Effects of Fire Frequency and Soil Temperature on Soil CO 2 Efflux Rates in OldField Pine-Grassland Forests

Soil CO2 efflux (Rs) is a significant source of carbon dioxide from soils to the atmosphere and is a critical component of total ecosystem carbon budgets. Prescribed fire is one of the most prevalent forest management tools employed in the southeastern USA. This study investigated the influence of prescribed fire on Rs rates in old-field pine-grassland forests in north Florida, USA, that had been managed with prescribed fire annually and biennially for over 40 years, or left unburned for approximately the same period. Monthly measurements were taken of Rs, soil temperature (Ts), and soil moisture from August 2009 to May 2011. Results showed that sites managed with annual and biennial dormant season prescribed fire had significantly lower monthly mean Rs rates and estimated annual soil carbon fluxes than sites where fire had been excluded. While Ts explained a significant amount of the temporal variations in Rs, it did not explain the differences in Rs among prescribed fire treatments. Our results provide new insight into the effects of prescribed fire and fire exclusion on soil carbon fluxes, and suggest that future methods to model ecosystem carbon budgets should incorporate not only current vegetative conditions, but also prescribed fire management activities.


Introduction
The influence of long-term forest management practices on soil CO 2 efflux (R s ) rates plays a key role in determining total ecosystem carbon budgets [1][2][3].It has been estimated that soil CO 2 efflux represents one of the largest global terrestrial fluxes of carbon to the atmosphere, with the total annual R s flux (75 Pg C year −1 ) an order of magnitude greater than current annual anthropogenic carbon (C) emissions from fossil fuel combustion (6 Pg C year −1 ) [4].Soil CO 2 efflux is a function of various interrelated biogeochemical factors that govern the production of autotrophic soil CO 2 efflux by plant roots and associated mycorrhizal fungi (R a ) and heterotrophic soil CO 2 efflux by soil micro and macro biota(R h ) [3][4][5][6].
Fire can influence R s rates by differentially impacting the R h and R a sources of CO 2 [4,7].For example, short-term autotrophic production of CO 2 can be reduced by fire due to both aboveground and belowground plant mortality and injury.While the long-term impacts of fire on R a are variable, R a has been shown in some cases to increase with time since fire as vegetation recovers following disturbance [8].In many cases, the burning of vegetation and surface fuels reallocates nutrient resources via incomplete combustion and subsequent deposition of ash, char, and other residues [8,9].In the short-term period following the deposition of those residues, both plants and soil microbes may respond positively to the availability of such resources, subsequently increasing R s rates.Long-term prescribed fire management regimes may also impact R h sources of CO 2 , by influencing vegetation composition, structure and associated litter and duff quality, production, and accumulation rates [10] which in turn have been shown to influence soil microbial populations and metabolic activity [11].Fire can also reduce R h by killing soil microbes through heating of litter and duff layers and upper soil horizons [4].Previous studies in multiple ecosystems have shown that both fire and forest management can influence R s rates, soil carbon pools, and various coupled biogeochemical processes [12][13][14][15][16].For example, in a study of a mixed conifer forest in California, USA, Ryu et al. [17] found that prescribed fire reduced R s rates while simultaneously altering soil conditions that would otherwise be associated with increased R s rates.However, two studies that investigated the influence of prescribed fire in a different mixed conifer forest in California and an upload oak (Quercus spp.L) forest in Missouri, USA, found that while prescribed burning significantly altered forest floor conditions, there was no clear effect on R s rates [18].The contrasting results reported by these studies demonstrate the complex nature of predicting the influence of prescribed fire on R s .While many factors influencing R s rates have been identified, much remains to be determined regarding the effects of specific forest and land management practices on overall R s rates [2][3][4], especially where management has been long-term rather than experimental.
Frequent prescribed fire is one of the dominant tools for forest management in the pine-grassland forests of the southeastern USA [19,20].Many of the current pine-grassland forests in the region are old-field forest assemblages that exist on former agricultural lands estimated to cover up to 21 million ha across the southeastern USA [21].Across many of these forests, frequent (1-3 year return interval) prescribed fire is used to perpetuate native species assemblages, reduce the risk of destructive wildfires, and promote and restore wildlife habitat [22].Although less frequently cited, the impacts of prescribed fire regimes on soil carbon sequestration are likely to increase in importance as important commodity trade partners adopt carbon credit exchanges (e.g., Canada).The long term consequences of varying fire return intervals for carbon cycling in pine grasslands can inform fire management decisions and projections of carbon sequestration capacity.
This study sought to investigate the impacts of fire regime, specifically annual burning, biennial burning, and prolonged fire exclusion, on R s rates in old-field pine-grassland forests.In addition, this study sought to interpret the potential response of R s to biotic and abiotic factors, including soil temperature, soil moisture, forest stand characteristics, and soil physical properties and chemistry.The intent of this research is to build upon our understanding of the effects of frequent fire on carbon dynamics and sequestration in pine-grasslands and similar woodland and savanna communities worldwide.Studies such as this also provide insight into the response of ecosystem carbon dynamics to forecasted changes in temperature and moisture regimes due to global climate change.

Study Site
The study was conducted on Tall Timbers Research Station (TTRS) in Leon County, FL, USA, approximately 30 km from the cities of Tallahassee, Florida (to the south) and Thomasville, Georgia (to the north) (30 • 39 N, 84 • 12 W; Figure 1).The study utilized research plots established on old field land for long-term study of fire regimes.The Stoddard Fire Plots, which are 0.2 ha plots established in 1960, have each been managed with a constant fire return interval until the present [10,23,24].The study also utilized a 9.2 ha plot named NB66 that had been fire-excluded since 1966 [24].Prior to establishment of the plots, the areas had been burned at 1-2 year intervals since agricultural abandonment which occurred from the late 1800s to the 1920s.Soils within the sites were heavily cultivated for growing corn and cotton from the 1820s until abandonment with subsequent understory and overstory vegetation assemblages highly influenced by past agricultural practices [23] but largely representing a subset of species occurring in native pine communities of the region [10,25].Soils were classified as fine-loamy, kaolinitic, thermic Typic Kandiudults of the Orangeburg and Faceville series (Natural Resource Conservation Service (NRC) Soil Survey Geographic Database (SSURGO)).
and Faceville series (Natural Resource Conservation Service (NRC) Soil Survey Geographic Database (SSURGO)).For this study, sampling took place within three annually burned (1YR), three biennially burned (2YR), and two fire-excluded (UB) Stoddard Fire Plots, and an additional 0.2 ha fire-excluded study area (UB) was established within the Tall Timbers NB66 study site, making three replicates of each fire regime [10,23].Replicate sets of plots each containing the three fire regimes were grouped within soil units, such that two sets were within Faceville (Fine, kaolinitic, thermic Typic Kandiudult) soil units and one was in an Orangeburg (Fine-loamy, kaolinitic, thermic Typic Kandiudult) soil unit.The study sites were located approximately 60 m a.s.l.Average annual precipitation was 137 cm with the majority falling during the summer months of June, July and August (National Climate Data Center 2009, Thomasville, GA, USA).Mean maximum and minimum temperatures for January and July for the area from long-term records   The overstory of the 1YR and 2YR burned plots consisted of a mixture of naturally regenerated shortleaf pine (Pinus echinata P. Mill), loblolly pine (P.taeda L.), and, to a lesser extent, longleaf pine (P.palustris P. Mill).The understory was composed of a mixture of grasses, forbs, and broadleaf woody plants which are typically topkilled by prescribed fire and then resprout [10,19,23,26].The unburned plots, due to the prolonged fire exclusion, contained a closed midstory and overstory canopy of broadleaf deciduous trees of species including but not limited to: water oak (Q.nigra L.), laurel oak (Q.laurifolia Mich.), sweetgum (Liquidambar styraciflua L.), black cherry (Prunus serotina Ehrh.), and flowering dogwood (Cornus florida L.), in addition to mature pine trees.

Sampling
The Stoddard Fire Plots combined with NB66 representing each of three prescribed fire regimes (1YR, 2YR, UB) were grouped into blocks replicated three times for a total of nine plots.Sampling for this study took place within subplots randomly located within each of the Stoddard Fire Plots and NB66.To account for spatial variability within the subplots, measurements were taken at nine locations within in a square grid arrangement with 5 m separation, following Kobziar [16].For this study, sampling took place within three annually burned (1YR), three biennially burned (2YR), and two fire-excluded (UB) Stoddard Fire Plots, and an additional 0.2 ha fire-excluded study area (UB) was established within the Tall Timbers NB66 study site, making three replicates of each fire regime [10,23].Replicate sets of plots each containing the three fire regimes were grouped within soil units, such that two sets were within Faceville (Fine, kaolinitic, thermic Typic Kandiudult) soil units and one was in an Orangeburg (Fine-loamy, kaolinitic, thermic Typic Kandiudult) soil unit.The study sites were located approximately 60 m a.s.l.Average annual precipitation was 137 cm with the majority falling during the summer months of June, July and August (National Climate Data Center 2009, Thomasville, GA, USA).Mean maximum and minimum temperatures for January and July for the area from long-term records   The overstory of the 1YR and 2YR burned plots consisted of a mixture of naturally regenerated shortleaf pine (Pinus echinata P. Mill), loblolly pine (P.taeda L.), and, to a lesser extent, longleaf pine (P.palustris P. Mill).The understory was composed of a mixture of grasses, forbs, and broadleaf woody plants which are typically topkilled by prescribed fire and then resprout [10,19,23,26].The unburned plots, due to the prolonged fire exclusion, contained a closed midstory and overstory canopy of broadleaf deciduous trees of species including but not limited to: water oak (Q.nigra L.), laurel oak (Q.laurifolia Mich.), sweetgum (Liquidambar styraciflua L.), black cherry (Prunus serotina Ehrh.), and flowering dogwood (Cornus florida L.), in addition to mature pine trees.

Sampling
The Stoddard Fire Plots combined with NB66 representing each of three prescribed fire regimes (1YR, 2YR, UB) were grouped into blocks replicated three times for a total of nine plots.Sampling for this study took place within subplots randomly located within each of the Stoddard Fire Plots and NB66.To account for spatial variability within the subplots, measurements were taken at nine locations within in a square grid arrangement with 5 m separation, following Kobziar [16].R s (µmol CO 2 m −2 •s −1 ) was sampled at each subplot using a LI-COR Biosciences LI-8100 automated soil CO 2 sampling instrument with a 20 cm diameter survey chamber (LI-COR Biosciences Inc., Lincoln, NE, USA).The survey chamber fit onto 20 cm diameter × 10 cm height PVC collars inserted 8 cm into the ground.Sampling of R s for all plots occurred monthly.During each monthly visit, all measurements were taken within two days.Over the course of study, any vegetative growth within the sample collars was clipped and removed prior to R s measurement.The PVC collars were installed in June and July 2009 and sampling began in August.The PVC collars and soil surfaces contained within them were protected from fire during burns in March 2010 (1YR plots) and March 2011 (1YR and 2YR plots) by placing slightly larger diameter sheet-metal cylinders around them.As such, the measurements were not expected to respond to direct impacts of prescribed fire, e.g., changes to soil chemistry or injury of microorganisms near the soil surface, but rather general environmental conditions corresponding to the respective fire regime.Sampling consisted of a 120 s measurement initiated by a 15 s "dead-band" during which gases were allowed time to mix within the chamber.During R s measurements, adjacent to each PVC collar, soil temperature (T s ) ( • C) was measured at 10 cm depth using an Omega 8831 type E T-Handle temperature probe, and soil moisture content (M s ) (m 3 /m 3 ) was measured at 5 cm depth using a Decagon Systems EC-5 soil moisture probe mounted on the LI-8100 (Omega Inc., Stamford, CT, USA; Decagon Systems Inc., Pullman, WA, USA).From August of 2009 until February 2010, each monthly sampling period involved measurement of R s , T s , and M s eight times per day in order to assess for diurnal variability.An analysis of the preliminary results found that fewer daily measurements would be sufficient to capture the variability, so daily measurements were scaled back to three times per day (morning, mid-day and late afternoon-early evening) from March 2010 until the end of the study in May 2011.Equipment problems resulted in no field measurements being taken during the month of October 2010, but otherwise there were only few interruptions due to equipment problems or severe thunderstorms.The resulting dataset for the entire twenty-one-month study totaled 7566 R s measurements.Some strong outliers in R s , T s , and M s were attributed to measurement or equipment error and were isolated and excluded from the analyses.
Within a 15 m radius circular plot (0.07 ha) centered on the middle PVC collar in each subplot, basal area (BA) (m 2 •ha −1 ), pine basal area (PBA) (m 2 •ha −1 ), hardwood (broadleaf) tree basal area (HW BA) (m 2 •ha −1 ), and stand density (TPH) (trees ha −1 ) were measured (only counting trees diameter at breast height > 10 cm).Forest stand characteristics were assessed once per subplot in the winter or spring of 2011.Soil total carbon and total nitrogen were measured in January 2013 (Table 1).Subplot level mean litter depth (Litter) (cm), duff depth (Duff) (cm) and total litter and duff depth (DL) (cm) were recorded as the average of three measurements taken from random locations with 30 cm of each PVC collar.Soil was sampled using a 2 cm diameter soil corer to 10 cm depth at 30 locations spread throughout each plot then combined, homogenized, and split to obtain subsamples for analysis.Samples were analyzed at Auburn University to determine percent total carbon and nitrogen with a TruSpec CN analyzer (Leco Corp., St. Joseph, MI, USA) using the dry combustion method, and mg −1 •kg −1 of phosphorus, calcium, magnesium, and potassium were determined with an iCAP analyzer (Thermo Fisher Scientific Inc., Waltham, MA, USA) using Mehlich 3 extraction.Bulk density to 10 cm depth was sampled at five locations within each plot using a 4.5 cm diameter soil sampler (Eijkelkamp Corp., Giesbeek, The Netherlands).Bulk density measurements were used to convert soil chemistry measurements to mass per unit area (kg −1 or Mg −1 per•ha −1 ).

Analysis
Analyses sought to test for significant differences among the three fire regimes with regard to R s , T s , M s and forest environmental conditions measured in the field.For each month, daily measurements per PVC collar were averaged, and the nine PVC collar means were averaged to produce a subplot-level mean value for each month.Repeated measures analysis of variance (ANOVA) was used to test for differences in these monthly means among the three fire regimes (n = 3 per fire regime) for R s , T s , and M s over the twenty-one monthly sampling periods between August 2009 and May 2011.Significant treatment effects were identified at p-value < 0.05.To assess for differences in forest environmental conditions among the three fire regimes, one-way ANOVA tests were applied to each variable separately.Where significant differences were identified, differences were further investigated using Tukey's HSD test.Additionally, linear and nonlinear regression models (Equations ( 1) and ( 2)) were developed to predict the monthly response of R s to T s as well as R s to M s in separate equations within each fire regime and within each season.Non-linear models of the relationships between R s rates and T s per fire regime were explored using an exponential equation (Equation ( 2)) frequently used to describe the response of R s rates to soil temperature [15,18,27,28].The β 1 estimates developed using Equation (2) were used to estimate Q 10 (Equation (3)), which describes the response of R s to a 10 • C change in soil temperature [4, 15,27].All statistical analyses were performed using JMP 9.0 (SAS Institute, Cary, NC, USA).
where MTemp was air temperature, β 0 , and β 1 were coefficients estimated through regression analysis.
Total monthly and annual soil carbon emissions per fire regime were estimated using the nonlinear regression models of R s responses (Equation ( 2)) to changes in ambient temperature (MTemp) following Samuelson et al. [28].Twenty-four hour 2 m elevation ambient temperature measurements recorded hourly from 1 August 2009-31 July 2010 at the Quincy, FL Automated Weather Network site located approximately 30 km from the study sites were used as the MTemp input to estimate hourly R s rates.The estimated R s rates (µmol CO 2 m −2 •s −1 ) were then converted to hourly soil carbon fluxes (g C m −2 •h −1 ) which were then summed to estimate monthly and annual soil carbon fluxes (Mg C ha −1 •year −1 ).

Differences in R s , M s , T s , and Vegetation among Fire Regimes
Mean R s rates varied significantly between fire regimes (p = 0.0007), with the highest R s rates typically in the UB treatment and the lowest typically in the 1YR treatment (Figure 2; Tables 2 and 3).The treatment effect of fire regime on R s rates varied monthly (treatment × time p < 0.0001), with the greatest difference between fire regimes observed during the summer months and the least during the winter months (Table 3).Across all treatments, R s ranged from 0-11.98 µmol CO 2 m −2 •s −1 during the study period, with the highest R s rates during the warmer months and the lowest rates during the cooler months (Figure 2).Although soil temperature did not differ significantly among the fire regimes (p = 0.1007), the lowest temperatures were in the UB sites and the highest in the 1YR sites (Figure 3).A distinct seasonal Ts trend among the fire regimes was observed, with the 1YR sites recording the highest mean Ts in the spring through fall seasons and the UB sites recording the highest mean Ts in the winter seasons.Soil moisture (Ms) did not differ significantly among the fire regimes (p = 0.11), although the lowest soil moisture contents were observed in the UB sites and the highest observed in the 1YR sites (Figure 4; Table 2).Although soil temperature did not differ significantly among the fire regimes (p = 0.1007), the lowest temperatures were in the UB sites and the highest in the 1YR sites (Figure 3).A distinct seasonal T s trend among the fire regimes was observed, with the 1YR sites recording the highest mean T s in the spring through fall seasons and the UB sites recording the highest mean T s in the winter seasons.Soil moisture (M s ) did not differ significantly among the fire regimes (p = 0.11), although the lowest soil moisture contents were observed in the UB sites and the highest observed in the 1YR sites (Figure 4; Table 2).Although soil temperature did not differ significantly among the fire regimes (p = 0.1007), the lowest temperatures were in the UB sites and the highest in the 1YR sites (Figure 3).A distinct seasonal Ts trend among the fire regimes was observed, with the 1YR sites recording the highest mean Ts in the spring through fall seasons and the UB sites recording the highest mean Ts in the winter seasons.Soil moisture (Ms) did not differ significantly among the fire regimes (p = 0.11), although the lowest soil moisture contents were observed in the UB sites and the highest observed in the 1YR sites (Figure 4; Table 2).Forest stand characteristics and litter and duff depths varied significantly among fire regimes (Table 4A).Patterns were related to higher tree stocking and duff and litter accumulation with lower occurrence of fire.Duff depth was over an order of magnitude higher in the UB sites than the 1YR sites (Table 4A).Soil bulk density was significantly higher in the 1YR and 2YR return interval treatments and percent total carbon was higher in the 2YR treatment than in the unburned treatment (Table 4B).Values for nitrogen, phosphorus, calcium, magnesium, potassium, and pH all showed a decreasing trend from the 1YR to unburned treatments, although results were significant only for magnesium (Table 4B).  1 for abbreviations and units.Data in parentheses are standard deviation.Letters per column show significant differences between fire return intervals (Tukey's HSD test); p-Values were considered significant at α = 0.05.

Drivers of Rs within Treatment
Linear regressions indicated monthly mean Ts significantly predicted monthly mean Rs within each fire regime (R 2 > 0.62; Figure 5).In contrast, Ms did not predict Rs (R 2 < 0.14) [29].Nonlinear exponential models also used to predict Rs from Ts and Ms by fire regime showed similar results as the linear models, in which case Ms was also not significant and not reported (Table 5).Model coefficients β0 and β1 from the nonlinear models were similar to those reported by Samuelson et al. [28] and Kobziar and Stephens [15].Q10 values ranged from 1.65 to 2.16 and differed among fire regimes (Table 5).The UB sites had the greatest Rs sensitivity to changes in soil temperature (Q10 = 2.16), while the 1YR sites were the least sensitive to soil temperature changes (Q10 = 1.65).Forest stand characteristics and litter and duff depths varied significantly among fire regimes (Table 4A).Patterns were related to higher tree stocking and duff and litter accumulation with lower occurrence of fire.Duff depth was over an order of magnitude higher in the UB sites than the 1YR sites (Table 4A).Soil bulk density was significantly higher in the 1YR and 2YR return interval treatments and percent total carbon was higher in the 2YR treatment than in the unburned treatment (Table 4B).Values for nitrogen, phosphorus, calcium, magnesium, potassium, and pH all showed a decreasing trend from the 1YR to unburned treatments, although results were significant only for magnesium (Table 4B).  1 for abbreviations and units.Data in parentheses are standard deviation.Letters per column show significant differences between fire return intervals (Tukey's HSD test); p-Values were considered significant at α = 0.05.

Drivers of R s within Treatment
Linear regressions indicated monthly mean T s significantly predicted monthly mean R s within each fire regime (R 2 > 0.62; Figure 5).In contrast, M s did not predict R s (R 2 < 0.14) [29].Nonlinear exponential models also used to predict R s from T s and M s by fire regime showed similar results as the linear models, in which case M s was also not significant and not reported (Table 5).Model coefficients β 0 and β 1 from the nonlinear models were similar to those reported by Samuelson et al. [28] and Kobziar and Stephens [15].Q 10 values ranged from 1.65 to 2.16 and differed among fire regimes (Table 5).The UB sites had the greatest R s sensitivity to changes in soil temperature (Q 10 = 2.16), while the 1YR sites were the least sensitive to soil temperature changes (Q 10 = 1.65).

Seasonal Variation in Rs and Ts Relationships
In the simple linear models and non-linear models predicting mean monthly Rs from Ts within each fire regime showed that the relationship was strongest during the fall and winter (R 2 = 0.69-0.90)and weakest during the spring and summer (R 2 = 0.11-0.63) in all fire regimes (Tables 6 and 7).Q10 also varied seasonally, with the lowest Q10 values observed during summer and the highest Q10 values observed during the winter in all fire regimes (Table 7).

Seasonal Variation in R s and T s Relationships
In the simple linear models and non-linear models predicting mean monthly R s from T s within each fire regime showed that the relationship was strongest during the fall and winter (R 2 = 0.69-0.90)and weakest during the spring and summer (R 2 = 0.11-0.63) in all fire regimes (Tables 6 and 7).Q 10 also varied seasonally, with the lowest Q 10 values observed during summer and the highest Q 10 values observed during the winter in all fire regimes (Table 7).older stands with greater aboveground biomass [30,31].In a trenching and exclusion experiment along a chronosequence of temperate forests in China, Luan et al. [32] found that R s rates were significantly correlated with site basal area (R 2 = 0.59, p < 0.05).
Higher soil respiration rates in the unburned sites may have also been influenced by tree species composition and associated litter quality, specifically dominance by deciduous broadleaf trees which were essentially absent in the burned plots.Soil CO 2 efflux rates have been shown in other studies to be lower in coniferous forests than in broad-leafed forests of the same soil type [5].In a study of R s rates in a mixed conifer-deciduous forest in Belgium, Yuste et al. [33] found that mean R s rates were lower under conifer tree canopies than under deciduous canopies, with total estimated annual carbon flux approximately 50% greater in the deciduous sites (8.8 ± 2.2 Mg C ha −1 •year −1 ) than in the coniferous sites (4.8 ± 0.7 Mg C ha −1 •year −1 ).In their review, Raich and Tufekciogul [5] suggested that the observed differences in R s between coniferous and deciduous forests may have been driven by forest litter production and quality, carbon allocation, and autotrophic contributions to total R s .
In our study, litter and duff depths were highest in the UB sites, reflecting higher tree stocking and lack of fire, in turn possibly contributing to higher R s rates in the UB sites.Other studies have reported that frequent fire greatly reduces the accumulation of litter and duff [34].The lower soil carbon levels and lowest R s in the 1YR plots compared to the 2YR plots likely reflect limited inputs of organic matter because of annual burning.The 2YR plots in turn are assumed to have lower organic matter inputs to the soil than the UB plots because frequent fire, yet the total carbon was higher in the 2YR plots, suggesting lower rates of decomposition than the UB plots consistent with the lower R s measured.It appears that some characteristic of the frequently burned environment suppressed soil respiration, whether through its effect on R a or R h .Increased bulk density associated with burning, found in this and other studies [35][36][37] has been attributed to infiltration of micropore space by ash and char, which might negatively influence microbial activity.Similar patterns of higher total carbon in the mineral soil of burned than unburned areas have been seen in many long-term fire experiments in the southeastern USA [36,[38][39][40][41].
The R s rates observed in this study resulted in lower estimated total annual soil carbon emissions in the 1YR (37%) and 2YR (25%) sites relative to the UB sites.The estimated monthly carbon emissions for all sites were similar to those reported by Samuelson et al. [28] for an unburned loblolly pine plantation in southwestern Georgia, USA, while the estimated total annual soil carbon emissions reported in our study were similar to those (14.10Mg ha −1 •year −1 ) reported by Maier and Kress [42] for an unburned loblolly pine plantation, but greater than those (7.78-9.66Mg ha −1 •year −1 ) reported by the Samuelson et al. [28] study.
Temporal variability in R s rates was explained more by soil temperature than any other recorded parameter.These results were consistent with other studies in southeastern USA forest systems in which R s rates were found to have strong correlations with soil temperature [28,42,43].The R s correlation with T s reported in this study (R 2 = 0.62-0.78 in linear models and R 2 = 0.60-0.80 in nonlinear models) were higher than those reported by Samuelson et al. [28] for a Georgia, USA loblolly pine plantation (R 2 = 0.38-0.56),much higher than those reported by Templeton et al. [43] across a range of loblolly pine plantations in 11 southeastern USA states (R 2 = 0.45), and similar to those reported for a loblolly pine plantation in North Carolina, USA (R 2 = 0.70) [42].Caution must be used in comparing correlations among studies, as variations in the modeled R s and T s relationship may be influenced by statistical modeling techniques regardless of the biogeochemical couplings observed in the field.While T s explained the majority of the temporal variability in R s rates, a lack of significant differences in T s among treatments suggests that temperature was not the key factor contributing to the differences in R s rates among the treatments.
Over the course of this study's sampling period, relationships between R s and T s varied seasonally, with stronger relationships during the winter seasons.This pattern was evident in the R 2 values of the seasonal linear and nonlinear models as well as in the seasonal Q 10 values.Distinct seasonal variations in Q 10 values have been noted by others [44,45] and such variations have potential implications for modeling R s values based on temperature measurements.Yuste et al. [45] suggested that while annual Q 10 values may sufficiently model annual R s rates, seasonal or shorter-term estimates of R s should be based on season specific Q 10 functions that capture the seasonal variation in R s temperature response, as observed in our study.
The seasonal variation in the relationship between R s and T s may be the result of phenology-related shifts in the relative contributions of R a and R h to R s .Previous research in partitioning the sources of R s have shown that during periods of aboveground vegetative growth, R a contributions to R s can increase relative to R h , as plants allocate recent C photosynthate belowground, driving higher root maintenance, root growth, and mycorrihizal fungal respiration rates [6,46].In addition, other studies have shown that during periods of aboveground vegetative growth, the T s and R s relationship weakens as other variables such as soil moisture and available photosynthetically active radiation (PAR) become important in governing belowground C allocation by plants [47][48][49].Similar to our findings, Fenn et al. [50] found that soil temperature explained less of the variation in R s during the summer than during the spring in a multi-season study of R s rates in a woodland in Oxfordshire, UK.

Conclusions
In this study comparing long-term fire disturbance intervals, soil temperature explained the majority of the temporal variation in R s rates, but neither temperature nor moisture explained the differences in R s observed among the fire frequencies.As such, we suggest that future efforts to model soil carbon fluxes in otherwise similar burned and unburned forests should account for the impacts of both seasonal variability and prescribed fire management regime on the relationships between R s and T s .Differences in CO 2 efflux rates, as well as soil chemistry and bulk density, reflect the long-term impact of contrasting fire disturbance regimes, as well as the biophysical consequences of those regimes.Lower total annual soil C emissions, combined with lower rates of soil C efflux and higher stored soil C, suggest that even biennially burned sites are effective at sequestering soil carbon.

Figure 1 .
Figure 1.The research site, Tall Timbers Research Station, was located in Leon County, Florida, USA.The site is approximately 30 km north of the city of Tallahassee, Florida, USA.

Figure 1 .
Figure 1.The research site, Tall Timbers Research Station, was located in Leon County, FL, USA.The site is approximately 30 km north of the city of Tallahassee, FL, USA.
are 16.8 • C and 4.6 • C for January and 33 • C and 21.8 • C for July (National Climate Data Center 2009, Thomasville, GA, USA).

Figure 2 .
Figure 2. Monthly mean soil CO2 efflux rates (Rs) (μmol CO2 m −2 •s −1 ) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, Florida, USA.Data were from 27 sample points per fire regime, measured three-times daily once per month.Measurements were not taken in October of 2010.

Figure 3 .
Figure 3. Monthly mean soil temperature (Ts) (°C) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, Florida, USA.Mean monthly 2 m air temperature also shown for comparison.Data were from 27 sample points per fire regime, measured three-times daily once per month.Equipment problems eliminated measurements taken in January 2009 and February 2011.Measurements were not taken in October of 2010.

Figure 2 .
Figure 2. Monthly mean soil CO 2 efflux rates (R s ) (µmol CO 2 m −2 •s −1 ) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, FL, USA.Data were from 27 sample points per fire regime, measured three-times daily once per month.Measurements were not taken in October of 2010.

Figure 2 .
Figure 2. Monthly mean soil CO2 efflux rates (Rs) (μmol CO2 m −2 •s −1 ) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, Florida, USA.Data were from 27 sample points per fire regime, measured three-times daily once per month.Measurements were not taken in October of 2010.

Figure 3 .
Figure 3. Monthly mean soil temperature (Ts) (°C) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, Florida, USA.Mean monthly 2 m air temperature also shown for comparison.Data were from 27 sample points per fire regime, measured three-times daily once per month.Equipment problems eliminated measurements taken in January 2009 and February 2011.Measurements were not taken in October of 2010.

Figure 3 .
Figure 3. Monthly mean soil temperature (T s ) ( • C) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, FL, USA.Mean monthly 2 m air temperature also shown for comparison.Data were from 27 sample points per fire regime, measured three-times daily once per month.Equipment problems eliminated measurements taken in January 2009 and February 2011.Measurements were not taken in October of 2010.

Figure 4 .
Figure 4. Monthly mean soil moisture content (Ms) (m 3 /m 3 ) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, Florida, USA.Mechanical difficulty resulted in erroneous data collected in August 2009.Measurements were not taken in October of 2010.

Figure 4 .
Figure 4. Monthly mean soil moisture content (M s ) (m 3 /m 3 ) for three prescribed fire regimes at the Tall Timbers Research Station near Tallahassee, FL, USA.Mechanical difficulty resulted in erroneous data collected in August 2009.Measurements were not taken in October of 2010.

Forests 2017, 8 , 274 9 of 14 Figure 5 .
Figure 5. Linear regression of the relationships between monthly mean soil CO2 efflux rates (Rs) (μmol CO2 m −2 •s −1 ) and monthly mean soil temperature (Ts) (°C) for three prescribed fire intervals at the Tall Timbers Research Station near Tallahassee, Florida, USA.Each point represents monthly mean values per sample plot.p-Values were considered significant at α = 0.05.

Figure 5 .
Figure 5. Linear regression of the relationships between monthly mean soil CO 2 efflux rates (R s ) (µmol CO 2 m −2 •s −1 ) and monthly mean soil temperature (T s ) ( • C) for three prescribed fire intervals at the Tall Timbers Research Station near Tallahassee, FL, USA.Each point represents monthly mean values per sample plot.p-Values were considered significant at α = 0.05.

are 16.8 °C and 4.6 °C for January and 33 °C and 21.8 °C for
July (National Climate Data Center 2009, Thomasville, GA, USA).

Table 1 .
Environmental variables measured in each research plot, with abbreviation, unit of measurement, and measurement regime provided.Monthly measurements were made from August 2009 to April 2011.

Table 2 .
Results of the repeated measures ANOVA showing effects of prescribed fire regime (FR); and time (monthly means of measurements) on soil CO 2 efflux (R s ), soil temperature (T s ) and soil moisture content (M s ) (n = 9 three replicates within three treatments).
p-Values were considered significant at α = 0.05, df = degrees of freedom.

Table 4 .
Mean forest characteristics per prescribed fire treatment (FR).See Table

Table 4 .
Mean forest characteristics per prescribed fire treatment (FR).See Table

Table 5 .
Linear and nonlinear regression relationships between soil CO2 efflux rates (Rs) and soil temperature (Ts) by fire return interval (FR).

Table 5 .
Linear and nonlinear regression relationships between soil CO 2 efflux rates (R s ) and soil temperature (T s ) by fire return interval (FR).