We performed a systematic Web of Science literature review in May 2024 (in accordance with Foo et al., 2021) using the search terms: “Wood anatomy” and “Time series” to obtain a concrete idea on the amount of manuscripts working on QWA and which inter-annual parameters they analyse. Literature spanned from 1989 until 2024 was considered which was published in peer-reviewed English journals. Each source was inspected for relevance and afterwards stored in a database. For each stored source, we noted the assessed plant material, the location of the study, the year of publication, specifications on the sample number and annual range, the sample width (i.e., increment core width) and data treatment, the main environmental conditions driving the inter-annual variability (if reported), and the specific anatomical parameter of interest. A total of 145 studies were deemed relevant and were screened, as they dealt with wood anatomical time series, and MLA, VD, VN, D _h and RCA were selected for the inter-annual analysis within this study ( Figure 1 ).

Wood samples were selected based on their maximum lumen size (in μm) and whether they cover a consistent range in the angiosperm vessel-area distribution ( Figure 2 ). Carpinus betulus was selected at the lower end of the vessel-area distribution, while Quercus robur wood has the largest maximum lumen area. With this selection we also cover a large porosity gradient (i.e., from diffuse-porous to ring-porous wood). From the selected trees, one wood increment core (using an increment borer; Haglöf, Sweden) with a diameter of 5 mm and length of ≈15 cm was taken ( Table 1 ). During the selection we ensured that there is no clear signs of tension or reaction wood. Sample selection was based on the quality of the sample (i.e., no cracks or rotten sections), covering at least 10 years of annually distinct rings. With these criteria a collection of angiosperm wood samples was obtained where for each core the three image acquisition methods were applied ( Table 1 ). For each of these samples QWA analysis was performed using the three specified image acquisition methods, surface pictures, micro-sections, and micro-CT volumes. Due to the nature of the methods, it is impossible to make images of exactly the same wood surface area due to vertical offsets.

The wood surface of each core was prepared by using a core microtome (Gärtner et al., 2015) to create a level transversal surface. The wood surface was blackened using a black marker and the cell lumina were filled by carefully rubbing white chalk powder onto the sample to enhance the contrast between cell wall and lumen (i.e., Fonti et al., 2007; Klesse et al., 2021). The prepared samples were positioned within an adjusted Atrics system (Levanič, 2007), referred to as “Skippy”, consisting of a motorized stage (minimum precision 0.01 mm) with a controlling unit used for moving the camera from the bark to pith direction (Gärtner et al., 2024). A high-resolution Canon 5DsR camera with a 100 mm macro lens (Canon Inc. Tokio, Japan) was used to obtain surface pictures. Images along the radial direction were stitched using PTGui (New House Internet Services B.V., Rotterdam, NL; see von Arx et al., 2016). The camera settings were set to an f-value of 7.2, with an obtained theoretical maximum resolution of 4.14 μm (or 0.2381 pixels per μm, 6,148 dpi) defined by the camera’s image sensor and the true 1:1 image reproduction ratio of the macro lens. Calibration values were generated by performing multiple measurements of mm-paper for each image. On average our pictures have a resolution of 6000 ± 25 dpi (based on laboratory-tested user-based in-accuracy during image resolution calibration). Stitched images of the entire sample were exported as JPEG files usable within image analysis software. Due to the difficulty in identifying tree rings on a species like C. betulus, we were unable to use the surface picture of this species within the comparison.

Classical wood anatomical micro-sections were taken from the transversal surface (Gärtner and Schweingruber, 2013). Wood samples were split into ≈4 cm radial samples using a cutter and prepared for paraffin embedding, to stabilize the tissue for obtaining high quality thin sections (von Arx et al., 2016; but see Frigo et al., 2024). Almost all samples, except the samples from A. glutinosa (see Anadon-Rosell et al., 2022 for details on processing), were embedded in paraffin according to standard protocol using a tissue processor (Leica TP1020, Leica Biosystems, Nussloch, Germany; detailed in Hamann et al., 2011). Each paraffin block was placed within a rotatory microtome (Leica RM2245, Leica Biosystems, Nussloch, Germany) to cut 12 μm thick cross-sections. The thin sections were stained with 1:1 safranin and astrablue solution and permanently fixed on glass slides using Euparal. Digital images of the sections with a resolution of 0.44 μm (or 2.27 pixel per μm, 57,600 dpi) were taken using a slide scanner (Axio Scan Z1, Carl Zeiss AG, Germany). Automatically stitched images were exported as JPEG files for further image analysis.

Each wood sample was measured for transverse wood anatomical features along the radial direction using the Nanowood X-ray CT scanner. This scanner was custom-built at the UGent Centre for X-ray Tomography (UGCT, www.ugct.ugent.be; Ghent, Belgium) and recently refurbished in collaboration with TESCAN-XRE (TESCAN ORSAY HOLDINGS a.s.), a former UGCT spin-off company. It is specifically designed to study materials made of wood or derived from wood. The X-ray source, a nano focus tube, was used to scan the samples at an average voltage of 50kV and a current of 40 μA, using 700 ms exposure time per projection and 3600 projections. All scans in this study were reconstructed using the Octopus Reconstruction software (Vlassenbroeck et al., 2007). With the described setup, the obtained reconstructed voxel pitch, further referred to as resolution, was 3.6 × 3.6 × 3.6 μm, which is similar to the 0.277778 pixel per μm and was used for the comparison between the image acquisition methods (7,060 dpi). Although submicron resolutions are obtainable with this system (Van den Bulcke et al., 2019), we optimized the resolution for both comparability and scanning time-efficiency. Scanning was performed along the entire wood sample length. A “slice” was selected parallel and closest to the surface, to maintain comparable anatomical features compared to the other methods (i.e., reducing variability in both cell size and position when moving away from the surface). The slice was pre-processed to remove noise and enhance the image, by applying histogram equalization to transform the values of the grey-scale image such that contrast improved. Subsequently the images were binarized using the topological derivative of Larrabide et al. (2008). All image processing was performed using MATLAB^® and Octopus Analysis+ (Vlassenbroeck et al., 2007), to binarize the image and de-speckle the image, and export the image into the JPEG format for further analysis.

The specialized software ROXAS (von Arx and Carrer, 2014), based on Image-Pro Plus (Media Cybernetics, Rocksville, MD, USA), was used to semi-automatically measure the lumen size of conductive cells for each method (fibres and vessels in broad-leaved species; von Arx et al., 2016). Each imaging method was analysed with their specified resolution (in pixel per μm), while using a species-specific ROXAS configuration file (see von Arx and Carrer, 2014). Selecting the exact same area for analysing the wood was not possible, as the analysed surface was not at the same sample depth between methods, moreover some methods allow for a wider area of interest which we deem part of the merits of a specific method. For all methods, care was taken to manually detect any previously undetected vessels within the section. Here, we assumed high sample quality for our comparison. Moreover, for the thin section with the highest resolution, detailed manual corrections were performed to draw most fibre lumina which serve as the true reference for comparison. A general distinction between fibres and vessels was used where elements with a lumen area above 500 μm² were considered as vessels (see Table 1 ; Supplementary Figure S3 ). This threshold was selected based on manual observations and the fact that we wanted a uniform threshold across species, instead of smaller species-specific thresholds (see Peters et al., 2020a). Ring boundaries were manually drawn and this information was stored for informing the inter-annual and sectorial approaches, described below. For each detected cell, we recorded: 1) the year in which the cell is positioned, 2) the calibrated positional information (X _cal and Y _cal in μm within the annual ring), 3) the lumen area (A _l in μm), and 4) the radial distance from the ring boundary (RadDistR in μm and % of the annual ring).

To investigate the results of the different imaging methods to capture ring-specific QWA parameters at the tree-ring scale, we calculated for each ring the top five most commonly used parameters ( Figure 1 ): Mean Lumen Area (MLA), Vessel Density (VD), Vessel Number (VN), Mean hydraulic diameter (D _h), and Relative Conductive Area (RCA). As the different imaging methods will detect different minimum sized cells, we performed a threshold analysis across all rings where we assessed when the MLA of the average tree rings converged between the different methods ( Supplementary Figure S1 ). Specifically, within this analysis we inspected at which cut-off value the mean lumen area is different between the methods, which is caused by the difference in the methods capability to detect smaller xylem elements. For the diffuse porous species (C. betulus, A. glutinosa, and F. sylvatica) the cut-off was set at 1000 μm², while 5000 μm² was used for the ring-porous species (Q. robur). These thresholds are commonly used thresholds in literature (i.e., Granda et al., 2018; Klesse et al., 2021). For each ring the cell properties and the total number of visible features were obtained from ROXAS, where VD and RCA were calculated using the area of interest, measured manually for each ring. VN was calculated with the VD, by multiplying this number by the ring area which was calculated using a fixed 5 mm width of the core surface and the ring width measurement. This was done to avoid that wider ring areas would bias VN. For each annual ring we also considered xylem anatomical parameters which relate to hydraulic efficiency. The key parameter describing this efficiency in literature is the hydraulically weighted mean vessel diameter (D _h) calculated as described in Equation 1 (Kolb and Sperry, 1999). Thereby, the cell-specific hydraulic diameter (D _h.cell) was used to account for elliptical lumen areas (Lewis and Boose, 1995).

To see the effect of the method on the intra-annual tree-ring variability, we performed a sectorial approach, generating tracheidograms for the samples based on distance instead of on cell count (Peters et al., 2018), and determined ecophysiologically relevant parameters such as hydraulic conductance and wood density. We used the positional information obtained from the vessels and fibres (X _cal and Y _cal of the cell centroid) in combination with the ring boundaries. Each sector was assigned a fixed radial width (from the pith to the bark direction) for which a “sector box” was determined. For all species, except for Q. robur, we used a fixed sector width of 120 μm which corresponds to the average diameter of larger vessels in these species. For Q. robur we selected a larger sector width of 450 μm, as the earlywood vessels are substantially larger ( Table 1 ). After isolating all cell elements, with their centroid in the sector box, three QWA features were extracted and linked to the central radial distance of the sector box. First, the median lumen area was determined, as initial testing confirmed that this was more robust than using the MLA. Second, the theoretical hydraulic conductance (K _H in m⁴ MPa^-1 s^-1; see Equation 2) was determined as a useful hydraulic proxy to define rehydration potential (Peters et al., 2023), which relates to the D _h. K _H was calculated according to the Hagen-Poiseuille law (Prendin et al., 2018b; Tyree and Zimmermann, 2002), where the conductance of a capillary tube is defined by its length (l) and lumen diameter (d _lum), with η being the dynamics viscosity of water (0.001 Pa S at 20°C). K _H of the sectors was then divided by the area of the sector box to obtain the specific theoretical hydraulic conductivity (K _s; m² s^-1 MPa^-1), which is independent from the size of the sector. Third, wood density (ρ in g cm^-3) of each sector was determined as detailed in Peters et al. (2020a) where the optical density, of the total non-lumen area divided by sector box area, was calculated to obtain ρ by assuming a fixed density of wall material of 1.504 g cm^-3 (Kellogg and Wangaard, 1969). Specifically, for the thin-section images, we generated two radial profiles of these QWA parameters, 1) isolating vessels only, and 2) including vessels, fibres and estimating the number of undetected fibres according to Peters et al. (2020a). The latter was applied to get an estimation of the absolute values of the parameters when being able to detect all non-vessel structures (excluding cracks), while accounting for cracks within the sample and assuming a fixed cell wall thickness of 4 μm to determine the total area covered by cells against the sector box area (see https://deep-tools.netlify.app/2020/11/24/sector-intro/; see Supplementary Figure S2 as an example). See Peters et al. (2020a) for a more detailed description on the sectorial approach.

All data analyses were performed using the R software (version 4.0.2, R Core Team, 2017), including the inter-annual uncertainty analysis and the sectorial approach. For this analysis we chose the wood anatomical thin section as a reference, due to its high resolution, and time spent annotating in ROXAS. For the wood anatomical thin section (or WA) we calculated all parameters once with and once without the cut-off, whereas for the other two methods we calculated the parameters considering the cut-off value. The other methods were only considered using the cut-off value for the cell-specific lumen area. Linear regression analyses were performed between the annual time series of MLA, VD, VN, D _h, and RCA, quantifying the goodness of fit (R ²), providing information on the matching inter-annual variability, the linear regression slope between the image acquisition methods, providing information on the slope, and the relative offset to the 1:1 line (in %), to show the over- or underestimation of the method compared to the reference. Significant relationships (P < 0.05) were reported where relevant.

We moved beyond inter-annual analysis by assessing the within tree-ring patterns of wood anatomical parameters which are relevant for ecophysiology, more specifically K _s and ρ. We calculated the mean time-series of these parameters across sectors for each year and aggregated them to describe the general radial profile per species and method. To describe the general behaviour of K _s and ρ along the tree-ring across years, we used a generalized additive mixed model (gamm). The gamm model was fitted using the “mgcv” package (Wood, 2017), while considering method as a fixed effect and the year as a random intercept. For each fit we calculated the standard error and subsequently the upper and lower confidence interval. Significant differences (P < 0.05) were reported in comparison with the wood anatomical thin section which included the cut-off threshold.

We found strong similarities between ring-width measurements obtained from different image acquisition methods (P < 0.001; Supplementary Figure S4 ), proving that the image calibration was appropriate and the differences between methods should be mainly caused by variability in the cell size, image resolution, and subsequent detection. When assessing the frequency distribution of cell-specific lumen area across all tree rings, the general shape of the relationship is similar across methods ( Figure 3 ). C. betulus shows a consistent skewness towards lower vessel lumen area, while Q. robur shows a bimodal distribution with smaller latewood vessels and large earlywood vessels. These findings highlight that across years the general distribution of cell lumen area is well captured by all methods when applying a consistent cut-off value (see Supplementary Figure S5 for results without a cut-off). When analysing the detection of specific cell sizes, it was clear that the micro-CT and surface picture techniques (in their current setting) are not capable of detecting most fibres, in contrast to wood anatomical thin sections ( Supplementary Figure S5 ).

We now focussed on the inter-annual tree-ring variability, with a special emphasis on the hydraulic parameters (i.e., D _h). The inter-annual variability in D _h was well captured, when comparing the raw wood anatomical thin section (including fibres and vessels) to the other methods (treated with the cut-off; Figure 4 ). Here we used all wood anatomical elements (both fibres and vessels) as a reference to highlight the robustness of the parameter, however in analysing the other parameters we used the WA parameters calculated when using a cut-off. All species considered, we found no significant effect of the method on the linear relationship (C. betulus, P > 0.45; A. glutinosa, P > 0.20; F. sylvatica, P > 0.10, and Q. robur P > 0.33). The highly significant slopes (P < 0.0001) between the linear relationships ranged from 0.7 to 1.1, with the weakest relationships apparent for Q. robur (average slope of 0.8 for all methods), likely caused the relatively low amount of detect vessels and its variance between methods. The average goodness of fit (R ²) across species was 0.83, with an average offset for the residuals to the linear model line of around -1.1 μm.

For the five most commonly analysed wood anatomical parameters ( Figure 1 ), we compared all methods to the wood anatomical thin-section to which we applied a cut-off ( Figure 5 ). For D _h, it is clear that both the R ² and slope are high when comparing the micro-CT and the surface pictures to the wood anatomy on which a cut-off was applied (on average 0.73 and 0.86, respectively). Especially, the deviation from the 1:1 line was exceptionally small for this parameter (ranging from -2.5% to 2.9%). For this parameter the inter-annual variability was best described when comparing the raw thin section (including vessels and fibres) with the cut-off thin section (mean across species; R ² = 0.99, slope = 0.96, residual offset = -0.91). Again, the weakest common pattern in inter-annual D _h was found for Q. robur. For the diffuse-porous species, the most commonly used MLA also appeared to be well captured by all methods, even when including fibres ( Figure 5 ; mean R ² = 0.84, slope = 0.92, residual offset = -160 μm², or -1%). For these species it was interesting to see that even when including fibres, which caused a mean residual offset of -342 μm², the inter-annual variability was well maintained (mean R ² = 0.90). Q. robur showed more discrepancy between the methods, where there was poorly explained inter-annual relationship of MLA when considering fibres within the thin section (R² < 0.01). When considering the micro-CT and surface picture this inter-annual commonality with the thin section improved slightly (mean R ² = 0.48, slope = 0.73), although it was still the lowest value among the species (see Figure 5 ).

Wood anatomical parameters related to the frequency of detected vessels were more variable in their performance, compared to cell-size related parameters such as MLA or D _h. Particularly the VD showed a low R ₂ of 0.58 for C. betulus with a highly variable slope range across species (min = 0.64, max = 0.93; Figure 5 ). The VN parameter (which was calculated based on VD and the ring area) showed better inter-annual comparability for C. betulus and A. glutinosa, and F. sylvatica, but less for Q. robur ( Figure 5 ). The RCA is also impacted by the image acquisition method, with a reduced common inter-annual variability when considering the surface picture for A. glutinosa (R ² = 0.40), which underestimated the RCA by 14%. However, Q. robur showed a high inter-annual agreement for both VD, VN, and RCA, although when including the fibres particularly the inter-annual variability be more consistent for VD and VN compared to the other species.

For each year the radial pattern matched well among the different species (i.e., Figure 6 for F. sylvatica). A _l and K _s showed highly similar patterns and as such we only considered K _s for further analyses. When considering the general radial profile across years, it is clear that all methods showed highly similar K _s trends ( Figure 7 ). Slight differences can be found for the surface picture when considering A. glutinosa and F. sylvatica where K _s is slightly underestimated at the beginning of the tree ring (P < 0.001). This is however not the case for Q. robur (P = 0.955), where only K _s is significantly reduced in the first few sectors when the fibres are considered (P < 0.001). For the latter the variability in K _s in the latewood cannot be captured with the micro-CT or surface picture method, visible as low values close to zero ( Figure 7 ). For ρ, across all methods it is clear that the absolute density is lower when considering the fibre lumina compared to solely considering the vessels ( Figure 7 ). This effect is less apparent in a species like F. sylvatica, where we do see an increase in density in the latewood. For Q. robur the density dynamics are also different when considering fibres with a more gradual increase in ρ compared to when considering only the larger vessels.

Our inter-annual analyses of common wood anatomical parameters revealed that the mean hydraulic diameter (D _h) matched well among methods ( Figure 4 ). This is likely due to the overall common cell size distribution and the appropriate size estimation of the largest cells for this parameter across the three methods ( Figure 3 ). Both the goodness of fit (R ²) and slope of the regression between methods were high. Most notable was the strong mean prediction accuracy (the offset from the 1:1 line), which confirms that studies using either of these methods can be compared for their climate-D _h relationships (i.e., Nolin et al., 2021; Castagneri et al., 2017b). Moreover, the fact that D _h is not affected by the in- or exclusion of fibres ( Figure 4 ) displays that this parameter is robust and not strongly depending on a size cut-off commonly used in ring-specific analyses (i.e., Fonti and García-González, 2004; Akhmetzyanov et al., 2019). This is further illustrated by the accurate capturing of the within ring variability of the specific theoretical hydraulic conductivity (K _s) across methods ( Figure 7 ), paving the way for future analyses on the specific sectors and their climatic sensitivity (Castagneri et al., 2020a).

Ring-porous Q. robur showed the lowest R ², especially when comparing D _h variability among image acquisition methods ( Figure 5 ). One explanation could be the relatively lower number of vessels per analysed surface area and the fact that we do not measure exactly on the same location and extent within the wood. This highlights an important limitation with ring-porous species, as one would require the inclusion of more cells to get more robust signals. While for F. sylvatica it was found that a QWA signal is relatively stable for a 1 mm wide radial strip (Diaconu et al., 2017), for ring-porous species like Q. robur or Fraxinus excelsior this might require a larger sampling width (García-González and Fonti, 2008). This impact is also presented within the most widely used parameter for ring-porous species, the mean lumen area (MLA), indicating the need for a more detailed assessment on the sampling width impact across wood anatomical parameters. Nevertheless, the robustness of these hydraulic parameters is highly promising as they describe the efficiency of water transport though the xylem tissue of the plant (Tyree and Zimmermann, 2002) and have more recently been linked to the rehydration efficiency of tree species after drought (Peters et al., 2023).

The most common applied image acquisition method from the literature review for angiosperms is the surface picture method, followed by the wood anatomical thin section method. From the inter-annual analysis, it is clear that with surface pictures at 6000 dpi most variability is well captured, as long as the cells are not too small (see A. glutinosa in Figure 5 ; maximum lumen area < 4000 μm²) or the ring-boundaries are properly visible (i.e., not the case for C. betulus). On the intra-annual level, the surface picture method performs well in capturing the decrease in K _s across rings ( Figure 7 ). However, the wood anatomically derived density (ρ) is clearly overestimated compared to the reference (a thin section which includes vessels and fibres) due to the fact that this method is unable to detect the lumen of the fibres (as discussed by Peters et al., 2020a). This is related to the general downside of this method, which is the time-consuming preparation of the surface, using for example a core microtome (Gärtner and Nievergelt, 2010). This forces the practise of sub-selecting wood samples due to the wood quality issues which do not provide the appropriate contrast (i.e., Skiadaresis et al., 2021). Moreover, for surface pictures there is a technological resolution limit (defined by the resolution of the camera sensor and magnification of the lens), which hampers the use of higher resolution images to detect wood anatomical structures. This is specifically illustrated by the relatively high underestimation of VN and RCA for the surface picture method for a species like A. glutinosa ( Figure 5 ). This resolution issue is also important for ring-porous species, as the MLA is highly sensitive to different size cut-off values ( Supplementary Figure S1 ). Nevertheless, the equipment requirements of this method are low and sample processing is fast, which explains its common use for ring-porous species (St. George et al., 2002; Fonti et al., 2009; Klesse et al., 2021). Moreover, the method can be applied on larger surface areas, such as parts of a stem disk, alleviating some of the issues due to limiting amounts of vessels captured with a 5 mm wide core (García-González and Fonti, 2008; Fonti et al., 2009). Expanding the use of this method to (semi-)diffuse porous species like F. sylvatica would thus be recommended.

Wood anatomical thin cross-sections show great common patterns with the other methods and allow for the disentanglement of the impact of vessels and fibres on the wood anatomical parameters ( Figures 5 , 7 ). However, the creation of wood anatomical micro-sections requires a multitude of processing steps and specialized equipment (von Arx et al., 2016; Fonti et al., 2025). Moreover, not all wood material is easily processed due to either the brittle nature of material (Reinig et al., 2018) or the hardness of the wood sample. Despite these challenges, thin sections are the ideal standard to measure wood anatomical features due to the contrast enhancing staining and high optical resolution (i.e., 2.27 pixel per μm; see also Gärtner and Schweingruber, 2013). Moreover, due to the easy detection of fibre lumina at the ring boundary, ring detection was not difficult for C. betulus when using this method ( Figure 3 ). Furthermore, this method is most commonly used for gymnosperms and shrubs more generally (i.e., Buras et al., 2017; Seftigen et al., 2022) and shows great potential in extracting wood anatomically derived density ( Figure 7 ; Peters et al., 2020a). Expanding the use of this method to other angiosperms, in combination with the sectorial approach, would provide a novel milestone for exploring the use of intra-annual variability in wood anatomical parameters as climate reconstruction proxies.

The X-ray microtomography is the least applied method in QWA literature, yet shows great potential in reconstructing the inter- and intra-annual variability of wood anatomical parameters ( Figures 5 , 7 ). Besides showing great potential in accurately reconstructing wood density at high resolutions (Björklund et al., 2019), the method is also capable of correcting for tilted vessel due to the presence of 3D information. Its rarer application is due to the need for expensive equipment and specialized analysis software. An additional downside of this method is that it requires data acquisition and reconstruction times as the resolution increases (Jacquin et al., 2017), yet very low labour cost as it has a high automation potential. It is due to this reason that we selected our target resolution (0.28 pixel per μm), even though that the resolution can be increased to also include the fibre dimensions (Van den Bulcke et al., 2009; Dierickx et al., 2024). Yet, the potential use for micro-CT has been proven an effective tool in obtaining a multitude of wood anatomical characteristics (Steppe et al., 2004; Mayo et al., 2010). Micro-CT allows us to obtain wood density profiles which can be directly compared to the wood anatomical features (De Mil et al., 2016). The new workflow, as described in Lehnebach et al. (2021), could for example be a critical new technique in estimating biomass increments within the tree ring for any tree species. This method also allows for the detection of wood anatomical features in 3D, permitting the correction of changing angles of the vessels and fibres (Trtik et al., 2007; Brodersen, 2013). Finally, micro-CT can be applied to tropical hardwood species, for which normally it would be challenging to use the thin-section method, due to difficulties in cutting the sample (i.e., Van Camp et al., 2018; Tarelkin et al., 2019), allowing micro-CT to boost tree ring research (Van den Bulcke et al., 2019). The development of high-resolution X-ray computed tomography is thus recognized by the scientific community as a critical and non-destructive technical advancement in QWA (Gennaretti et al., 2022) and our results should further motivate its implementation within QWA research, especially given the continuous progress in acquisition and reconstruction speed.

Our systematic uncertainty analysis of the three most commonly applied image acquisition methods in analysing QWA time-series reveals their potential in effectively extracting wood anatomical parameters. Particularly, hydraulic parameters related to cell number and lumen dimensions appear easily extractable with the surface picture method, making it possible to process large quantities of samples in a relatively short time compared to wood anatomical thin sections. However, it is not advised to apply such a method on species with a maximum lumen area below 4000 μm², even with comparably high resolution of 6000 dpi used in this study (see Figure 2 with A. glutinosa), or on species with unclear visual ring boundaries. We also presented an effective way to assess within-tree radial profiles with all methods. Since the temporal dynamics of angiosperm wood formation are still under early exploration (Balzano et al., 2018; Noyer et al., 2023), the presented results provide a first benchmark on the accuracy of these different methods for angiosperm species and should motivate their combined application in QWA, especially when using the sectorial approach.

From our results, we estimate that it is possible to compile wood anatomical parameters collected with various methods, as long as the maximum vessel size is above 4000 μm², paving the way forward to create a generalized database which includes these measurements for large-scale spatial analyses, similar to existing tree ring databases (e.g., Babst et al., 2017; Babst et al., 2019; Klesse et al., 2023). However, a clear issue from our results is that species with larger vessels, like ring-porous species, tend to be sensitive in their wood anatomical parameters to the number of vessels considered in the calculation. This sensitivity is highly dependent upon the tangential width of the sample. A systematic assessment of this effect on the climate sensitivity would be helpful in establishing the exact width needed to obtain robust climatic signals. Also, our uncertainty assessment can be expanded by creating QWA chronologies from each of the methods to show how climatic responses are impacted, as demonstrated in the comparison by Björklund et al. (2019). This would however require higher sample replication and a higher and standardized number of tree rings per sample than in our study. With the sectorial approach presented here, it is possible to move beyond isolating solely the first row of vessels (Souto-Herrero et al., 2017) and delve into more detail on the radial pattern changes of QWA parameters and inspect their individual climate sensitivities. These investigations could be of great value as they would provide even more robust recommendations for method selection and their specific use in the QWA community (von Arx et al., 2021).