Fine dead fuel moisture shows complex lagged responses to environmental conditions in a saw palmetto (Serenoa repens) flatwoods

Fine dead fuel moisture shows complex lagged responses to environmental conditions in a saw palmetto (Serenoa repens) flatwoods

Agricultural and Forest Meteorology 266–267 (2019) 20–28 Contents lists available at ScienceDirect Agricultural and Forest Meteorology journal homep...

NAN Sizes 0 Downloads 0 Views

Agricultural and Forest Meteorology 266–267 (2019) 20–28

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet

Fine dead fuel moisture shows complex lagged responses to environmental conditions in a saw palmetto (Serenoa repens) flatwoods

T



J. Kevin Hiersa, , Christina L. Stauhammerb, Joseph J. O’Brienc, Henry L. Gholzd,1, Timothy A. Martine, John Homf, Gregory Starrb a

Tall Timbers Research Station, 13093 Henry Beadel Dr., Tallahassee, FL, 32312, USA Department of Biological Sciences, University of Alabama, Tuscaloosa, AL, 35487, USA c Center for Forest Disturbance Science, USDA Forest Service, Athens, GA, 30602, USA d Division of Environmental Biology, National Science Foundation, 2415 Eisenhower Avenue, Alexandria, VA, 22314, USA e School of Forest Resources and Conservation, University of Florida, PO Box 110410, Gainesville, FL, 32611, USA f Climate, Fire, and Carbon Cycle Sciences, Northern Research Station, USDA Forest Service, Newtown Square, PA, 19073, USA b

A R T I C LE I N FO

A B S T R A C T

Keywords: Fire weather Wildfire Prescribed fire Fuels Environmental conditions Fuel moisture

Fine dead fuel moisture has a major influence on wildland fire behavior yet the dynamics driving water exchange of fuel particles in forested environments remain poorly understood. Most fire behavior models rely on simple, stand-level fuel moisture estimates, ignoring potentially important variation occurring within fuelbeds that could influence fire behavior. This is especially true in surface fire regimes where variation in fine-scale fuel properties drive fire behavior and subsequent fire effects. Saw palmetto [Serenoa repens (Bartr.) Small] dominated fuelbeds in the pine forests of the southeastern United States have high within stand variation in one of the most fire prone habitats in the world. Pine needles and palmetto fronds dominate the biomass of fine dead fuel types that produce extreme fire behavior. To assess predictors of fine dead fuel moisture, we analyzed fuel moisture dynamics of these two fine dead fuel types over a two-year period in conjunction with under- and overstory forest meteorological data. Using multiple models and time lag analysis of within-stand moisture dynamics, the results indicate that saw palmetto and pine dramatically differ in drying regimes, primarily resulting from different responses to cumulative rainfall, net radiation, and antecedent atmospheric moisture content. Despite being responsive to changes in relative humidity, saw palmetto was significantly dryer than pine under nearly all meteorological conditions, and it was capable of maintaining extremely low fuel moisture despite high relative humidity or rainfall. Our results point to the need to capture additional drivers of microclimatic variation to aid fire managers in accurately predicting within-stand fuel moisture and subsequent fire behavior. Improving the scientific community’s understanding of variation in complex fuel beds is critical for effectively managing risk in fire prone ecosystems.

1. Introduction Fine dead fuel moisture is a critical factor governing fire behavior and is an input to nearly all wildland fire models (Matthews, 2014). Understanding the environmental drivers of fuel moisture exchange is key in modeling fuel combustion and fire behavior (Burgan and Rothermel, 1984). It is also important for prescribed fire planning (Wade et al., 1989) and development of fire management tactics. Wildland fire combustion is sensitive to fine fuel moisture due to the high specific heat of water and energy needed to vaporize it. Fine dead fuel moisture has been shown to be highly sensitive to relative humidity

(RH), rainfall, shade and diurnal changes in temperature (Viney, 1991; Catchpole et al., 2001), yet modeling how in situ fuels respond to these variables has remained elusive. Understanding the limitations of current fuel moisture inputs to common fire behavior models remains a critical knowledge gap. Currently fuel moisture is often calculated on a stand basis, ignoring much of the critical spatial variability that drives fire behavior (Banwell et al., 2013; Kreye et al., 2013). Within fuel beds, moisture content is also highly dependent on bed depth, bulk density and the plant species composition contributing to the fuel load (Nelson and Hiers, 2008). Yet, these characteristics are highly variable within stands (Ottmar et al.,



Corresponding author. E-mail address: [email protected] (J.K. Hiers). 1 Deceased. https://doi.org/10.1016/j.agrformet.2018.11.038 Received 14 February 2018; Received in revised form 24 November 2018; Accepted 30 November 2018 Available online 11 December 2018 0168-1923/ © 2018 Elsevier B.V. All rights reserved.

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

2016). The equilibrium moisture time-lag theory (Fosberg and Deeming, 1971; Nelson, 1984) has been the dominant technique for predicting fine dead fuel moisture using air temperature and RH measurements. Under constant conditions of temperature and humidity, fuels dry along a negative exponential curve, with the first standard deviation used to denote a fuel’s response time associated with time-lag categories (Nelson, 1984; Nelson, 2000; Nelson and Hiers, 2008). Typically, these categories are expressed as 1, 10, and 100-hr fuels (Fosberg and Deeming, 1971). In an attempt to create a consistent system for assessing wildfire potential, The National Fire Danger Rating System (NFDRS) fuel moisture estimates were based on drying rates of 10-hr fuels (Burgan, 1988). The 10-hr time-lag has become the standard by which fuel moisture is tracked through remote automated weather station (RAWS) networks and used to plan prescribed burns. The method most commonly used to assess fuel moisture in the United States is the Nelson (2000) 10-hr fuel stick model, and the established equations are used to adjust these standard outputs to estimate worstcase fine dead fuel moisture used in fire behavior modeling (Rothermel et al., 1986). This reliance on 10-hr fuels to derive fine dead fuel moisture or simple empirical equations related to environmental variables (Fosberg and Deeming, 1971) lends to high levels of uncertainty in fuel moisture estimates when conditions are dynamic (Weise et al., 2005). This uncertainty is particular problematic with rapidly changing RH (Viney, 1991; Matthews, 2014). While fine dead fuels are often characterized as 1-hr time-lag fuels to represent their rapid response to changing environmental conditions, in reality they represent a diverse suite of vegetation that range from woody debris < 0.6 mm in diameter to pine needles, broadleaf litter and grass litter. Each of these litter types may further be affected by other factors such as litter age (Van Wagner, 1969, 1979, 1982), fuel structure (Nelson and Hiers, 2008), and position within the forest floor (Banwell et al., 2013). In studies of equilibrium moisture conditions, Nelson and Hiers (2008) showed that longleaf pine (Pinus palustris Mill.) litter under constant conditions ranged from 10 to 36-hr timelag fuel, rather than the assumed 1-hr response of fine dead fuel particles. In another study of humid climates that experienced frequent rain, Hough and Albini (1978) suggested that pine dominated fuel beds approximated 100-hr fuels, despite being characterized as fine dead fuel particles. These observations point to variation in fire dead fuels drying modalities that are not captured by stand level averages. Fuel moisture is also affected by rainfall and dew formation which vary spatially and temporally within a stand (Banwell et al., 2013) but these aspects are not often included in fire behavior models (Viney and Hatton, 1990), in which fine dead fuel moistures are specified as a function of temperature and humidity (Andrews, P. 1986). While fuel stick moisture sensors on NFDRS RAWS stations attempt to incorporate precipitation events to more accurately predict fuel moisture (Nelson, 2000), current NFDRS and fire behavior prediction protocols in the United States do not consider variability in fuel moisture within stands. Additionally, the standards for locating RAWS stations place sensors in the open, creating differences between measured conditions and those found within the stand (National Wildfire Coordinating Group (NWCG, 2014). These differences may then lead to improper estimates of fine fuel moisture content and ultimately fire behavior. In contrast to the focus on time-lag response concentrating on diurnal fluctuations of fuel moisture, relatively less attention has been given to how litter species composition and fuel arrangement affects moisture exchange (Matthews et al., 2012; Nelson and Hiers, 2008). Fine dead fuel moistures in humid environments of the southeastern United States continue to produce fire behavior and intensity that exceeds the standard modeled predictions (Wade et al., 1989). Vapor exchange between fuels and the environment using equilibrium moisture theory may be less accurate in humid or dry environments as predicted equilibrium moisture content is asymptotic (Matthews, 2014; Kreye et al., 2018, in press). Saw palmetto–gallberry [Ilex glabra (L.)

Fig. 1. Mean (+/- St. Error) of mid-day pine litter and saw palmetto fine dead moisture content. For comparison, biweekly mid-day average sensor-derived understory fuel moisture (black line) are shown. Saw palmetto moisture content averaged approximately half that of pine needles.

Gray] flatwoods of the Southeastern United States represent one of the most widespread and dynamic wildland fire fuel types in North America (Wade et al., 1989). These fuels (Fig. 1) are represented as Fuel Model 7 in the fire behavior prediction system (Anderson, 1982) and SH4 by Scott and Burgan (2005). They also produce among the most extreme fire behavior potential (e.g., Georgia-Florida Bay Complex 2007, Impassable Bay Fire 2004, Blackjack Bay Complex 2002) in the eastern United States (Hough and Albini, 1978; Wade et al., 1989). Flatwoods are often associated with monospecific or mixed canopies of slash pine (Pinus elliottii Engelm.) or longleaf pine with variable density throughout the southeastern US Coastal Plain (Myers and Ewel, 1990). These forests also represent an estimated 1 million ha of prescribed fire activity annually (Wade et al., 2000; Melvin, 2015). Observations of fire behavior in this ecosystem highlight the within-stand variability at fine scales and document its connection to variation in fuelbed properties (Loudermilk et al., 2012; O’Brien et al., 2016a; Hiers et al., 2009). Moreover, recent studies have documented significant differences in gravimetric moisture content among fine dead fuel types in these ecosystems (Wright, 2013; Ottmar et al., 2016; Prichard et al., 2017). This study assesses the complex interactions between dominant fuel types and environmental drivers of within-stand mid-day fine dead fuel moisture for a pine-saw palmetto flatwoods. The specific objectives of this study are to: 1) compare the drying dynamics of the two main surface fuel components in these flatwood ecosystems: dead saw palmetto litter and pine needle cast, 2) evaluate the impact of antecedent meteorological drivers of fine dead fuel moisture in these two dominate fuel types, 3) compare traditional fine dead fuel moisture estimates (Rothermel et al., 1986) to observed mid-day saw palmetto and pine fuel moisture, and 4) evaluate how weather station location affects estimates of fuel moisture. In doing so it was our overarching goal to understand environmental parameters controlling moisture exchange dynamics of fine dead fuels, which in turn will aid in modeling efforts to reduce fire behavior uncertainties. 2. Materials and methods 2.1. Study site The study was conducted at the University of Florida’s Austin Cary 21

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

2.3. Fuel moisture collections

Forest (ACF) from July 2004 to December 2005. The site is a 41 ha pine flatwoods ecosystem located 15 km northeast of Gainesville, Alachua County, Florida, USA (29°44′17″N, 82°13´8“; elevation 50 m). The soils are poorly drained ultic alaquods (sandy, siliceous, thermic) with a discontinuous spodic horizon 30–60 cm deep and argillic horizon approximately 1.25 m deep (Gaston et al., 1990). Prior to state purchase in 1936, the site had been selectively harvested for timber and used for low-intensity cattle grazing. In 1936 the cattle were removed from the site and the area was allowed to naturally regenerate from remnant seed trees. The resulting stand is comprised of an overstory of longleaf pine and slash pine (72% and 28% of tree basal area, respectively). The understory is dominated by saw palmetto, with lesser species of gallberry, wax myrtle (Myrica cerifera, L.), and wiregrass (Aristrida stricta Michx). Site conditions and biomass data are presented in more detail by Powell et al. (2008), but are summarized here. At the time of this study, tree canopy height averaged 22 m and the understory was ∼1.5 m with no mid-story present; saw palmetto and gallberry dominated ∼90% of the cover. Tree ages ranged from 25 to 85 years old, with an average stem density of 363 trees ha−1. Pine litter loading averaged ∼290 g/m-2, at a depth of 0.25 cm, and saw palmetto loading averaged ∼135 g/m-2, with 25% of that total in dead biomass perched under living fronds (Powell et al., 2008). Pine litter dominated surface litter material and dead saw palmetto fronds were perched just above the pine litter layer under living saw palmetto shrubs (Ottmar and Vihnanek, 2000). The management objective for this site was to restore an uneven-aged, mixed slash and longleaf pine stand by encouraging natural regeneration through prescribed burns every 3 to 5 years. The site was burned prior to the study in January 2003 using backing fire and strip head fires set approximately 30 m apart. This prescribed fire ensured that fuels were not significantly weathered and were typical of conditions reported in fire maintained saw palmetto flatwoods (Ottmar and Vihnanek, 2000).

From July 2004 to December 2005, excluding days with local thunderstorms, bi-weekly samples of saw palmetto fronds and pine litter were randomly collected at the site (n = 5 for each litter type) between the hours of 10:00-15:00. This is the time of day when most prescribed fires are conducted and is considered peak burning conditions for wildfires. Longleaf and slash pine litter was bulked and considered proportional to species composition reported by Powell et al. (2008). Sampling included only dead frond and needle material, resulting in a total of 250 samples of each species over the measurement period. The samples were placed in plastic bags and returned to the field lab at ACF within 20 min where they were initially weighted. Samples were then placed in a drying oven at 70 °C for a minimum of 72 h and then reweighed. Pre- and post-drying weights were then used to calculate gravimetric moisture content of the fuels (Fosberg, 1970; Nelson and Hiers, 2008). Time-stamped fuel moisture sample measurements from each species were matched to the nearest half hour of both the understory (within forest) and overstory (above forest) micrometeorological data. Due to a data logger failure in the understory caused by hurricane activity at the site, July and August 2005 understory measurements were not available to be matched during this time period. Thus, 244 and 242 fuel moisture samples were matched with overstory micrometeorological data for statistical analyses of saw palmetto and pine, respectively; whereas 184 and 183 samples matched with understory micrometeorological data for saw palmetto and pine, respectively. 2.4. Cross-correlation analysis of fuel moisture and lagged variables As a first step, we investigated the degree to which fuel moisture measurements were related to logged micrometeorological variables, explicitly assessing the possibility of time-lagged effects. While the usual Spearman or Pearson correlation coefficient measures the synchronous correlation between two data series, the cross-correlation function (CCF) measures the asynchronous correlation between two series as time is lagged between them. Sensor-derived moisture content data from fuel moisture sticks measured in the understory were available continuously over the study period, every half-hour (excluding the outage). The CCF was computed via the SAS procedure PROC ARIMA for each input series with continuous fuel moisture, using only crossproducts where both series were non-missing (version 9.3; SAS Institute, 2011). However, when data series exhibit autocorrelation, current values of those series depend on their previous values. Since CCFs derived with autocorrelated series can exhibit spurious patterns, we used procedures to specifically account for autocorrelation in the input series themselves (Dean and Dunsmuir, 2016). Following Brocklebank and Dickey (2003), all micrometeorological data series were “pre-whitened” to white noise prior to computing CCFs. Autoregressive models were identified for each input series utilizing the smallest canonical (SCAN) correlation method, and the residual, prewhitened series were used in computing each CCF (Brockelbank and Dickey, 2003). Because the sample-derived fuel moisture data were not time series data, in the sense of having repeated measurements over time for each sample, traditional CCF analyses were not appropriate for these measurements. Instead correlations were computed between the sample measurement and concurrent values of the suite of micrometeorological variables, as well as lagged values of those variables. Each micrometeorological variable (over- and understory) was lagged by half-hour increments up to 10 days, and the pattern of correlation over lagged time was examined. Because all variables were non-normally distributed (Kolmogorov-Smirnov p < 0.01), we used the SAS procedure PROC CORR to calculate Spearman rank correlation coefficients for these analyses, utilizing lagged values of each independent variable with the measured fuel moisture by species. When values of the

2.2. Meteorological and soil measurements A 30-m walkup scaffolding tower was installed with an above-canopy meteorological station. This station included: global radiation [LI200, LI-COR Biosciences, Lincoln, NE, USA], barometric pressure [Vaisala PTB110, Campbell Scientific, Logan, UT, USA], photosynthetically active radiation (PAR) [LI-190, LI-COR Biosciences, Lincoln, NE, USA], net radiation (Rn) (Q7, Radiation and Energy Balance Systems, Inc., Seattle, WA), wind speed and direction (No. 3001-5, R. M. Young Company, Traverse City, MI), air temperature, RH and derived vapor pressure deficit (VPD) (HMP 23 UT, Vaisala, Inc., Helsinki, Finland), and precipitation (tipping bucket, Sierra Misco, Inc., Berkeley, CA). Measurements from the instruments were recorded using a CR1000 datalogger (Campbell Scientific, Logan, UT). A second meteorological tower was placed in the understory at 3.3 m in height. This system had equipment identical to the overstory tower. In addition, this system also included soil heat flux plates (HFT3.1, Radiation and Energy Balance Systems, Inc.) buried 10 cm below the soil surface in three locations within 10 m of the understory tower. We also measured volumetric water content within the top 20 cm of the soil surface using a water content reflectrometer probe (CS616, Campbell Scientific), barometric pressure (PTB110, Vaisala, Helsinki, Finland), and fuel moisture and fuel temperature with 3 paired CS505 and CS107 sensors (Campbell Scientific, Logan, UT) were randomly placed within 15 m of the understory meteorological tower at 30 cm above the soils surface, respectively. The understory meteorological tower data was collected on a CR5000 datalogger (Campbell Scientific, Logan UT). All overstory and understory environmental variables were measured every 15 s and averaged and stored every 30 min on the data loggers.

22

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

Table 1 Statistically significant lags (in hours) between understory and overstory micrometeorological variables and fuel moisture as indicated by cross-correlation analysis, for saw palmetto, pine litter, and fuel moisture sticks. In all variables, lag amounts differed by species and understory versus overstory meteorological station values. An “S” indicates synchronous (zero-lagged) relationship; where understory and overstory values were highly correlated (e.g., PAR and Rn at 1-hour and 25-hour lags), the location with the highest correlation was selected during the model selection process. Rn = net radiation, RH = relative humidity, VPD = vapor pressure deficit. Variable

Air Temperature Precipitation Rn PAR RH VPD Soil heat flux

Saw palmetto

Pine litter

Fuel moisture sticks

understory

overstory

understory

overstory

understory

overstory

10.5

11 3 2, 11.5, 25 2.5, 25 1, 23 1 S, 9.5

10.5

12 27.5 3.5, 14, 17 3.5, 16.5 1.5, 18, 23.5 18 S, 10

0.5

0.5 0.5 1 1 0.5 0.5

1, 14.5 1 S S S, 9

5, 16 8 5.5, 18, 23.5 4.5, 18, 23 S, 10.5

1 1 0.5 0.5 S

and pine. Similarly, models were formulated to describe continuously measured fuel moisture sticks, using the all possible micrometeorological variables as potential predictors. To make these models more comparable to those of the sampled fuel moisture, we considered only those dates and times when sample data were collected. This resulted in 111 half-hour samples; however, due to data outages, only 80 sample times had complete data. To test the applicability of more traditional methods of assessing fuel moisture, we further estimated models of field-based moisture content as a function of (1) continuous fuel moisture sticks and (2) Fosberg’s fine dead fuel moisture (FDFM; Rothermel et al., 1986). These common measurements were used as predictors for two simple mixed model via the SAS procedure PROC MIXED, accounting for repeated measurements. Each of the final models by species included understory RH, accumulated precipitation, and both under- and overstory Rn (see Supplementary Information, Table S1 and S4 for model selection). The inclusion of both under- and overstory meteorological variables resulted in much better fitting models, as evidenced by their lower AIC values. The final selected model for saw palmetto had an AIC of -176, whereas inclusion of only understory or only overstory variables resulted in AICs of -141 and -163, respectively (Supplementary Information, Tables S2 and S3). Similarly for pine, the understory-only and overstory-only models had much less support (AIC=-109 and -123, respectively) than the final selected model (AIC=-154) (Supplementary Information, Tables S5 and S6). Models of fuel moisture sticks were much simpler than that of fuel samples due to the lack of significant lagged correlations beyond one hour in the cross-correlation functions (Table 1, and also see Supplementary material Table S7).

independent variable were missing, or the dependent variable was not measured, measurements were excluded from the analysis. We generated CCFs over a range of time lags to determine the timedependent nature of relationships between the continuous fuel moisture and each pre-whitened input variable, as well as the sample-derived fuel moisture and lagged micrometeorological variables. Graphical representations of each series’ correlation function were used to identify lags using a window of ten days. In the case of field-based moisture measurements, these graphical representations were further used to identify cyclical patterns indicating the influence of autocorrelation on CCFs. 2.5. Models of moisture content considering micrometeorological variables Linear mixed models were formulated to describe the relationships between fuel moisture content and environmental variables. The moisture content data were arcsine transformed prior to analysis to aid in meeting assumptions of homoscedasticity and normality of the residuals. Independent variables were standardized to ensure model stability. First, models were formulated for sampled fuel moisture content via the SAS procedure PROC MIXED, separately for saw palmetto and pine. We first considered using all possible micrometeorological variables as potential predictors in the model, utilizing data reduction techniques to avoid multicollinearity. While principal components analysis (PCA) has been successfully used to reduce the number of independent variables in similar situations, preliminary PCA did not reveal meaningful components for these data. Moreover, analyses of lagged correlations revealed complex non-synchronous correlations between fuel moisture and several micrometeorological variables. Thus, independent variables considered for each species were a subset of micrometeorological variables and highly correlated lags thereof, chosen on the basis of their bivariate correlations with moisture content, as well as their lack of correlation with the other independent variables in an effort to avoid multicollinearity. Multicollinearity was also formally tested via simple regression techniques via the SAS procedure PROC REG, examining the variance inflation factor (VIF) with a cut-off of 10 (Neter et al., 1989). Repeated measures structures were included in each model, such that samples collected on the same date were grouped together. A modified backwards-stepwise selection method was used, dropping non-significant variables sequentially, and evaluating the Akaike Information Criteria (AIC) at each step. The selected models were those which had the lowest AIC and least number of parameters (Burnham and Anderson, 2002). A single model with both species was then formulated to investigate common predictors between species. Using the significant predictors from the species-level models, interactions of each predictor with species were tested. Non-significant interactions which could be eliminated based on significance and AIC values were interpreted to indicate driving factors common to both saw palmetto

3. Results Analyses of pine litter and dead saw palmetto frond fuel moisture showed significantly different minimum fine dead fuel moisture content, with mid-day saw palmetto moisture content being 17% lower than pine on average (Fig. 1). Over the 50 dates of sample collection, the moisture content of pine litter was consistently higher than that of saw palmetto, with the exception of 5 sample dates, 4 of which were influenced by fiber saturating rain events within the preceding 24 h. When analyzing lagged correlations of fuel moisture with micrometeorological variables, pine litter and dead saw palmetto fuel moisture showed divergent responses to antecedent weather variables and environmental conditions. Time-lag analysis of patterns revealed complex lags in response to both current and prior day weather variables. Moreover, weather station position (under the canopy or in the open) revealed different patterns in lags between micrometeorological variables and fuel moisture (Table 1, Fig. 2–4). Our analysis showed pine litter moisture content had much higher 23

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

8 h for pine (Figs. 3 and 4). For the overstory met data, PAR correlations with fuel moisture were more complex, with primary peaks at 2.5 and 3.5 h lag for saw palmetto and pine, respectively, and a secondary peak at 16.5 h lag for pine. There was also a second peak for PAR at ∼25hours for saw palmetto, likely the result of autocorrelation of the daily pattern in PAR signal. Rn also showed complex correlation patterns with fuel moisture. In the overstory, patterns of correlations between fuel moisture and Rn were similar to that observed in overstory PAR (Fig. 3). For saw palmetto and pine litter, respectively, there were moderate local peak negative correlations of the understory Rn values with fuel moisture content at ∼1 and 5 h, and positive peak correlations at in the understory at 14.5 and 16 h (Fig. 4). Air temperature, both in the overstory and at ground level showed positive lagged correlations with fuel moisture at an approximate half-day lag, but peak lags occurred ∼1 h sooner in saw palmetto versus pine when considering the overstory (Figs. 3 and 4).

Fig. 2. Analysis of time-lagged correlations for pine litter and saw palmetto litter with cumulative rainfall. Pine litter shows a stronger, lagged relationship with cumulative rainfall in the previous 24 h, while saw palmetto shows an immediate but relatively weaker overall response to cumulative rainfall.

3.1. Analyses of correlations with fuel moisture sticks

correlations with cumulative rainfall (Fig. 2). While correlations between cumulative rainfall and moisture content in saw palmetto peaked within ∼2.5-3 h and again at ∼14.5–16 hours, correlations with pine litter peaked at lags of ∼26-28.5 h. Similarities in the two species’ correlation patterns reflect autocorrelation within the rainfall series itself. While correlations were computed out to 10 d, meaningful correlations of both understory and overstory micrometeorological variables with moisture content occurred at lags of 25 h or less. For RH and VPD, the peak maximum (absolute value) correlation with saw palmetto fuel moisture content was approximately concurrent (immediate-1 h) in both overstory and understory (Figs. 3 and 4). On the other hand, the peak correlation of RH with pine litter moisture content occurred at a lag of 23.5 h for both understory and overstory. For pine litter and VPD, the peak lag was at 23 h in the overstory, and 18 h in the understory (Figs. 3 and 4). Moreover, significant secondary peaks of pine fuel moisture with RH occurred at 1.5 and 5.5 h’ lag (overstory and understory, respectively) (Table 1). While this result may have resulted by autocorrelation within each of the RH and VPD signals themselves, the fact that pattern differs by species argues that there are different drying processes at work. For understory PAR, the peak maximum (absolute value) correlation with fuel moisture content was lagged by ∼1 h for saw palmetto and

The 10-hour fuel moisture probe, used on remote weather stations (NWCG, 2014) were also analyzed for antecedent environmental predictors. Though we computed the cross-correlations out to 240 lags (5 days), significant lags were not observed after 12 h, and the most substantial overstory cross-correlations with fuel stick moisture occurred with very short time lags (Table 1, Fig. 5). Considering understory measurements, the most substantial cross-correlation occurred between the fuel stick moisture and air temperature at ½ hour and RH at ½ hour (Fig. 5a). Most other cross-correlations were negative. For the overstory “open” station, the most significant cross correlation was with VPD at ½ hour lag (Fig. 5b). For both understory and overstory measurements, the strongest cross-correlations with the continuous fuel stick moisture for PAR and Rn were at one-hour lags; for air temperature, RH, VPD, precipitation and soil temperature at 0 cm, the strongest lags were at ½ hour. For soil temperature at 5 cm and 10 cm, the strongest correlations were synchronous. 3.2. Models of fuel moisture samples Based on the lagged correlation analyses, models describing sampled fuel moisture were formulated with subsets of the highest correlated independent variables and lags by species (Table 1). However, not all predictors could be included due to their inter-correlations. For

Fig. 3. Analysis of time-lagged correlations for pine litter and saw palmetto litter with environmental variables from overstory (open) weather station at lags up to 30 h. RH = relative humidity, Rn = net radiation, VPD = vapor pressure deficit, PAR = photosynthetically active radiation. 24

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

Fig. 4. Analysis of time-lagged relationships for pine litter and saw palmetto litter with within-stand understory weather station at lags up to 30 h. RH = relative humidity, Rn = net radiation, VPD = vapor pressure deficit, PAR = photosynthetically active radiation.

example, understory and overstory air temperatures lagged by ∼11 h and were highly correlated with both sampled fuel moisture and each other. Based on the cross-correlation analyses, the understory value was the stronger predictor of sampled fuel moisture for both species and was therefore used as an input to both pine litter and saw palmetto litter models. Additionally, several significant relationships between RH and fuel moisture by species were indicated in the understory and overstory. Although the lagged values were not significantly correlated within variable, RH values were significantly correlated with those of VPD. Since RH had stronger correlations with fuel moisture, models were therefore formulated with multiple lagged values of RH in both species. Each of the final models by species included understory RH, accumulated precipitation, and both under- and overstory Rn. The inclusion of both under- and overstory variables resulted in much better fitting models, as evidenced by their lower AIC values. The final selected model for saw palmetto had an AIC of -176, whereas inclusion of only understory or only overstory variables resulted in AICs of -141 and -163, respectively. Similarly, for pine, the understory only and overstory only models had much less support (AIC=-109 and -123, respectively) than the final selected model (AIC=-154). While many of the same environmental variables were implicated in both species’ models, the lags at which predictor variables were significant varied by species (Table 2). Whereas synchronous understory RH was a significant predictor of saw palmetto moisture content, pine litter moisture content showed multiple significant lags to RH understory. Not only was pine litter moisture significantly lagged at 5.5 and 18 h to understory RH, the overstory RH showed a significant lag of approximately 1 day as a predictor of pine litter moisture. For both species, higher RH and cumulative precipitation was associated with higher moisture contents, though these effects were also lagged for pine

(Table 2). The effects of Rn and PAR were somewhat complex. For saw palmetto, there was a strong decrease in moisture content as 14.5-hour lagged understory Rn increased, and a much weaker increase in moisture content as 11.5-hour lagged overstory Rn decreased. Since most measurements were taken between 10:00am and 3:00pm, the former variable corresponds to post-dusk values of Rn, whereas the latter corresponds to nighttime values of Rn. Thus, higher values of evening understory Rn resulted in drier fuels, while higher nighttime Rn in the overstory was correlated with slightly higher fuel moisture according to model results. This may reflect the effects of re-radiation of energy in the subcanopy in the evening, and possibly capture conditions of dew formation overnight. Dew is a significant sources of daily moisture dynamics in this ecosystem (Kreye et al., 2018, in press). For pine, there were decreases in moisture content with increased lagged Rn (5 h and 14 h), but increased with 16.5-hour lagged overstory PAR (Table 2). While this may seem incongruous, this lag corresponds to PAR measured near dusk, due to the timing of the moisture content measurements. Therefore, lagged PAR is likely a proxy for day length; when days are longer (and PAR is higher earlier in the day), there is higher moisture content of pine litter. Combined models for pine litter and saw palmetto yielded few common predictors. All effects from the separate species models and their interactions, except one, were significant (p < 0.05). Only Rn lagged at 14.5 h had a common (negative) effect, which did not interact by species (data not shown). 3.3. Models of fuel moisture sticks Models of fuel moisture sticks were much simpler than that of fuel Fig. 5. Analysis of cross-correlations with standard 10hr fuel sticks and environmental variables from (A) understory and (B) overstory meteorological stations for comparison to continuously sampled fuel moisture. Significant variables from both stations include PAR, net radiation (Rn), air temperature (Tair), and RH. The understory variables included soil heat flux (SHF) and soil temperature at 0-cm, 5-cm, and 50-cm (Ts0, Ts5, Ts10, respectively), while the overstory variables included vapor pressure deficit (VPD) and precipitation.

25

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

Table 2 Results of Type 3 Tests of Fixed effects for models of mid-day fuel moisture content of: saw palmetto, pine litter, and fuel moisture sticks. Significant effects depended on fuel type. Fuel type

Effect

Estimate

Std. Error

DF

t Value

Pr > |t|

Saw palmetto

Intercept RH of understory Rn of overstory (11.5 hr lag) Rn of understory (14.5 hr lag) Cumulative precipitation (3 hr) Intercept RH of understory (5.5 hr lag) RH of understory (18 hr lag) RH of overstory (23.5 hr lag) PAR of overstory (16.5 hr lag) Rn of understory (5 hr lag) Rn of overstory (14 hr lag) Cumulative precipitation (27.5 hr) Intercept RH of understory (0.5 hr lag) Temperature understory (0.5 hr lag)

0.1893 0.0904 0.0347 −0.0616 0.0612 0.3384 0.0462 0.0945 0.0882 0.0278 −0.0387 −0.0584 0.0736 1.5212 0.0128 −0.0074

0.0102 0.0152 0.0159 0.0145 0.0096 0.0105 0.0162 0.0145 0.0116 0.0118 0.0148 0.0160 0.0097 0.0022 0.0022 0.0022

179 179 179 179 179 175 175 175 175 175 175 175 175 77 77 77

18.5 5.95 2.19 −4.24 6.38 32.28 2.85 6.53 7.58 2.35 −2.60 −3.65 7.62 701.68 5.87 −3.4

< .0001 < .0001 0.0301 < .0001 < .0001 < .0001 0.005 < .0001 < .0001 0.0199 0.01 0.0003 < .0001 < .0001 < .0001 0.0011

Pine litter

Moisture stick

results show that under dynamic environmental conditions, fine dead fuel moisture is a fuel-type dependent result of complex lags operating on multiple timescales not captured by traditional methods. As such, stand level estimates cannot simply be captured by an arithmetic mean of these two dominant fuels. Standard tools for estimating fine dead fuel moisture from remote automated weather stations (RAWS) and the Fire Behavior Prediction System provided poor estimates of fuel moisture in pine litter and saw palmetto dominated fuelbeds common to the Southeastern United States. While this problem has been recognized in humid ecosystems (Wade et al., 1989), field data to validate fuel moisture estimates has rarely been collected in situ, instead relying on equilibrium moisture conditions in the lab to generate relationships (Nelson and Hiers, 2008). These results show categorically different minimum daily dead fuel moisture contents between pine litter and saw palmetto. Despite a responsiveness to near instantaneous changes in RH and short-term rainfall accumulation (3-hr lag), saw palmetto litter consistently had lower fuel moistures under nearly all combinations of antecedent environmental conditions. Thus, while saw palmetto appeared to have some characteristics of 1-hr time-lag fuels in a relative sense, absolute moisture content was consistently below the moisture of extinction of 40% (Blackmarr, 1972) and was always lower than that predicted by traditional estimates. Saw palmetto is known for its highly flammable litter and ability to exceed traditional estimates of fire spread under predicted fuel moisture scenarios (Hough and Albini, 1978; Burgan, 1988). One potential explanation for this difference is that saw palmetto is shielded from short-duration rain events because of the positioning of litter below the live fronds. Furthermore, the waxy cuticle and vertical orientation efficiently sheds water even in dead fronds (Abrahamson, 2007). Such differences in basic litter characteristics may also help to explain variability not attributable to environmental conditions in this study. Pine litter is critical to patterns of energy release from frequent surface fires in these pine dominated ecosystems (O’Brien, et al., 2016a; 2016b), and pine litter in this study revealed a complex response in moisture content to multiple lagged environmental predictors. Analysis of lagged correlations revealed three dynamic patterns with this dominant fuel type. Pine litter moisture showed highest correlations with Rn, RH, and VPD at ∼23-26-hr lag, i.e. the values from the day before, and a shorter term 5-hr and 5.5-hr lag response to Rn and RH, respectively. Pine litter also appeared to be very slow in responding to precipitation, with a significant response to rainfall accumulations in the preceding 27.5 hs. While pine litter moisture exchange has been shown to approximate 10 to 30-hour time lag fuels in the laboratory (Nelson and Hiers, 2008), the multiple lagged responses to RH, Rn, and rainfall capture the challenge of extending laboratory predictions to the

samples due to the lack of significant lagged correlations beyond one hour in the cross-correlation functions (Table 1). Moreover, ½-hour and 1-hour PAR and Rn in the overstory and understory were highly intercorrelated, and thus models could be formulated with just one of these variables. Likewise, correlations among over- and understory RH and VPD were high, allowing just one of these variables to be included. In all cases, correlations with fuel sticks were higher when considering understory micrometeorological variables. Models were therefore fit with understory values of 1- hr lagged Rn, as well as ½-hr lagged values of RH, air temperature, and precipitation. The final model included just two significant effects (Table 2). Half-hour lagged RH had a positive association with fuel stick moisture, and ½-hr lagged air temperature had a negative association with fuel stick moisture. 3.4. Predicting sampled fuel moisture from fuel sticks By comparison, models of field-based fuel moisture as a function of fuel stick-derived moisture had much less support. For saw palmetto, the model with micrometeorological variables had an AIC much lower than that of the fuel stick model (-176 versus -113.7). For pine litter, model AICs were also reduced when using micrometeorological variables (-152, versus -103.4 with the fuel moisture stick; results not shown). These results suggest that fuel stick sensors embedded within the understory do more to improve prediction of the moisture of pine needles on the ground (ΔAIC = 48.6) in comparison to saw palmetto (ΔAIC = 62.3); however, there was substantial evidence that using the more complex environmental variables produced better fitting models for observed fuel moisture data. The empirical formula commonly used to calculate fine dead fuel moisture for wildland fire, the FDFM, was a poor predictor of field moisture content for either pine litter or saw palmetto litter in this study. Looking very coarsely at the patterns of correlations and lagged FDFM, there appears to be no variable that gains appreciable correlation as time goes on (i.e., no lags). Using a simple model with FDFM as the single predictor of fuel moisture, AICs indicated a low level of evidence to support these models. FDFM was not a significant predictor (p = 0.8044) for pine litter moisture, and had an AIC of -25.9 (ΔAIC = 126 versus the full model with micrometeorological variables). Yet, for saw palmetto, FDFM was a significant predictor (p < 0.0001), but the AIC of -101 indicated substantially lower support than that of the full model (ΔAIC = 75). 4. Discussion Between dominant fine dead fuel types, the dynamics and drivers of within-stand variation in fuel moisture varied significantly. These 26

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

field. These multiple lags of in situ pine litter moisture exchange may appear analogous to tide calculations where predicted high or low tides are the non-linear outcome of short and long-term wave functions, but stochasticity of rainfall duration and amount challenge simple estimates in humid environments. In this study, a full consideration of environmental variables from the two instrumented in situ weather station revealed more nuanced drivers of fuel moisture than are commonly available to fire managers. The location of the weather station measurements produced different drying models. Since standard RAWS placement can either be located in the open or within representative forests, environmental variables used to describe fuel moisture dynamics must be modified by the context of the station. Nonetheless, the location of micrometeorological weather stations did little to enhance overall prediction of local fuel moisture, with substantial improvements to the AIC when both overstory and understory were included in models. The inclusion of variables from both stations (open and within stand) point to moisture dynamics within stands not captured by standard micrometeorological station positions. Such variables could include canopy mediated patterns of shade and sun flecks, live fuel impacts on boundary layer RH via transpiration, and soil influences on fuel moisture of surface litter. Net radiation was also a strong driver of fuel moisture dynamics for in situ fuels but not for 10-hour fuel sticks affixed to weather stations. For saw palmetto litter, we observed a strong decrease in moisture content as 14.5-hour lagged understory Rn increased, and a weaker increase in moisture content as 11.5-hour lagged overstory Rn decreased. Since measurements were taken midday, the responses correspond to post-dusk and nighttime values of Rn, respectively. Less negative Rn would indicate conditions where dew would be less likely to form allowing the fuels to continue drying. Conversely, a clear night sky with a highly negative Rn may likely also be accompanied by dew that moistens fuels. For pine, there were decreases in moisture content with increased lagged Rn (5 h and 14 h), but pine moisture also increased with 16.5-hour lagged overstory PAR (Table 2). This lag corresponds to PAR measured the evening prior, likely a proxy for seasonally changing day length. When days were longer (and PAR is higher earlier and later the day), there was higher moisture content of pine litter. The relationship between negative sensible heat flux at night and fuel moisture, likely driven by the influence of dew formation common to spring and summer conditions, was captured more effectively in the understory meteorological station. This relationship of fuel moisture with evening and nighttime Rn is a novel observation in this ecosystem that could vary significantly with canopy and midstory characteristics. Our results reveal a window into the acknowledged problem of modeling fire spread in high-humidity, low-latitude forests and how generalizing across scales can yield inaccurate fuel moisture estimates (Burgan, 1988). Anecdotal evidence of extreme fire behavior despite high humidity or even during rain is likely a product of the two modes of drying and wetting found within these dominant fuels. Pine needles tend to reflect the prior day’s sun-driven moisture dynamics, and the physical interception of rain by live palm fronds tends to decouple fine dead palm moisture from all but the most intense rain events. Thus, fire is capable of spreading quickly after recent rainfall or under high humidity due to the synergy among these fuels. Having such large pools of flammable fuel to sustain combustion across varied environmental conditions would allow the heat of combustion to rapidly volatilize fuel moisture in the adjacent fuel types, despite differences in their moisture dynamics. In this way, variable spatial distributions of fine dead fuels at fine scales could enhance fire spread under conditions normally considered sub-optimal for combustion, reinforcing the critical importance of fine-scale fuel variation as a process driving variation in combustion (Hiers et al., 2009; Mitchell et al., 2009; O’Brien et al., 2016b).

southeastern US specifically, the ability to capture variation in mid-day fuel moisture predictions are challenging. This study points to the critical differences in fine dead fuel moisture dynamics of slash pines and saw palmetto in one of the most fire prone fuels in the US (Hough and Albini, 1978). These differences in drying dynamics between commonly combined fine dead fuels illustrates the difficulty in using traditional approaches to develop prescription parameters or predict fire behavior, particularly following rain events. On-site weather measurements are capable of predicting general dynamics for moisture during peak midday burning conditions, but large differences in drying characteristic among dominant fine dead fuel types cautions against the overreliance on fuel sticks or commonly used tools for estimating fuel moisture in these environments. In addition to significant differences in the absolute fuel moisture of these two dominant fuel types, this study recognizes two distinct drying modes within the fine dead fuels responsible for the majority of biomass and fire behavior in this forest (Ottmar and Vihnanek, 2000). Despite a history of research into the fire behavior of this fuel bed (Hough and Albini, 1978) and the recognition that saw palmetto spatial coverage has a large impact on fire behavior, the assumption of fuel moisture homogeneity has never been challenged (Anderson, 1982; Scott and Burgan, 2005). Pine litter and saw palmetto dead fronds are the dominant biomass of fine dead fuels in many southeastern ecosystems (Ottmar et al., 2007). These divergent responses of fine dead fuels to precipitation, net radiation, and atmospheric moisture confound attempts to parameterize fuel moisture effects in fire behavior models. Live saw palmetto fronds shade dead and dying fronds from exposure to sunlight and precipitation creating the potential for large differences in fine dead fuel moisture from predictions derived from traditional methods. Fine scale variation in fuels in this ecosystem has been shown to drive fire behavior (Hiers et al., 2009; Loudermilk et al., 2012; O’Brien et al., 2016a, 2016b), and this study suggests fine scale variation in fuel moisture content is a critical mechanism behind those patterns. Acknowledgements We dedicate this work to the memory of Dr. Henry Gholz who passed too soon. His mentorship of ideas and experimentation will maintain a lasting legacy. Funding for this research was provided by the USFS under agreement 03-CA-11242343-114. Special thanks is given to the University of Florida for access and funding of portions of this research. References Abrahamson, W.G., 2007. Leaf traits and leaf life spans of two xeric-adapted palmettos. Am. J. Bot. 94 (8), 1297–1308. Anderson, H.E., 1982. Aids to Determining Fuel Models for Estimating Fire Behavior. USDA Forest Service General Technical Report INT-122. pp. 22. Andrews, P., 1986. BEHAVE : Fire behavior prediction and fuel modeling system - BURN subsystem, part 1. USDA Forest Service, Intermountain Forest and Range Experiment Station, General Technical Report INT-194. 130 pp. Banwell, E.M., Varner, J.M., Knapp, E.E., 2013. Spatial, seasonal, and diel forest floor moisture dynamics in Jeffrey pine-white fir forests of the Lake Tahoe Basin, USA. For. Ecol. Manag. 305, 11–20. Blackmarr, W.H., 1972. Moisture Content Influences Ignitability of Slash Pine Litter. USDA Forest Service Southeast Forest Experiment Station, Asheville, NC. Research Note SE-173. pp. 1–7. Brocklebank, J.C., Dickey, D.A., 2003. SAS System for Forecasting Time Series, 2nd ed. SAS Institute Inc., Cary, North Carolina. Burgan, R.E., 1988. 1988 Revisions to the 1978 National Fire Danger Rating System. USDA Forest Service, Research Paper SE-273. Burgan, R.E., Rothermel, R.C., 1984. BEHAVE: Fire Behavior Prediction and Fuel Modeling System–FUEL Subsystem. USDA Forest Service General Technical Report INT-167. pp. 1–124. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multi-Model Inference: A Practical Information-Theoretic Approach, Second ed. Springer-Verlag, New York, NY, USA. Catchpole, E.A., Catchpole, W.R., Viney, N.R., McCaw, W.L., Marsden-Smedley, J.B., 2001. Estimating fuel response time and predicting fuel moisture content from field

5. Conclusions In humid environments generally and in fuels of the subtropical 27

Agricultural and Forest Meteorology 266–267 (2019) 20–28

J.K. Hiers et al.

Andrew, Strother, Dexter, Rowell, Eric, Bright, Benjamin C., 2016b. Canopy derived fuels drive patterns of in-fire energy release and understory plant mortality in a longleaf pine (Pinus palustris) sandhill in Northwest FL, USA. Can. J. Remote. Sens. 42 (5), 1–12. Ottmar, R.D., Hudak, A.T., Prichard, S.J., Wright, C.S., Restaino, J.C., Kennedy, M.C., Vihnanek, R.E., 2016. Pre-fire and post-fire surface fuel and cover measurements collected in the south-eastern United States for model evaluation and development–RxCADRE 2008, 2011 and 2012. Int. J. Wildland Fire 25 (1), 10–24. Ottmar, R.D., Vihnanek, R.E., 2000. Stereo Photo Series for Quantifying Natural Fuels Volume VI: Longleaf Pine, Pocosin, and Marshgrass Types in the Southeast United States. National Wildfire Coordinating Group, National Interagency Fire Center PMS 835. Powell, T.L., Gholz, H.L., Clark, K.L., Starr, G., Cropper Jr, W.P., Martin, T.A., 2008. Carbon exchange of a mature, naturally regenerated pine forest in north Florida. Global Change Biol. 14 (11), 2523–2538. Prichard, S.J., Kennedy, M.C., Wright, C.S., Cronan, J.B., Ottmar, R.D., 2017. Predicting forest floor and woody fuel consumption from prescribed burns in southern and western pine ecosystems of the United States. For. Ecol. Manag. 405, 328–338. Rothermel, R.C., Wilson, R.A., Morris, G.A., Sackett, S.S., 1986. Modeling Moisture Content of Fine Dead Wildland Fuels: Input to the BEHAVE Fire Prediction System. USDA Forest Service Research Paper, INT-359. Intermountain Research Station, Ogden, UT. Scott, J.H., Burgan, R.E., 2005. Standard Fire Behavior Fuel Models: a Comprehensive Set for Use With Rothermel’s Surface Fire Spread Model. Gen. Tech. Rep. RMRS-GTR153. US Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fort Collins, CO 72 pp. Viney, N.R., 1991. A review of fine fuel moisture modeling. Int. J. Wildland Fire 1, 215–234. Van Wagner, C.E., 1969. Drying rates of some fine forest fuels. Fire Control Notes 12, 5–12. Van Wagner, C.E., 1979. A laboratory study of weather effects on the drying rate of jack pine litter. J. For. Res. 9, 267–275. https://doi.org/10.1139/x79-044. Van Wagner, C.E., 1982. Initial moisture content and the exponential drying process. Can. J. For. Res. 12, 90–92. https://doi.org/10.1139/ x82-013. Viney, N.R., Hatton, T.J., 1990. Modelling the effect of condensation on the moisture content of forest litter. Agric. For. Meteorol. 51 (1), 51–62. Wade, D.D., Lunsford, J.D., Dixon, M.J., Mobley, H.E., 1989. A Guide for Prescribed Fire in Southern Forests. Technical publication R8-TP-US Department of Agriculture, Forest Service, Southern Region (USA). Wade, D.D., Brock, B.L., Brose, P.H., Grace, J.B., Hoch, G.A., Patterson, W.A., 2000. Fire in eastern ecosystems. In: Brown, J.K., Smith, J.K. (Eds.), Wildland Fire in Ecosystems: Effects of Fire on Flora. GTR RMRS-42. USDA Forest Service, Rocky Mountain Research Station, Ogden, UT, pp. 53–96. Weise, D.R., Fujioka, F.M., Nelson, R.M., 2005. A comparison of three models of 1-h time lag fuel moisture in Hawaii. Agric. For. Meteorol. 133, 28–39. Wright, C.S., 2013. Fuel consumption models for pine flatwoods fuel types in the southeastern United States. South. J. Appl. For. 37 (3), 148–159.

data. Int. J. Wildland Fire 10 (2), 215–222. Dean, R.T., Dunsmuir, W.T.M., 2016. Behavioral Research 48, 783. https://doi.org/10. 3758/s13428-015-0611-2. Fosberg, M.A., Deeming, J.E., 1971. Derivation of the 1- and 10-hour Timelag Fuel Moisture Calculations for Fire-danger Rating. USDA Forest Service, Research Note RM-RN-207. Rocky Mountain Forest and Range Experiment Station, Fort Collins, CO. Fosberg, M.A., 1970. Drying rates of wood below fiber saturation. For. Sci. 16, 57–63. Gaston, L., Nkedi-Kizza, P., Rao, P.S.C., Sawka, G., 1990. Spatial variability of morphological properties at a Florida flatwoods site. Soil Sci. Soc. Am. J. 54 (2), 527–533. Hiers, J.K., O’Brien, J.J., Mitchell, R.J., Loudermilk, E.L., Cropper, W., Slatton, C., 2009. The Wildland Fuel Cell Concept: an approach to characterize fine-scale variation in fuels and fire in frequently burned longleaf pine forests. Int. J. Wildland Fire 18, 315–325. Hough, W.A., Albini, E.A., 1978. Predicting Fire Behavior in Palmetto-gallberry Fuel Complexes. USDA Forest Service Research Paper SE-174. Kreye, J.K., Kobziar, L.N., Zipperer, W.C., 2013. Effects of fuel load and moisture content on fire behaviour and heating in masticated litter-dominated fuels. Int. J. Wildland Fire 22, 440–445. Kreye, J.K., Hiers, J.K., Varner, J.M., Hornsby, B.S., Drukker, S., O’Brien, J.J., 2018. Effects of solar heating on the moisture dynamics of forest floor litter in humid environments: composition, structure, and position matter. Can. J. For. Res (In Press. Loudermilk, E., O’Brien, J.J., Mitchell, R.J., Cropper, W.P., Hiers, J.K., Grunwald, S., Grego, J., Fernandez, J.C., 2012. Linking complex forest structure and fire behaviour at fire scales. Int. J. Wildland Fire 21, 882–893. Matthews, S., Sullivan, A.L., Watson, P., Williams, R.J., 2012. Climate change, fuel and fire behaviour in a eucalypt forest. Global Change Biol. 18 (10), 3212–3223. Matthews, S., 2014. Dead fuel moisture research: 1991-2012. Int. J. Wildland Fire 23, 78–92. Melvin, M., 2015. National Prescribed Fire Use Survey Report. National Coalition of Prescribed Fire Councils and National Association of State Foresters, Washington, DC. Mitchell, R.J., Hiers, J.K., O’Brien, J.J., Starr, G., 2009. Ecological forestry in the Southeast: understanding the ecology of fuels. J. For. 107, 391–397. Myers, R., Ewell, J.J. (Eds.), 1990. Ecosystems of Florida. University of Central Florida Press, Orlando. National Wildfire Coordinating Group (NWCG), 2014. Interagency Wildland Fire Weather Station Standards and Guidelines. National Wildfire Coordinating Group PMS 426-3. Nelson, R.M., 2000. Prediction of diurnal change in 10-h fuel stick moisture content. Can. J. For. Res. 30, 1071–1087. Nelson, R.M., Hiers, J.K., 2008. Fuel moisture in longleaf pine litter: the influence of fuel bed properties on timelags and drying rates. Can. J. For. Res. 38, 2394–2404. Neter, J., Wasserman, W., Kutner, M.H., 1989. Applied Linear Regression Models. Irwin, Homewood, IL. O’Brien, J.J., Loudermilk, E.L., Hornsby, B.S., Hudak, A.T., Bright, B.C., Dickinson, M.B., Hiers, J.K., Teske, C., Ottmar, R.D., 2016a. High-resolution infrared thermography for capturing wildland fire behaviour: RxCADRE 2012. Int. J. Wildland Fire 25 (1), 62–75. https://doi.org/10.1071/WF14165). O’Brien, J.J., Loudermilk, E.L., Hiers, J.K., Hornsby, Benjamin, Pokswinski, Scott, Hudak,

28