Study area
The study area was located in eastern Puerto Rico, USA (Fig. 1). Nine sites were sampled and partitioned into two forest assemblages representing the subtropical dry forests in Puerto Rico.
The first forest assemblage was referred to as the northeastern subtropical dry forest (Ewel and Whitmore 1973) and included three sites: Ceiba I, Ceiba II, and Las Cabezas, between 18.22° and 18.38° N, and 65.60° and 65.67° W (Fig. 1), which were situated between sea level and up to nearly 100 m in elevation, with a 15.2° average slope inclination (Gould et al. 2006). The average annual temperature was 25.7 °C, and the average annual precipitation was 1416 mm. The soils were generally composed of Alfisol and Hapustalfs (Ping et al. 2013), had an average depth of 0.38 m and a pH of 6.51. The parent material of this soil is Colluvium and Andesitic residuum (Ping et al. 2013). Most abundant trees included Bucida buceras L., Guapira fragrans (Dum. Cours.) Little, Bourreria succulent Jacq., and Gymnanthes lucida Sw.; the shrub layer was dominated by Triphasia trifolia (Burm. f.) P. Wilson, Chamaesyce articulate (Burm.) Britton, Lantana camara L., and Argythamnia stahlii Urb.; and several species of lichens included Macfadyena unguis-cati (L.) A.H. Gentry, Tragia volubilis L., and Serjania polyphylla (L.) Radlk. (Gould et al. 2006).
The second forest assemblage was referred to as the southeastern subtropical dry forest (Ewel and Whitmore 1973) and included six sites in the USDA Forest Service, Institute of Tropical Forestry’s Guayama Research Area, between 18.04° and 18.05° N, and 66.16° and 66.17° W (Fig. 1). These sites were located between 270 m and up to nearly 640 m in elevation, with 28.2° average slope inclination. The average annual temperature was 22.72 °C, and the average annual precipitation was 1693.18 mm. The soils were generally composed of shallow Typic Haplustalfs (Muñoz et al. 2017), with an average depth of 1.1 m and soil pH of 5.32. The parent material of this soil is semiconsolidated volcanic rocket (USDA Soil Conservation Service 1977). The most abundant tree species in this subtropical dry forest included Bucida buceras, Casearia guianensis (Aubl.) Urb., Pictetia aculeate (Vahl) Urb., Nectandra coriacea (Sw.) Griseb., Andira inermis (W. Wright) Kunth ex DC., Guapira fragrans (Dum. Cours.) Little, Randia aculeata L., Zanthoxylum monophyllum (Lam.) P. Wilson, Eugenia foetida Pers., and Leucaena leucocephala (Lam.) de Wit.
Soil sampling and charcoal extraction
A square plot (10 m × 10 m) was positioned at each site. At each plot, a 20 cm × 20 cm area of surface organic matter was removed to expose the mineral soil. Soils were sampled from surface to parent material, at 20 cm intervals per sample layer, to maintain a fine vertical resolution of the extracted charcoal assemblages. In the laboratory, the mineral soils were suspended in 10% potassium hydroxide (KOH) solution for at least 24 h in order to disperse soil aggregates (Inoue et al. 2016). The soils were wet-sieved using superimposed sieves of 5 mm and 2 mm. The macrocharcoal fragments were extracted from the sieves, washed, and weighed.
Wood litterfall collection and branch sampling
Within each plot in the northeastern subtropical dry forest, we randomly installed three baskets of 0.25 m2 at 1 m above ground level. The wood collected in the baskets from each plot was combined into a single sample. Wood was collected every month from January to December 2015.
Five dominant species (Andira inermis, Zanthoxylum monophyllum, Guapira fragrans, Casearia guianensis and Nectandra coriacea) and two other species (Ardisia obovata Desv. ex Ham. and Ficus citrifolia Mill.) in the southeastern subtropical dry forest were selected for wood sampling. Three plants per species were randomly chosen in the southeastern subtropical dry forest. From 10 to 20 Dec 2015, 10 first-year branches per plant were collected. Wood litterfall and branch samples were 65 °C oven dried, and ground through a 1 mm sieve.
Radiocarbon dating
Before radiocarbon dating, charcoal samples were cleaned with 1M hydrochloric acid (HCl) and 1M sodium hydroxide (NaOH) to remove any adsorbed dissolved organic matter. All samples were dried prior to analysis. The radiocarbon ages of 20 charcoal samples from the northeastern subtropical dry forest were determined by AMS (Accelerator Mass Spectrometry) at the Earth System Science Department, University of California, Irvine, USA; and the radiocarbon ages of 58 charcoal samples from the southeastern subtropical dry forest were determined by AMS at the Lawrence Livermore National Laboratory, California, USA.
The calibrated age of charcoal was obtained using the Calib 704 software (Queen’s University Belfast, Belfast, Northern Ireland, United Kingdom). The determination of the calibrated age of each radiocarbon date was based on the weighted average of the highest probability distribution within the 2σ ranges of the starting and ending calendar dates. For each forest assemblage, all of the calibrated radiocarbon dates were pooled in a cumulative probability analysis using the sum probabilities option in Calib 704 to plot the probability that a given event occurred at a particular time to visualize the fire chronology on the Holocene temporal scale. All carbon-14 (14C) dates were presented in cal yr BP (Frégeau et al. 2015).
Estimation of charcoal decay rate
The decrease of charcoal weight with time was caused by charcoal decay and burning of charcoal during subsequent fires. For charcoal found in mineral soils below the surface layer, this decrease of charcoal weight over time was most likely due to charcoal decay, not by fire. Soil charcoal decay is a function of microbial activity wherein soil charcoal is colonized and consumed by soil microbial communities (Moskal-del Hoyo et al. 2010, Tilston et al. 2016). The decay curve of soil charcoal over time is best described as an exponential function. We proposed a novel approach to estimate the charcoal decay rate over time by assuming that the maximum initial size of charcoal that gets into mineral soil in each time interval remains invariant over a 1000-year period. Soil environment for charcoal deposition, such as pore size, drying-rewetting cycles, soil erosion, and burial rates, should be similar over a 1000-year period because soil development is extremely slow and most residential soils are aged for millions of years in the tropics (Birkeland et al. 1992). Thus, we have:
$$ Y={Y}_0\ \left({e}^{-\mathrm{bx}}\right), $$
(1)
where Y corresponds to the maximum weight of charcoal at each age class, Y0 is the maximum weight of charcoal in mineral soil at age zero, b is the decay rate of charcoal (and its inverse value is the average turnover time of charcoal), and x is the calibrated age of charcoal. We used a time interval of 200 yr to identify charcoal with the maximum weight in each age class for the northeastern dry forest, and a time interval of 1000 yr for the southeastern dry forest, ensuring a minimum number of 5 age classes with charcoal presence. We counted the number of dated charcoal samples within each age class and selected two charcoal samples with maximum weight from each age class. We then obtained charcoal decay rate b using linear regression after natural logarithm transformation for both the northeastern dry forest (b1) and the southeastern dry forest (b2).
Estimation of charcoal abundance
The minimum detectable weight was 1.4 mg for analysis at the Earth System Science Department, University of California Irvine; and was 3.5 mg at the Lawrence Livermore National Laboratory. Thus, the number of charcoal particles as a function of time was likely underestimated because there were charcoal particles <1.4 mg that were ≥1.4 mg at their initial weight in the northeastern subtropical dry forest, and <3.5 mg that were ≥3.5 mg at their initial weight in the southeastern subtropical dry forest. To estimate real charcoal abundance as a function of time, we first employed Eq. (1) to estimate the initial weight of charcoal particles that were heavier than 1.4 mg at the time of sampling in the northeastern subtropical dry forest and heavier than 3.5 mg at the time of sampling in the southeastern subtropical dry forest.
The initial weight of the charcoal pieces (Y01) that were older than 200 yr and heavier than 1.4 mg in the northeastern subtropical dry forest was estimated through the equation:
$$ {Y}_{01}=\frac{1.4}{e^{-{\mathrm{b}}_1\left(\mathrm{x}-200\right)}}, $$
(2)
where b1 corresponds to the decay rate of charcoal in the northeastern dry forest, and is the dated charcoal age.
The initial weight of the charcoal pieces (Y02) that were older than 1000 yr and heavier than 3.5 mg in the southeastern subtropical dry forest was calculated using the following equation:
$$ {Y}_{02}=\frac{3.5}{e^{-{\mathrm{b}}_2\left(\mathrm{x}-1000\right)}}, $$
(3)
where b2 corresponds to the decay rate of charcoal in the southeastern dry forest, and x is the dated charcoal age.
We then assumed that the abundance-size distribution of original charcoal (before decay occurs) in mineral soils in each age class remains invariant because depositional environments of soil charcoal are unlikely to change much over the course of soil development within multiple millennium years in a residential soil. Thus, the number of undetected charcoal particles (nud1) that were initially heavier than 1.4 mg and became lighter than 1.4 mg due to charcoal decay in the northeastern dry forest were estimated using the equation:
$$ \frac{n_{\mathrm{ud}1}}{n_{\mathrm{d}1}}=\frac{n_{<{\mathrm{Y}}_{01}}}{n_{>{\mathrm{Y}}_{01}}}, $$
(4)
where nd1 corresponds to the number of charcoal particles that were heavier than 1.4 mg at the time of sampling in each age class >200 yr in the northeastern dry forest; n<Y01 is the number of charcoal particles that were lighter than Y01 but heavier than 1.4 mg at the time of sampling in the 200 yr age class; and n>Y01 is the number of charcoal particles that were heavier than Y01 at the time of sampling in the 200 yr age class. The sum of nud1 and nd1 is the number of corrected charcoal particles in each age class >200 yr in the northeastern dry forest.
Similarly, the number of undetected charcoal particles (nud2) as a function of 1000 yr age interval in the southeastern subtropical dry forest was estimated using the equation:
$$ \frac{n_{\mathrm{ud}2}}{n_{\mathrm{d}2}}=\frac{n_{<{\mathrm{Y}}_{02}}}{n_{>{\mathrm{Y}}_{02}}}, $$
(5)
where nd2 corresponds to the number of charcoal particles that were heavier than 3.5 mg at the time of sampling in each age class >1000 yr in the southeastern dry forest; n<Y02 is the number of charcoal particles that were lighter than Y02 but heavier than 3.5 mg at the time of sampling in the 1000 yr age class; and n>Y02 is the number of charcoal particles that were heavier than Y02 at the time of sampling in the 1000 yr age class. The sum of nud2 and nd2 is the number of corrected charcoal particles in each age class >1000 yr in the southeastern dry forest.
Reconstruction of paleofire history
The random sampling of charcoal does not necessarily assure that all fires will be detected (Frégeau et al. 2015). Therefore, we used EstimateS 9 software (Colwell and Elsensohn 2014) to calculate the estimated fire events based on the observed or corrected charcoal particles. The number of randomizations was set to 100 in the Diversity Settings screen of EstimateS 9. This type of analysis has been used to determine an expected number of species in pooled samples, given the reference sample. The accumulation curves were created according to the relationship between the observed or corrected fire events and dated or corrected charcoal particles. When the curve forms an asymptote, it suggests that most of the fires that occurred at the site have been theoretically estimated. An index was produced based on a nonlinear regression of the mean number of fires detected in relation to the number of dated or corrected charcoal pieces using the following equation:
$$ F(n)=F\left(\mathit{\max}\right)\ \left(1-{e}^{\mathrm{kn}}\right), $$
(6)
where F(n) corresponds to the number of fires observed or corrected, n is the number of charcoal pieces dated or corrected, F(max) is considered here as an estimator of the actual number of fires, and k is the constant controlling the shape of the curve (Fregeau et al. 2015). The F(max) index and the constant k were calculated using the equation of exponential regression in Sigmaplot 14.0 software (Systat Software Inc., San Jose, California, USA). The mean fire interval (I), that is, the average in calibrated years of all the fire intervals, was calculated for each site:
$$ I=\frac{P}{n_{\mathrm{f}}-1}, $$
(7)
where P corresponds to the fire period defined here as the time elapsed between the youngest and oldest fires and nf is the number of fires.
Stable carbon isotope analysis
The ground samples of wood litterfall, live branches, and charcoal were sent to Michigan Technological University’s Forest Ecology Stable Isotope Laboratory, Houghton, USA, for the analyses of carbon isotope composition (δ13C) values using a Costech Elemental Combustion System 4010 (Costech Analytical Technologies Inc., Valencia, California, USA) connected to a continuous flow isotope ratio mass spectrometer. δ13C values were reported in reference to the international Pee Dee belemnite standard (Slater et al. 2001). The Δ13C values of charcoal, wood litterfall, and live branches were calculated through the equation:
$$ {\varDelta}^{13}C=\frac{\updelta^{13}{C}_{\mathrm{air}}-{\updelta}^{13}{C}_{\mathrm{plant}}}{1+\frac{\updelta^{13}{C}_{\mathrm{plant}}}{1000}}, $$
(8)
where δ13Cplant is the isotopic value of the wood litterfall, live branches, or charcoal; and δ13Cair is the isotopic value of the atmospheric CO2 in a specific time period corresponding to a smoothed δ13C curve of atmospheric carbon dioxide (CO2) from 16 100 BC to the present (available at http://web.udl.es/usuaris/x3845331/AIRCO2_LOESS.xls).
There are two stable carbon isotopes in the air: 12C (carbon-12) and 13C. During photosynthesis, plants preferentially take in 12C instead of 13C (i.e., discrimination of the heavy isotope in favor of the lighter one; Fiorentino et al. 2014). In C3 plants under optimal conditions, the stomata are fully open and the flow of CO2 inside the intercellular spaces of the leaf is not limited, leading to discrimination, and thus low δ13C and high ∆13C (Fiorentino et al. 2014). Under environmental stress (e.g., drought), plants typically defend against water stress through stomatal closure, increasing water use efficiency and δ13C, consequently decreasing Δ13C in C3 plants (Fiorentino et al. 2014). This is the basis for the extensively reported relationships between plant Δ13C and environmental variables. In many environmental studies, it is assumed that carbon isotope ratios derived from naturally occurring and anthropogenic charcoal are a direct representation of the isotopic values of the wood tissues from which they were formed, and hence a record of environmental and climatic signals (Hall et al. 2008). C4 plants are not robust enough to be easily applicable to archaeobotanical remains (Tieszen and Fagre 1993). So, for the northeastern subtropical dry forest, the annual mean ∆13C of wood litterfall of 2015 was compared with charcoal ∆13C to infer paleoclimate. For the southeastern subtropical dry forest, the ∆13C of the first-year live branches of 2015 was compared with charcoal ∆13C to deduce paleoclimate. The year 2015 was an extreme drought year in Puerto Rico (Mote et al. 2017).