Pinus sylvestris (Scots Pine) is a sub-boreal evergreen conifer with one of the largest distribution ranging between Scotland and northeast Asia (Nikolov and Helmisaari, 1992) at altitudes between 0 to 2700 m asl. As most conifers, it contains uniseriate rays characterized by a single layer of parenchyma cells (Lev-Yadun and Aloni, 1995; Figure 1) facultatively embraced by tracheid cells on the upper and lower extremities, and sometimes enclosing resin ducts (Brown et al., 1949).

Wood samples for this study were obtained in May 2013 in the xeric Pfynwald forest located in the Swiss Rhone Valley (Valais). The climate at this site is continental with 600–700 mm of annual precipitation and a mean annual temperature of 10.1°C (data from 1981 to 2010 of the weather station of Sion, at 20 km distance from the site, MeteoSwiss). Cores with a diameter of 10 mm were taken at stem breast height from 40 mature Scots pines of 10–13 m height, 12–30 cm dbh, and aged between 45 and 135 years. Ring widths in the cores were measured using a LinTab device and time series were cross-dated using COFECHA (Grissino-Mayer, 2001) to assign each ring to the correct calendar year. The 20 outmost annual rings from all the 40 cores were considered for cross-sectional analyses, whereas a subset of six cores was used for additional tangential and radial analyses at three locations each separated by five to seven annual rings. In addition, a single stem disc of 14 cm diameter (58 years old) was collected from a felled individual, split in three radial bars of 20 mm width and 20 mm height at offsets of 120°, and used for tangential and cross-sectional analyses in the sapwood rings 1989, 2002, and 2012.

Wood sample where processed according to Schweingruber and Poschlod (2005) for the anatomical characterization and quantification of the rays. Therefore, 10–15-μm permanent thin sections were produced with a sledge microtome (Gärtner et al., 2014), placed on a slide and stained with Alcian blue (1% solution in acetic acid) and safranin (1% solution in ethanol) to differentiate between unlignified (blue) and lignified (red) cells. Afterwards, sections were dehydrated using a series of ethanol solutions of increasing concentrations, washed with xylol, and finally permanently preserved by embedding them into Eukitt glue (Gärtner and Schweingruber, 2013). Overlapping images covering the entire samples were captured with a Nikon D90 digital camera mounted on a Nikon Eclipse 50i optical microscope with 40 × magnification and merged using PTGUI v8.3.10 Pro (New House Internet Services B.V., Rotterdam, the Netherlands). In the (subset) of the 40 cores, the average measured width and surface in cross-sectional, tangential and radial sections were 6.69 mm, 4.87 and 9.39 mm², respectively. In the stem disc the average measured width in cross-sections was 19 mm, while the average measured surface in tangential sections was 28 mm².

The outlines of all rays and tree ring borders were vectorized manually in the images using a tailored version of ROXAS 1.6 (von Arx and Dietz, 2005; von Arx and Carrer, 2014), a specialized image analysis tool for wood cell anatomical measurements based on Image-Pro Plus (Media Cybernetics, Silver Spring, Maryland, USA). As a result ROXAS provided several statistics in a global and annual resolution such as individual ray area (all planes), individual ray length (cross- and radial sections) and ray height (tangential sections), the number and position of initiating (NEWRAY) and disappearing rays within the ring (cross-sections), and the percentage of ray surface in the xylem (PERPAR; all cutting planes).

Several trials were used to identify reliable quantification methods and assess distinctness of ray metrics based on the subset of the six cores. First, the consistency of the measured percentage of ray surface in the xylem (PERPAR) in the three cutting planes (cross-, tangential, radial plane) was evaluated by comparing values from the same annual rings. Second, to explore the potential of more efficient ray quantification, the relationship of individual ray area and ray length (cross-sections) and ray height (tangential sections) was investigated by linear regression analyses. Third, the similarity of different ray parameters in the tangential sections such as PERPAR, ray density, ray area, ray height, and ray width was assessed by correlation analyses. Ray width was estimated from ray area through division by ray height.

The variability of ray parameters was assessed in two different ways. First, the inter-annual variability in cross-sectional PERPAR was determined in the full set of 40 cores and time series of 20 rings each by calculating the coefficient of variation (CV) and the mean sensitivity (Cook and Pederson, 2011), i.e., the average percentage change from one yearly value to the next. The same sample set was used to assess the intra-annual variability of cross-sectional PERPAR and the proportion of initiating rays (NEWRAY) between early- and latewood using t-test. Second, the variability of the obtained PERPAR values was assessed as a function of the measured wood surface by bootstrapping. Using tangential sections from each of the three locations in the three radial bars of the stem disc, one thousand 1-mm² measurement windows were randomly placed and the PERPAR values were extracted and pooled to a single data pool (n = 9000). Bootstrapped values based on 1000 replications for PERPAR mean and coefficient of variation of its estimation (CV) were obtained by randomly combining an increasing number of individual 1-mm² values from the data pool, thus simulating the measurement of increasing wood surface (from 1 to 15 mm²). The same procedure was repeated for the cross-sectional images from the same annual rings as the tangential sections for steps of 1-mm width (from 1 to 15 mm). Since mean ring width was around 1 mm (1.047 mm), a measured width of 1 mm in cross-sections corresponded to 1 mm² in tangential sections, thus allowing direct comparisons of the obtained variability between the cutting planes. The calculated CVs were used to calculate the relationship between measured wood width/surface, number of samples (n), and estimated accuracy. These calculations were based on Equation (1): (1)CI95=2·CV·n-0.5 which expresses the relative 95% confidence interval (CI95; given as a percentage of the mean) as a function of CV and n. In fact CI95 approximately corresponds to twice the standard error (SE), SE equals to SD divided by the square root of the number of samples (n), and SD is CV multiplied by the mean.

In this study we evaluated several methodologies for quantifying ray features in conifers. Tangential and cross-sections proved to be suitable, but with strengths and limitations in terms of accuracy and investment, and with differences in the information they register. In contrast radial sections were generally unsuitable for ray quantification. The large differences in percentage of ray surface (PERPAR) estimation (up to three times) among cutting planes likely reflects sampling artifacts linked to the sheet-like orientation and fusiform shape of uniseriate rays (see Figures 1D,E). In fact, in a radial section, the position and contour of the ray outline may change substantially from the lower to the upper side of the section. Because of the transparency of the tissue, the perceived ray outline is the maximum ray projection through the entire thickness of the section, which leads to a systematic overestimation (“radial overestimation artifact”). This effect–although much weaker—also occurs in cross-sections due to the tapering toward the upper and lower ray extremities, supposedly explaining the overestimated values (Figure 2). In contrast, the tangential sections are robust in this respect and therefore likely produced most accurate estimates of PERPAR, and therefore relative ray volume (Myer, 1922). The significant correlation between tangential and cross-sectional data confirms the systematic nature of the larger cross-sectional values. With respect to the quantification of the area of individual rays, the strong allometric relationship allows accurate estimation of ray area as a function of ray length in cross-sections (Olano et al., 2013) and ray height in tangential sections. Using such relationships could significantly increase the efficiency of the measurement procedure.

Among the different tangential ray parameters, PERPAR positively correlated with all other ray metrics but mean ray width, which suggests PERPAR incorporates the information from most other metrics (Figure 4). Notably, the significant correlations with mean ray area and height are in line with previously observed consistent patterns of these metrics in relation to tree vigor and growth conditions (e.g., Bannan, 1954, 1965; White and Robards, 1966; Fonti et al., 2015). Moreover, the missing relationship between mean ray height and width indicates that their ratio mainly varied among the analyzed images (i.e., trees and/or annual rings), since within the images individual ray height well scaled to the width (mean Pearson's r = 0.475; P < 0.001). This variability is mainly due to the variability in ray width: supplementary analyses revealed that the width of individual rays is less strongly correlated with ray area (mean Pearson's r = 0.750; P < 0.001) than ray height (mean Pearson's r = 0.899; P < 0.001). Mean ray width also showed a larger variability among trees and/or annual rings than mean ray height (CV = 0.169 vs. 0.094). Together with its independence from PERPAR, mean ray width might therefore be a more promising parameter for ecological studies than mean ray height.

The higher PERPAR and marginally larger number of initiating rays (NEWRAY) observed in the latewood compared to earlywood was unexpected when considering the radial orientation of the rays. In fact, once initiated, rays grow and extend to keep the connection with the cambium and phloem (Fischer and Höll, 1992; Spicer, 2014). Such small intra-annual differences could indicate slightly larger parenchyma cells (DeSmidt, 1922) and/or a higher ray initiation rate in latewood than in earlywood. They potentially evidence an advantage of having a larger storage and transport capacity close to the cambium to support the onset of cambial activity in the following growing season. However, the small intra-annual difference in PERPAR values also indicates that it should suffice to only roughly balance early- and latewood in a proportional way when analyzing PERPAR in tangential sections. Additionally, if the section was not taken fully parallel to the orientation of the rays, the number of NEWRAY in cross-sections may be overestimated (see Figure 1B). The extent of overestimation in ray initiation is probably directly related to the number of disappearing rays (“ending ray artifact”; Eckstein, 2013), because they are both directly linked to the orientation of the section (see Figure 1C). Nevertheless, time series of the number of initiating rays from samples with an orientation problem still represent valid data if standardized before statistical analysis.

The accuracy of PERPAR greatly changed depending on the measured wood width (cross-sections), surface (tangential sections), and the number of samples (Figure 5). Yet, our results suggest that common measurement strategies—10 cross-sections of 5-mm increment cores or 10 tangential planes of 1 mm²—provide rather inaccurate estimates of PERPAR. They may thus be often inappropriate to extract an ecological signal considering that year-to-year variability (MS) in PERPAR is about 20% (Olano et al., 2013; Fonti et al., 2015), or even only about 11% as observed in the 40 cores analyzed here. This study estimates how an increased number of samples [since, CI95 decreases with the square root of the number of samples (cf. Equation 1)], or expansion of the measured wood width/area will increase the accuracy. Although our assessment was based on a very detailed analysis of a single stem disc, it cannot be excluded that the CI95 curves may be shifted for Scots pines from other populations or different species. However, we speculate here, that a CI95 in the range between ±5 and ±10% (dotted horizontal lines in Figure 5), corresponding to approximately half the CV or MS, could represent a reasonable balance between data accuracy and measurement efficiency.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.