costly in either the labour in programming many single-channel dataloggers, or the expense of multi-channel systems.

Hardware notwithstanding, fire ecologists have applied little creativityand much less scrutinyto the interpretation of temperature-time curves in the last 60 years. Given that temperature is a poor measure of either intensity or heat flux and thermocouple data contribute little information once the maximum temperature is obtained, thermocouple data do little to describe fire behaviour. A major limitation is that one cannot parse the contributions of major modes of heat transfer in the fire environmentthermocouples are sensitive to their own heat loss (Walker and Stocks 1968; Shannon and Butler 2003; Mingchun Luo 1997; Roberts, Coney, and Gibbs 2011), thus their orientation relative to oncoming flames influences their temperature-time response (Coates et al. 2018). Thus, researchers focused on describing the mechanisms of heat transfer and fire behaviour in the wildland fire environment seek to go “beyond fire temperatures (Bova and Dickinson 2008)” and determine how thermocouple data can be calibrated to better predict fireline intensity. Going even farther beyond fire temperatures are techniques to measure dynamic energy flux via radiometers and heat flux sensors (Bova and Dickinson 2009; Rudz et al. 2009; Kremens, Dickinson, and Bova 2012; Frankman et al. 2013).

Another issue is that a fundamental principle underpinning residence timethat there is a lethal level of heating that organisms cannot endurehas recently been called into question. Pingree and Kobziar (2019) put it plainly in their review of “the myth of the biological threshold,” in which they found “both positive and negative responses to soil heating across temperature and duration gradients $$and$$ therefore discourage the use of the traditionally accepted metric of 60^oC for the duration of one minute.” Even the decades-old papers cited as evidence for thresholds in lethal heating described considerable variability in temperatures at different durations (Daniell, Chappell, and Couch 1969; Nelson 1952), and mortality across variable heat dosage appears to vary with species (Wright 1970; Kral et al. 2018). Perhaps an alternative to arbitrary above-ambient values is the 1980 definition by Rothermel and Deeming (1980, 2): “By monitoring fuelbed temperature with thermocouples …the time from initial temperature rise to time of definite drop can be used to indicate residence time.” This “time of definite drop” is consistent with the residence time endpoint RT$_\Omega$ in Fig. fig:flam, defined as the end of the truncated temperature-time curve. While Fig. fig:flam uses the initial reading of 60^oC for the beginning of residence time (RT$_\alpha$) purely for argument’s sake, basing an “initial temperature rise” on ambient temperature makes sense.

A conceptual model of biologically-relevant portions of a hypothetical temperature-time curve collected from a thermocouple, with annotations identifying potentially useful parts of the curve. Residence time: Most broadly applicable are the bounds of a biologically-meaningful residence time (RT, triangles), which begins when the curve crosses a threshold (RT$_\alpha$) and extends until the curve is truncated at the end of the biologically-relevant portion of the curve (RT$_\Omega$), defined as when the thermocouple no longer receives sustained heat input, begins to cool, and the recorded temperature is an artefact of the thermocouple’s physical properties (Fig. fig:concept). The value of the threshold is arbitrary and while basing a threshold on a lethal level of heating is questionable (Pingree and Kobziar 2019), 60^oC is used here for argument’s sake. Flammability: The annotations also identify sections of the temperature-time curve that might be useful in quantifying flammability concepts articulated by Pausas, Keeley, and Schwilk (2017 Box 1). Ignitability is the time between exposure to the heat source (e.g., pre-heating from the approaching fire front) and ignition of the fuel (flaming combustion). Sustainability, or burning time, is the duration of combustion. For fine fuels, sustainability can be very brief. Combustibility, defined as how well fuel burns, is given an * because it is the least obvious in this model. It is associated here with maximum temperature and tentatively bound on the lower end as the point of ignition with the idea that, all other conditions equal, additional heating beyond the point of ignition surely relates to how well the fuel burns. The lower bound might also be ambient temperature or the arbitrary RT threshold. fig:flam

Properly-interpreted temperature-time data from thermocouples might contribute to a growing field of study in fire ecology: flammability. Although flammability remains a nebulous concept through the physics-based lens of fire behaviour (Dickinson and Johnson 2001; Fernandes and Cruz 2012), fire ecologists are developing frameworks in which flammability can help understand the ecology and evolution of species in fire-prone ecosystems (Mutch 1970; Pausas and Moreira 2012; Schwilk 2015; Pausas, Keeley, and Schwilk 2017). Another element of prescience in the Rothermel and Deeming (1980, 2) definition of residence time is the reference to “initial temperature rise”, which suggests the early stages of the combustion process during which fuel particles take on heat at an accelerating rate (preheating) until they reach kindling temperature and ignite into open flaming combustion. In the parlance of flammability, the period between preheating and ignition is called ignitability (Pausas, Keeley, and Schwilk 2017) and can be quantified in the temperature-time curve (Fig. fig:flam). Fuel temperatureand the air around fuelsrises slowly as heating begins and then spikes dramatically once the fuel ignites (Rothermel 1972; Finney et al. 2013). Also potentially quantifiable in the temperature-time curve is sustainability, the duration of combustion (Fig. fig:flam). Less obvious, however, is combustibility, defined variously as how well a material burns (Pausas, Keeley, and Schwilk 2017), or how intensely it burns (Anderson 1970). Although neither definition lines up perfectly with temperature, combustibility could be conceived of as the additional temperature increase beyond ignition, or perhaps the total temperature increase above ambient (Fig. fig:flam). Other applications quantify combustibility as mass loss rate (Simpson et al. 2016).

The use of raw thermocouple data for some flammability components identified in Fig. fig:flam might be limited to the lab, given difficulty scaling up to field-level studies (the same might be said of flammability, in general (Fernandes and Cruz 2012)). Thermocouples have been applied widely in laboratory flammability studies (Simpson et al. 2016; Weir and Scasta 2014; Grootemaat et al. 2017; Gao and Schwilk 2018), where fuel and environmental conditions are easily controlled. An example of confounding environmental variables in the field is the variability in pre-heating rates among wind directionsfuels and downwind of heading fires heat more rapidly than those ahead of backing fires (Rothermel 1972). These conditions would have to be controlled or accounted for in the scaling of lab-based approaches to quantifying flammability.

Supplemental Information

Supplemental information is available at the online version of this paper. Table S1 References included in the systematic review.

Compliance with Ethical Standards

This paper complies with all relevant ethical standards.