Nonannual tree rings in a climate‐sensitive Prioria copaifera chronology in the Atrato River, Colombia

Abstract In temperate climates, tree growth dormancy usually ensures the annual nature of tree rings, but in tropical environments, determination of annual periodicity can be more complex. The purposes of the work are as follows: (1) to generate a reliable tree‐ring width chronology for Prioria copaifera Griseb. (Leguminoceae), a tropical tree species dwelling in the Atrato River floodplains, Colombia; (2) to assess the climate signal recorded by the tree‐ring records; and (3) to validate the annual periodicity of the tree rings using independent methods. We used standard dendrochronological procedures to generate the P. copaifera tree‐ring chronology. We used Pearson correlations to evaluate the relationship of the chronology with the meteorological records, climate regional indices, and gridded precipitation/sea surface temperature products. We also evaluated 24 high‐precision 14C measurements spread over a range of preselected tree rings, with assigned calendar years by dendrochronological techniques, before and after the bomb spike in order to validate the annual nature of the tree rings. The tree‐ring width chronology was statistically reliable, and it correlated significantly with local records of annual and October–December (OND) streamflow and precipitation across the upper river watershed (positive), and OND temperature (negative). It was also significantly related to the Oceanic Niño Index, Pacific Decadal Oscillation, and the Southern Oscillation Index, as well as sea surface temperatures over the Caribbean and the Pacific region. However, 14C high‐precision measurements over the tree rings demonstrated offsets of up to 40 years that indicate that P. copaifera can produce more than one ring in certain years. Results derived from the strongest climate–growth relationship during the most recent years of the record suggest that the climatic signal reported may be due to the presence of annual rings in some of those trees in recent years. Our study alerts about the risk of applying dendrochronology in species with challenging anatomical features defining tree rings, commonly found in the tropics, without an independent validation of annual periodicity of tree rings. High‐precision 14C measurements in multiple trees are a useful method to validate the identification of annual tree rings.


| INTRODUCTION
Dendrochronology is a powerful, inexpensive, and relatively fast methodology used to study ecology and growth of tree species that form annual rings, as well as to extend climate records where instrumental data are scarce (Fritts, 1976). In temperate regions, due to a marked seasonality, most of the trees have well-defined annual rings that allow for an extensive application of dendrochronology (Stokes & Smiley, 1996). Despite an increase in dendrochronological studies in the tropics Villalba et al., 2011), major efforts are still needed for its widespread application, due to the challenges of identifying ring boundaries defining the annual nature of tree rings that is essential for developing reliable chronologies, among other factors. The crossdating technique, a fundamental dendrochronological principle, is the matching of growth patterns between treering samples from the same tree and between trees in a given area, which guarantees that the absolute calendar year is assigned to each individual tree ring. While in temperate regions statistically significant crossdating has been accepted as sufficient proof for the identification of annual rings (Stahle, Mushove, Cleaveland, Roig, & Haynes, 1999), in the tropics the annual periodicity of tree layers cannot be taken for granted.
Bomb peak dating using 14 C has been successfully conducted not only in temperate (Biondi, Strachan, Mensing, & Piovesan, 2007;Pearson, Hua, Allen, & Bowman, 2011) and subtropical regions (Biondi & Fessenden, 1999), but also in trees and shrub species from the tropics (Fichtler, Clark, & Worbes, 2003;Groenendijk et al., 2014;Worbes & Junk, 1989). Nevertheless, calendar year misidentifications using 14 C bomb peak dating have been also reported in tree species such as Brachystegia cynometroides, Brachystegia eurycoma, and Daniellia oge in Africa (Groenendijk et al., 2014); Clarisia racemosa Peltogyne cf.heterophylla and Cedrelinga catenaeformis in Bolivia (Soliz-Gamboa et al., 2011); subtropical Callitris endlicheri and the temperate species Callitris glaucophylla in Australia (Pearson et al., 2011). Overall, radiocarbon dating provides an independent validation of the frequency of tree-ring formation for the recent past (e.g., for the last 60 years) and allows testing the synchrony of ring patterns when rings from several years are analyzed. The dendrochronological method allows assigning calendar dates to the tree rings by crossdating tree-ring series from a particular region. The radiocarbon method synchronizes 14 C signatures embedded in individual tree rings to those of the atmosphere obtained from samples of ambient air and/or tree rings around the globe during the bomb peak period (e.g., from 1955 to today). Once crossdated growth rings of a sample and its 14 C signature yield the same calendar age, one can be sure that the dendrochronological ages obtained are correct.
Prioria copaifera Griseb. (Leguminoceae) is an evergreen species that dwells in the Atrato River floodplains (Figure 1). In these environments F I G U R E 1 Study site in the low Atrato River (7°20′N, 76°57′W). Yellow points show hydroclimatic stations: Domingodó Station (77°02′N, 7°10′W;precipitation and water levels;1967temperature;1972-2005, and Bellavista Station (76°57′N, 5°57′W; streamflow; 1965-2002). This map was generated using GeoMapApp (URL: http://www.geomapapp.org) and the Global Multi-Resolution Topography (Ryan et al. 2009) P. copaifera trees are exposed to high flooding levels over more than 8 months per year, with a shorter period of low streamflow. These interannual streamflow fluctuations can be an environmental pacemarker leading to interannual tree growth variations (Grauel, 2004;Herrera & del Valle, 2011). Prioria copaifera growth and ecology has been relatively well studied due to its commercial value, but its phenology and climatic relationships are still unknown (del Valle, 1986;Giraldo & del Valle, 2011;Grauel, 2004). This work will evaluate the annual nature and the climate signal of tree rings in P. copaifera by: (1) generating a reliable tree-ring width chronology from trees located at the low Atrato River in northwestern Colombia using standard dendrochronological techniques; (2) assessing the climate signal recorded by the tree-ring records; (3) validating the annual periodicity of the tree rings using 14 C analyses.

| Site description and climate variability
We collected 15 cross sections from P. copaifera trees in the low Atrato River in Colombia (7°20′N, 76°57′W; Figure 1). Prioria copaifera does not show a clear transition from sapwood to heartwood and exhibits a diffuse-porous wood anatomy. The boundaries between growth layers are defined by marginal parenchyma, without a distinct earlywood-latewood transition ( Figure 2). The vessels are diffuse, solitary or in radial groups of 2-3 vessels, medium size (132 ± 25 μm), and just visible to the naked eye with less than five vessels per mm 2 (López, del Valle, & Giraldo, 2014).
The hydroclimatic records are from the Colombian Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM, Table 1).
The seasonal cycle of precipitation shows higher precipitation from May to November and lower precipitation from December to April, whereas high temperatures last from March to August, and lower temperatures persist from September to February (Figure 3a). The amplitude of the seasonal cycle for temperatures is 1°C and ranges from 26.5 to 27.5°C.

| Sample description and dendrochronological procedure
The sampled cross sections were analyzed at the Forest and Climate Change Laboratory at the Universidad Nacional de Colombia T A B L E 1 Mean Pearson correlation products (r) of the Prioria copaifera tree-ring width chronology with the annual instrumental records (total precipitation, mean temperature, mean water levels, and mean streamflow) and annual climate indices such as the Oceanic Niño (ONI), the Pacific Decadal Oscillation (PDO), the Southern Oscillation Index (SOI), and the North Atlantic Ocean (NAO) F I G U R E 2 Anatomical detail of P. copaifera tree rings: A) image at 75x B) image at 100x, and C) image at 40x. Blue arrows show the tree ring identified as real, white arrows show wedging rings and red arrow shows false rings at Medellín, and a preliminary tree-ring width chronology using 10 of the 15 trees was developed (Herrera & del Valle, 2011). In three trees, 14 C measurements were taken on wood from the ring of the 1994 calendar year for assessing the annual periodicity of the growth bands. Here, we re-processed, re-crossdated, and remeasured the 10 samples used by Herrera and del Valle (2011) and added five samples in order to attain a robust chronology. The samples were progressively polished with 80 to 2,000 grain sandpaper.
To unambiguously mark the tree rings, we evaluated the continuity of the growth layers around the entire circumference and the wedging rings were marked as true rings, identifying when the tree rings were locally mismatched or absent. Discontinuous parenchyma bands identified false rings. A total of 31 ring-width series resulted from measuring 2 or 3 radii from each of the 15 cross sections collected. The visual crossdating was performed by matching tree growth patterns within individual (from the same tree) and among individuals (from different trees) under a Meiji stereo zoom at 40× magnification using the Yamaguchi method (Yamaguchi, 1991 This guaranteed a precision of 0.01 mm in the measurements. The crossdating was validated using the statistical software COFECHA (Grissino-Mayer, 2001;Holmes, 1983) and the dplR package in R software (Bunn, 2008) that identifies possible errors using correlations by running segments between individual tree-ring series and the master chronology. After this validation, the ring-width raw individual series were standardized using a cubic spline with a 50% frequency response at 60 years and an autoregressive model was applied to remove the autocorrelation. The resulting individual residual series were used to build a mean chronology calculated with a biweight robust mean (Cook & Kairiukstis, 1990) using dplR (Bunn, 2008

| Relationships with climate
We used correlation analysis, conducted with the bootres package in R (Zang & Biondi, 2013), to evaluate the monthly and seasonal relationship between the developed residual chronology without autocorrelation and local climate and regional climate indices. The significance of correlations was evaluated with a 95% confidence interval from nonparametric test based on bootstrapping. Only time series that overcame normality, trend, and homogeneity tests were used. The climate variables were local precipitation, temperature, river water levels, and streamflow from the local meteorological stations (Table 1, Figure 1), as well as regional climate indices associated with ocean and atmospheric dynamics such as the Oceanic Niño Index (ONI), the Pacific Decadal Oscillation (PDO), and the Southern Oscillation Index (SOI). The North Atlantic Oscillation (NAO) was also included. However, the NAO index should not show any significant correlation with our chronology as its impacts are mainly centered on the Northern Hemisphere. We also used gridded precipitation data from GPCC V6 0.5° (Schneider et al., 2011) and HaddlSST1 1° sea surface temperature (SST; Kaplan et al., 1998) to evaluate the spatial relationship of gridded precipitation and SST with streamflow records and the residual tree-ring width chronology. The spatial correlations were calculated monthly, annually, and seasonally. The significance of the correlations was evaluated using a nonparametric 90% confidence interval, and only coefficients with p < .10 level were plotted in the maps.

| Radiocarbon dating
The application of 14 C bomb peak dating is based on the excess of  The tree termed Pg25 was randomly chosen, while the trees Pg20 and Pg11 were preselected due to their high correlation with the master chronology.
In each tree, we tested eight tree rings spaced about 10 rings apart to reproduce the 14 C bomb spike in the tropics around the calendar year of 1965. By analyzing samples from single preselected dendrochronological calendar years roughly spaced before and shortly thereafter of the 14 C bomb spike, it is possible to identify potential errors in crossdating and ring identification as well as any other ambiguities in the 14 C sample processing and measurements (Andreu-Hayles et al., 2015). While discrepancies in the dendrochronological calendar dating can be attributed to the presence of partial absent rings or false rings as a result of opportunistic growth not necessarily annual, problems in 14 C sample processing and measurement errors are normally associated with insufficient sample cleaning or contamination, or poor spectrometer performance. The 14 C problems (if any) can be addressed by processing and measuring reference materials alongside with the samples of interest (Santos, Linares, Lisi, & Tomazello Filho, 2015 facility (Beverly et al., 2010). The 14 C data were mass-dependent isotopic fractionation corrected using the online δ 13 C values measured by the spectrometer. Wood blank and reference samples subjected to the same holocellulose extraction procedure were also analyzed alongside with tropical tree-ring samples to aid on accuracy and precision, following established protocols . Measured 14 C values are given using the fraction modern carbon (Fm 14 C) notation (Stuiver & Polach, 1977). Uncertainties smaller than 0.3% were calculated based upon counting statistics, spectrometer isotopic fractionation and background corrections, and the scatter of results from primary and secondary standards, following the data analysis described in Santos et al. (2007). Finally, we compare the Fm 14 C (±SD) value of the samples with the 14 C calibration curve NH-zone 2 (Hua et al., 2013), which better represents the atmospheric 14 C levels of the Northern Hemisphere in the area where the Atrato River site is located.

| Assessment of the stability between climate and the tree-ring records
We where I t stands for the ring-width time series and X t the climatic data.
The "intercept" μ is traditionally a constant, but here it is replaced by a slowly bending trend model (μ t ), the integrated random walk model.
We replaced the constant response weight (α) by a stochastic counterpart α t based on random walk models for individual climate variables.
Both trend (μ t ) and response weights (α) were estimated using the discrete Kalman filter (Harvey & Shephard, 1993;Visser & Molenaar, 1988); this filter is ideal in the sense that it yields the minimum mean square error estimates (normally distributed noise processes) for μ t and α t along with maximum-likelihood estimates for unknown noise variances (Visser et al., 2010). The explained variance used here was computed as: where I t is the ring-width time series, X t is the climatic variable, whereas μ t and α t are the estimated values for μ t and α t , respectively.
It can be understood as the measure of the explanatory power of adding X t to the model, for example, how much of the phenomena is explained by adding an explanatory variable instead of only using the slowly bending trend ȗ t .

| Tree-ring chronology
The

| Climate signals
Prioria copaifera growth is significantly related to annual records of streamflow (positively) and ONI (negatively), while no significant relationships were found with annual precipitation and temperature series from the closest local meteorological station, nor SOI, PDO, and NAO (Table 1). Figure 5 illustrates that most of the significant monthly correlations occurred at the end of the rainy season. In agreement higher seasonal correlations were found during October-December (OND; Table 2) than annually (Table 1)

| Radiocarbon analysis
Sixteen of the twenty four 14 C measurements done in this study from three different trees show significant differences with the calendar dates obtained using dendrochronological methods, whereas eight of twenty-four show good agreement with the NH-zone 2 14 C curve (Hua et al., 2013; Figure 7). Best agreements were found for Pg11

| Instability in the relationships between P. copaifera growth and environmental variables
The correlations between the ring-width chronology and annual streamflow, temperature, ONI, PDO, and SOI were higher when a shorter period from 1980 onwards ( Table 2) was used instead of the maximum time span available for these records (Table 1). In agreement, the STMs calculated for streamflow and ONI (i.e., highest correlations) as dependent variables of growth, which explain 97% of the variance for streamflow and 21% for ONI, confirm these results (Figure 8).
Significant relationship between growth and these climatic records is occurring when the zero value falls outside the dotted line for the stochastic response weight α t (blue lines in Figure 8). The growth-climate relationship shows an early period of instability in both estimators α  Figure 8b). Overall, we detected a stable response after 1980 for both variables ONI and streamflow explaining P. copaifera growth.  1950-2006 .20 1980-2006 Correlations were conducted for the maximum time span available for each record (r 1 ) and for a shorter period from 1980 to 2006 (r 2 ). Significant correlations are indicated by * (p < .05).

| DISCUSSION
T A B L E 2 Pearson correlation products (r) of the Prioria copaifera tree-ring width chronology for the season October-November-December (OND) with instrumental records (total precipitation, mean temperature, mean water levels, and mean streamflow) and annual climate indices such as the Oceanic Niño (ONI), the Pacific Decadal Oscillation (PDO), the Southern Oscillation Index (SOI), and the North Atlantic Oscillation (NAO) with climate), the 14 C results demonstrated a lack of annual tree growth periodicity in three of the trees composing the chronology ( Figure 7).
The significant disagreement between the dendrochronological dates and the 14 C atmospheric signal of the tree rings of P. copaifera cannot possibly be attributed to incomplete removal of extraneous substances during chemical pretreatment. Our holocellulose extraction procedure is robust, and has been proven to produce a perfect tree-ring/ 14 C crossdating match in subtropical tree species during the bomb peak period, when 71 consecutive tree rings were processed F I G U R E 6 Spatial correlation from 1965 to 2002 between: streamflow and GPCC V6 0.5 gridded precipitation annual spanning from January-December (a) and for October-November-December, OND (b); the P. copaifera chronology and GPCC V6 0.5 gridded precipitation annual (e) and for OND (f); the streamflow records and HaddlSST1 annual (c) and for OND (d); the P. copaifera chronology and HaddlSST1 SST annual (g) and for OND (h). Note that only significant correlations (p < .10) are plotted (a) (e) The fraction modern carbon (Fm 14 C) values of the analyzed samples of Prioria copaifera selected from three different trees (Pg11, Pg20, and Pg25) are plotted against the atmospheric record compilations for NH-Zone2 from 1950 to 2010 (Hua et al., 2013) and measured . In addition, the targets produced here from the tree Pg25 were processed and 14 C-AMS measured together with another set of postbomb tropical wood samples inde- pendently validated by us that shows perfect annual frequency of tree-ring formation (Andreu-Hayles et al., 2015).
Problematic anatomical features, such as vague, discontinuous, missing, and false rings are very common issues in tropical species, and it is possible that they may hinder the correct identification of tree-ring boundaries (Brienen & Zuidema, 2006). However, we believe that the large offset in the 14 C dates observed in our study is mostly due to the presence of intra-annual parenchyma band formation that leads to rings with sub-annual resolution. Therefore, P. copaifera trees may have produced one or more rings per year, without a constant frequency, as suggested before by Grauel (2004). The formation of sub-annual rings could be the response to strong intra-annual variability in growth during a given year. In that case, it would be impossible to anatomically differentiate annual rings from sub-annual rings in years when the limiting factors (e.g., streamflow in P. copaifera) vary substantially within a given year.
The formation of fewer subannual rings, specially over the recent period, could be due to the age of the trees: older trees have lower rates of intra-annual structure formation (Dünisch, Ribeiro Montóia, & Bauch, 2003); stand/crown position: trees under suppressed growth conditions tend to form more intra-annual structures (Groenendijk et al., 2014); and/or physiological factors as root depth or less susceptibility to insect attacks (Copenheaver, Pokorski, Currie, & Abrams, 2006). Our results imply some important limitation to use the P. copaifera ring-width chronology for reconstructing the climate along the low Atrato River. The phenology of those trees in the low Atrato River environment makes reliable chronologies very difficult to be developed, but it does not preclude a potential annual pattern when this tree species is growing under distinct environmental conditions. Nevertheless, the annual nature of P. copaifera tree rings was previously reported by McKenzie (1972) using the periodic cambial wounding technique, as well as by Herrera and del Valle (2011) and Giraldo and del Valle (2011) by examining 14 C signature in a single ring of the recent period. This alerts about the danger of generating inaccurate climate reconstructions in the tropics (Herrera & del Valle, 2011) if a thorough independent validation of the annual nature of the tree rings is not performed.
Although obtaining multiple high-precision 14 C-AMS dates is still expensive, it is a good approach to corroborate the annual nature of tree rings for species with challenging anatomical features defining growth layers, never tested before with dendrochronological methods (Andreu-Hayles et al., 2015;Santos et al., 2015), and/or are growing in regions with a complex seasonality. Measurements before and after the bomb spike can be taken in order to obtain a precise assessment of the annual tree-ring identification over the entire chronology. Besides  (Callado, Roig, Tomazello-Filho, & Barros, 2013), periodic cambial wounding (Mariaux, 1967), successive microsampling for evaluations of cambial activity throughout a given period (Krepkowski, Bräuning, Gebrekirstos, & Strobl, 2011), or counting rings in trees with known ages (Dünisch et al., 2003).
Despite the discrepancies between the dendrochronological dates and the 14 C dates, the climate signals detected in the tree-ring chronology seem to be plausible and not accidental, providing insights of the sensitivity of P. copaifera growth to climate. Streamflow was found to be the most important environmental factor limiting P. copaifera growth (Figure 5a). We initially hypothesized that the seasonal extended floods would limit growth due to anoxic conditions (i.e., negative relationship), but we found that floods (high- ENSO is associated with higher than normal precipitation and streamflow over the entire watershed during La Niña events, but drier than normal conditions during El Niño events (Poveda, 2004).
The ONI, SOI, PDO, and SST signatures in the P. copaifera tree rings confirm that ENSO variability may promote larger tree rings during La Niña years; and narrower rings during the El Niño events. Clark, Piper, Keeling, & Clark (2003) and Grauel (2004) reported that P. copaifera diminished during the El Niño of 1997-1998 and increased during the La Niña period of 1999-2000. This pattern was reported in flooded stands similar to our study site, but also in more diverse and drained forests, supporting the climate-growth relationships described here.
Although spurious significant correlations cannot be discarded as the cause for the detection of hydroclimate signals in the P. copaifera chronology, other feasible explanations may be suggested.
We hypothesize that during the most recent years correlations over the correct calendar year between the chronology and climate series, or just lagged 1-2 years, may have occurred because: (1)

| CONCLUSIONS
In dendrochronology it is common practice to assume annual periodicity without further testing when: (1) there is common tree growth variability shared among trees; (2) significant correlations with meteorological records with annual resolution are found. Here, we show how these criteria are not valid for trees of the tropical species P. copaifera Griseb., located at the Atrato River floodplain in Colombia (7°20′N, 77°57′W). Despite that the P. copaifera chronology crossdated well and exhibited strong and coherent climatic signals, these evidences were not reliable proofs of the annual nature of tree-ring formation due to the presence of intra-annual growth bands not anatomically distinguishable from annual rings. The wide range of results reported by different dendrochronological studies in relation to the 14 C analysis lead to the recommendation to conduct multiple 14 C measurements per tree on wood tissues belonging to the sharp ascending and descending portions of the 14 C bomb curve when attempting to corroborate the annual nature of the tree-ring formation, and in more than one tree if possible. It is also advisable to test the stability of the climate signal recorded by the tree-ring chronologies before the development of climate reconstructions or other applications requiring high-precision dating. Larger sampling replication, different sampling sites, and longer reliable climatic records may also aid toward the generation of more reliable chronologies and climate reconstructions in tropical regions.

ACKNOWLEDGMENTS
This research was partially supported by the Department of Forest program. This is LDEO contribution no. 8079.

CONFLICT OF INTEREST
None declared.