Using adjusted Blue Intensity data to attain high-quality summer temperature information: A case study from Central Scandinavia

Abstract

The inexpensive Blue Intensity proxy has been considered a complement or surrogate to maximum latewood density (MXD), but is associated with biases from differential staining between sapwood and heartwood and also between deadwood samples and living-wood samples that compromise centennial-scale information. Here, we show that, with some minor adjustments, ΔBlue Intensity (ΔBI) is comparable with MXD or ΔDensity (Δ = the difference or contrast between latewood and earlywood density) in dendroclimatological reconstructions of summer temperatures in the Central Scandinavian region, using Pinus sylvestris L. (Scots pine), on annual and multi-centennial timescales. By using ΔBI, this bias is significantly reduced, but the contrast between earlywood and latewood in BI is altered with degree of staining, while for density it is not. Darker deadwood samples have a reduced contrast compared with the lighter living-wood samples that make ΔBI and ΔDensity chronologies diverge. Here, we quantify this behaviour in BI and offer an adjustment that can reduce this bias. The adjustment can be derived on independent samples, so in future work on BI, parallel density measurements are not necessary. We apply this methodology to two Central Scandinavian Scots pine chronologies that averaged into a composite is able to reconstruct summer temperatures with an explained variance in excess of 60% in each verification period using a split sample calibration verification procedure. Although the amount of data used to derive this contrast adjustment produces desirable results, more tests are needed to confirm its performance, and we suggest that future work on the BI proxy should aim for a small subset of parallel BI and density measurements while the bulk of the data is only measured with the BI technique. This is to ensure that the adjustment is continuously updated with new data and that the conclusions derived here are robust.

Keywords

Blue Intensity blue reflectance density late Holocene Scandinavia summer temperature tree rings

Introduction

Exploration of the new proxy Blue Intensity (BI), derived from optically scanned images of tree-ring data, is important because studies have shown its potential to become a powerful temperature proxy (e.g. Björklund et al., 2014; Campbell et al., 2007, 2011; McCarroll et al., 2002, 2013), similar to maximum latewood density (MXD, Schweingruber et al., 1978) which is considered to contain the strongest summer temperature signal in high latitudes (Grudd, 2008; Linderholm et al., 2010; Melvin et al., 2013; Wilson and Luckman, 2003). While MXD data are dependent on expensive equipment, BI can be obtained at much lower costs, and the relatively low effort with which data can be produced and replicated means that the technique is much more accessible to tree-ring researchers. However, one of the main concerns regarding BI is its ability to capture centennial-scale climate information of similar quality to MXD (Björklund et al., 2014).

While radiodensitometry provides wood density, BI provides information probably related to the absorptive properties of the lignin content in wood (McCarroll et al., 2002), which in turn is coupled to the density. Extractives like resin do, however, significantly affect BI measurements and produce systematic biases on multi-centennial timescales. Björklund et al. (2014) showed that, using regional curve standardisation (RCS; Briffa et al., 1992), a method intended to preserve centennial-scale variability in tree-ring chronologies, maximum latewood blue absorption intensity (MXBI, the BI analogue of MXD) chronologies would likely provide biased summer temperature information compared with MXD.

If BI is going to be used as a complement to MXD, the problems concerning its ability to express temperature variations also on centennial timescales needs to be resolved. Presently, the best way to ascertain information on these timescales is by using density data from the same trees, under the assumption that density is an appropriate predictor of temperature on centennial timescales. The main reason for the difference between density and BI data is because of the heartwood and sapwood within a tree having different degrees of staining (Björklund et al., 2014), which is caused by an active differential allocation of extractives (Raven, 2004). Also, the longer time the extractives reside inside the wood-structure, the more permanent the staining likely becomes. Alternatively, trees that produce more resins for various reasons, for example, because of injuries, will after death be better preserved, and consequently, these trees are more likely to be sampled because less resinous trees will be more susceptible to degradation. To reduce the effect of differences in colour within a tree, samples to be used for BI analysis are refluxed with ethanol before being scanned (Campbell et al., 2011), but still some staining can remain (e.g. Björklund et al., 2014). In combination, the above-mentioned effects will likely result in MXBI chronologies that contain persistent negative trends that are unrelated to climate.

In an attempt to overcome these problems, the ΔBI parameter was introduced, which was defined as the difference between latewood and earlywood measurements within an annual increment, intended to remove the effect of differential staining within trees as well as increasing the summer (June–August, JJA) temperature signal (Björklund et al., 2014). Still, the ΔBI parameter could not fully match the long-term trend in ΔDensity, which was assumed to contain the best estimates of JJA temperatures on centennial timescales. The mismatch is likely an effect of the contrast between earlywood and latewood BI measurements systematically differing from that in density.

In this study, we will examine how the contrast between earlywood and latewood BI changes as a function of degree of sample staining. Based on our findings, we propose a method to adjust the contrast of earlywood and latewood BI values for individual trees, so that it better reflects the earlywood/latewood contrast in density measurements. We show that contrast-adjusted ΔBI (ΔBI_adj) better matches the persistence in ΔDensity, and that by adjusting for the variable contrast in earlywood and latewood among samples, ΔBI_adj can be considered an acceptable complement to ΔDensity also on centennial timescales. Moreover, we provide a ‘manual’ on how to adjust BI measurements when corresponding density data are unavailable. Finally, we present the first, to our knowledge, JJA temperature reconstruction based entirely on BI data, for central Sweden. This reconstruction, derived from ΔBI_adj, is compared with those based on ΔDensity and MXD data from the same samples, as well as independent reconstructions of Northern Fennoscandian or Northern European summer temperatures (Esper et al., 2014; McCarroll et al., 2013).

Data and methods

Study area

We utilised Scots pine tree-ring data from two previously studied sites in central and northern Sweden: Arjeplog (66.3°N, 18.2°E, 550–700 m.a.s.l.) and Jämtland (63.1°N, 13.4°E, 600–650 m.a.s.l.). From both sites, MXD data have previously been explored (Björklund et al., 2013, 2014; Gunnarson et al., 2011). In this study, we used an updated Jämtland chronology, where by increasing the sample replication, the chronology now reaches back to 1300 CE based only on local density data analysed with the same instrumentation, rather than being a composite of two different datasets (see Gunnarson et al., 2011 for more details). Both Arjeplog and Jämtland show expressed population signals (EPS; Wigley et al., 1984) > 0.85 over the full 1300–2008 period, with the exception for Arjeplog in a few decades around 1600 CE, indicating adequate reliability (Björklund et al., 2014). Both study sites are located east of the main divide of the Scandinavian Mountains, close to the local altitudinal tree line. The climates of respective area are cool and temperate, with mean monthly temperatures ranging from −14°C in January to 13°C in July in Arjeplog and −9°C and 12°C in Åre-Björnänge (the closest meteorological station). The mean annual precipitation is 553 and 762 mm in Arjeplog and Åre-Björnänge, respectively (1961–1990 mean; Alexandersson et al., 1991).

Tree-ring data

Density and BI data were extracted from the same individual tree cores for both sites. A total of 143 trees were analysed from Arjeplog, and 79 trees from Jämtland (one core/tree). The minimum sample number through time in both chronologies was 9 trees/yr during 1300–2008 CE. The samples were prepared and dated according to standard dendrochronological methods (Stokes and Smiley, 1968). The density data were produced using an ITRAX multiscanner from Cox Analytical Systems (http://www.coxsys.se), described by Gunnarson et al. (2011). The optical BI data were produced following the standard protocol developed by Campbell et al. (2011), with some modifications according to the suggestions by Björklund et al. (2014). The BI parameters used in this study were MXBI and earlywood blue absorption intensity (EWBI), which are subsequently used to construct ΔBlue Intensity (ΔBI), which is EWBI subtracted from MXBI (Figure 1a). Also a new parameter – contrast-adjusted ΔBI (ΔBI_adj), which is described below, was obtained. For more detailed descriptions of MXBI and ΔBI, see Björklund et al. (2014). As reference chronologies for the BI data and for later reconstruction purposes, MXD and ΔDensity (Figure 1a) were also used. The Jämtland and Arjeplog data were initially analysed separately, but after contrast adjustment, they were merged into composite chronologies after scaling their means and standard deviations. Three composite BI chronologies were constructed from the different parameters (MXBI, ΔBI and ΔBI_adj), as well as two composite density chronologies (MXD and ΔDensity). To preserve as much of the centennial-scale variability as possible, RCS was used to remove age-related trends in the tree-ring data: a regional curve (representing the average pith-year growth evolution at each site) was smoothed with a cubic smoothing spline with a 50% cut-off at 100 years and subsequently subtracted from each data series prior to averaging data into chronologies. The Arjeplog and Jämtland chronologies had separate regional curves, and pith offsets were utilised. The standardisation was performed with the ARSTAN software (Cook and Krusic, 2005). For reference, an alternative standardisation procedure, based on the RCS principle in combination with the signal-free approach to standardisation (Melvin and Briffa, 2008), introduced in Björklund et al. (2013) was employed and parallel results are presented in supplementary material (available online).

Figure 1.

(a) Cartoon with the tree-ring parameters used in the study. (b) ΔBI versus ΔDensity from Arjeplog that shows the mismatch in trend (Björklund et al., 2014). (c) ΔBI and ΔDensity versus EWBI and EWD, respectively. It shows that ΔBI is dependent to a small degree on the EWBI value, while this is not the case for density. (d) The difference in contrast between earlywood and latewood BI changes with degree of staining. The δ indicates the change in BI with degree of staining, δ₁ indicates the change in EWBI and δ₂ indicates the change in MXBI. The δ₁ is different from δ₂ as the degree of staining changes; however, the cartoon illustrates that it is proportional. (e) The density profile for the same samples in (d) where all the profiles are ideally the same.

Proxy development

The ΔBI and ΔDensity chronologies from both sites display a mismatch in trend (Figure 1b, illustrated by Arjeplog data). To explore how the contrast in earlywood and latewood behaves in BI measurements relative to density measurements, EWBI was regressed against ΔBI, and earlywood density (EWD) was regressed against ΔDensity (Figure 1c). The analysis was made for each tree ring in both Jämtland and Arjeplog datasets amounting to 39,875 parallel observations. Figure 1c shows that there is a marked difference in how the contrast between earlywood and latewood behaves in BI relative to density. From the analysis, it is clear that the contrast between earlywood and latewood in BI measurements decreases as the earlywood darkens (r = −0.36, p ≤ 0.01). This phenomenon is schematically depicted in Figure 1d. This is, however, not the case for density (Figure 1c and e), where the density profile remains largely unchanged regardless of the degree of staining (r = 0.017, p ≤ 0.01).

There are large differences in trends when the MXBI and MXD chronologies are compared (Figure 2a and b). By introducing the Δ parameter, the trend difference is reduced (Figure 2c and d, note that ΔBI has a more positive trend compared with ΔDensity in both Arjeplog and Jämtland). The more positive trend in the ΔBI chronology compared with ΔDensity likely arises from the fact that older samples are generally darker than living tree samples. More distant time periods are thus associated with lower earlywood/latewood contrasts and modern periods with higher contrasts, leading to a positive trend in ΔBI.

Figure 2.

(a) and (b) Comparison between MXD and MXBI chronologies from Jämtland and Arjeplog, (c) and (d) ΔDensity and ΔBI, and (e) and (f) ΔDensity and ΔBI_adj. See Figure S1 (available online) supplementary information for results with an alternative standardisation procedure.

By pooling MXBI and EWBI for each sample into P_BI (i.e. for each tree, MXBI and EWBI measurements are combined into one dataset) and by regressing it against a corresponding pooling of MXD and EWD (P_Density; Figure 3a), it is possible to estimate the general earlywood–latewood contrast in BI compared with the contrast in density for every sample/tree.

Figure 3.

(a) The relationship between the pooled EWD and MXD (P_Density), and pooled EWBI and MXBI (P_BI) for one sample from the Arjeplog data. (b) R² values for all the regressions between P_Density and P_BI samples from Arjeplog and Jämtland. (c) The relationship between EWBI sample mean and slope (α), and intercept (β) from P_Density versus P_BI. (d) The effect of adjusting the slope and intercept according to sample-mean EWBI for some common intercepts in the datasets. (e) Equations needed to contrast adjust BI data when corresponding density data does not exist.

Examining all regressions of P_BI versus P_Density (Figure 3b), it is clear that there is a strong relationship between density and BI (on average R² = 0.9), note that all pairs of P_BI versus P_Density are regressed against each other and included in the average. It is also evident that the slope or contrast systematically changes when samples are dark or light. We chose the mean earlywood BI ( $\bar{EWBI}$ ) as a representative for the degree of staining in each sample. Regressing ( $\bar{EWBI}$ ) against the slope and intercept of the previous regression between P_BI and P_Density, it is suggested that ( $\bar{EWBI}$ ) can be used as a predictor for the change in contrast that occurs in BI. Applying the outcome of the second regression, those slopes and intercepts (Figure 3c), to a set of common $\bar{EWBI}$ from Arjeplog and Jämtland yields the functions shown in Figure 3d. Consequently, for a dark P_BI sample to match the contrast in the corresponding P_Density sample, the P_BI sample contrast need to be increased and is adjusted with a steep slope. The brighter P_BI samples are thus adjusted with a gentler slope so that dark and light samples have a similar contrast regardless of the original staining.

The ( $\bar{EWBI}$ ) is a conservative predictor for this adjustment since samples can have different staining in different places. The most obvious example is the difference in BI between heartwood and sapwood. These two components in the wood need to be separated for this adjustment to work. There are also other less obvious differences in BI nuances across samples, but they were disregarded to try to make the adjustment as simple as possible without loosing the objective, which was to neutralise the contrast bias in BI as far as possible without making the method too complex.

To verify the contrast-adjustment approach, the Arjeplog samples were contrast adjusted using the equation derived from the Jämtland data (green circles Figure 3c), and then the process was reversed so that the Jämtland samples were contrast adjusted using the equation derived from the Arjeplog data (yellow circles Figure 3c). The contrast-adjusted chronologies are presented in Figure 2e and f.

A short guide to contrast adjusting BI data

To help other researchers to apply this technique, we here provide a step-by-step guide to the process of deriving contrast-adjusted BI data:

Because of the pronounced difference in staining between heartwood and sapwood, the two components need to be separated for every sample. Heartwood is darker than sapwood and needs a steeper slope for the adjustment. This division can be qualified because heartwood and sapwood are clearly distinguishable components of a Scots pine cross section and the difference in appearance is unrelated to climate variations.

P_BI is then regressed against P_Density to derive a set of slopes and intercepts, α and β, respectively (see Figure 3a for an example; P denotes that the EWBI and MXBI have been pooled for each tree, the same is also true for EWD and MXD).

For each sample, the sample mean of EWBI ( $\bar{EWBI}$ ) is then used as a predictor for the contrast change required for every P_BI. Note that $\bar{EWBI}$ has a reasonable skill in predicting the slope and intercept of α and β. The relationship between slope and intercept is very strong (R² = 0.85; results not illustrated).

Inserting the predictions of α and β into Eq. 3, a contrast-adjusted BI (BI_adj) can be calculated

α = 0.048 \times \bar{EWBI} + 3.0

(1)

β = - 12.6 \times \bar{EWBI} + 838

(2)

P_{BI adjusted} = α \times P_{BI} + β

(3)

fter contrast adjustment of the BI data, ΔBI is calculated.

{ΔBI}_{adj} = {MXBI}_{adj} - {EWBI}_{adj}

(4)

The requirements to contrast adjust BI samples when parallel density measurements are unavailable are to follow steps 1, 4 and 5. However, before these steps can be taken, the BI measurements must be made according to the modified protocol in Björklund et al. (2014). Although the protocol has been developed using only measurements from the software WinDendro (Regent Instruments Canada Inc., 2009), it should be possible to use other software such as CDendro/CooRecorder (Cybis Elektronik, 2010) to derive the measurements as long as the scanner is calibrated prior to use with a colour target IT8.7/2 and inverted to be positively correlated with radiodensitometric measurements, values range between 0 and 255.

Furthermore, if the adjusted BI data is going to be used together with density data, it is important to note that the BI data will be biased towards the mean of the density data (have lower standard deviations). This is because the simple linear regression used here assumes that all the errors are in the BI data and not in the density data and regressing BI against density, the standard deviation of BI data can only be equal or lower than for density (inherent from the methodology). The lower standard deviation can, however, be scaled to match the density data standard deviation.

Reconstructing summer temperatures

The ΔBI_adj chronologies from the two sites were normalised (z-scored) and then averaged together into a composite chronology covering the period 1300–2008 CE. The same was done for MXD and ΔDensity. Because of the inherent problems described above, we made no reconstructions using the MXBI and ΔBI data. As ΔDensity from both Arjeplog and Jämtland have strong JJA temperature signals (Björklund et al., 2014), we chose this season as the target for the reconstructions.

The temperature record selected for this analysis was obtained from the global land surface 5° × 5° resolution CRUTEM.4.2.0.0 dataset covering the period 1850–2008 (Jones et al., 2012), averaged over the 10–20°E and 60–65°N region. To extend the temperature history back in time, a transfer model was developed using simple linear regression, where the JJA temperature anomalies (deviations from the 1961–1990 mean) were set as the predictand and the tree-ring data of the current year (t) as the predictor. The time stability and quality of the model were tested through a split sample calibration/verification procedure (Gordon, 1982), where the period of tree-ring and climate overlap (1850–2008) was divided into two periods of roughly equal length (1850–1928 and 1929–2008, respectively). Calibration and verification statistics were then calculated for the first and second half of the period, respectively. The calibration and verification periods were then exchanged and the process repeated. The validation was performed using the explained variance (R²), Reduction of Error (RE), Coefficient of Efficiency (CE) and Mean squared Error (MSE) statistics (National Research Council, 2006). The final models were calibrated over the full 1850–2008 period. The obtained regression coefficients were used to calibrate the chronologies. The ΔBI_adj reconstruction was then compared with instrumental Central Scandinavian temperatures, and its MXD and ΔDensity counterparts. Furthermore, comparisons with independent reference data were made and included a Northern Fennoscandian summer temperature reconstruction based on an ensemble of various tree-ring parameters (NFT; McCarroll et al., 2013) and one Northern European summer temperature reconstruction based on a composite of two state of the art MXD records from Northern Fennoscandia (Esper et al., 2014).

Analysis of coherency between the BI and the density data, as well as between the tree-ring proxies and observed temperatures, was made using the software Anclim (Štěpánek, 2008). Coherence analysis, that can be thought of as a squared correlation coefficient that depends upon frequency (Von Storch and Zwiers, 2004), was made to reveal the frequency association between different tree-ring chronologies and between tree-ring chronologies and instrumental data. The coherency between BI and density data was restricted to periods <300 years because periods longer than 30% of the data length should not be analysed, and when comparing tree-ring data with instrumental observations, periods longer than 30 years were disregarded. The spatial climate signal of each chronology was assessed through point-by-point correlations with gridded 0.5° × 0.5° resolution land surface JJA temperatures from the CRU TS 3.10 dataset (Harris et al., 2014) over the interval 1901–2008.

Results

The mismatch in long-term trend between Jämtland and, especially, Arjeplog MXBI and MXD (Figure 2a and b) is also highlighted in the coherence analysis where MXBI gradually loses coherency with MXD at lower frequencies (Figure 4a–c). The ΔBI chronologies match their ΔDensity counterparts better (Figure 2c and d), especially the Jämtland chronology, although both ΔBI chronologies display more positive trends than the ΔDensity ones. According to the coherence analysis, the ΔBI and ΔDensity chronologies are more coherent than the MXBI and MXD in most frequencies, especially on centennial timescales (Figure 4a–c). The ΔBI_adj chronologies are even more similar to ΔDensity (Figure 2e and f), which is also indicated by the coherency analysis (Figure 4a–c). Moreover, this good agreement between the ΔBI_adj and ΔDensity chronologies for both Arjeplog and Jämtland confirms the usefulness of the above described adjustment technique (section ‘Proxy development’). The implication of this is that; provided that the adjustment equation is valid for Scots pine in a wider region, it will become possible to contrast adjust BI data from additional sites without the need for density data.

Figure 4.

(a–c) Coherency between MXD and MXBI (darkblue), between and ΔBI (lightblue) and ΔDensity and ΔBI_adj (grey). Number of observations is 708 and significance levels refer to periods of less than 300 years; above that periodicity, no coherence coefficient is calculated. (d–f) Coherency between the Central Scandinavian composite tree-ring data (MXD, ΔDensity and ΔBI_adj) and summer temperatures (JJA) from Central Scandinavia. Number of observations is 158 and significance levels refer to periods of less than 30 years; above that periodicity, no coherence coefficient is calculated. Frequencies equal to 0.1 means 10 years, 0.01 means 100 years and so on. See Figure S2 (available online) supplementary information for results with an alternative standardisation procedure.

The composite ΔBI_adj reconstruction of Central Scandinavian JJA temperatures (hereafter CST) is verified in both independent periods in the split calibration/verification procedure (Table 1). The ΔDensity reconstruction has slightly weaker statistics but verify above or around 0.5 for R², RE and CE also in both periods (Table 1). The MXD has yet again slightly lower statistics than ΔDensity but passes all the statistical evaluation tests (Table 1). Comparing the coefficients of the highest correlation (ΔBI_adj vs JJA temp.) and the lowest correlation (MXD vs JJA temp.), it is not possible to separate them with a Fischer’s z-test (p = 0.052). Comparing the correlations using the first differenced data, the ΔBI_adj has a significantly higher correlation than MXD (p = 0.013). The final calibration linear models (Eq. 5–7, in the form CST = α × TRD + β) were built over the 1850–2008 period, where CST is the Central Scandinavian summer temperatures, α is the slope, TRD is the tree-ring data (predictor) and β is the intercept (complementary tests of the model residuals confirm that the modelled results satisfy the standard regression assumptions of normality, variance and autocorrelation).

CST = 0.77 \times MXD + (0.039)

(5)

CST = 0.822 \times ΔDensity + (- 0.223)

(6)

CST = 0.785 \times {ΔBI}_{adj} + (- 0.220)

(7)

Table 1.

Split sample calibration/verification statistics for Central Scandinavian summer temperature reconstruction (CST) based on the MXD data, ΔDensity data and the ΔBI_adj data. The early period covers 1850–1928 and the late period covers 1929–2008. The full period explained variance is in bold-faced font. See Table S1 (available online) supplementary information for results with an alternative standardisation procedure.

Reconstruction version	Calibration/verification	Explained variance (R²)	Reduction of error (RE)	Coefficient of efficiency (CE)	Mean squared error (MSE)	Slope	Intercept	F-statistic model
MXD	Late/early	0.49	0.62	0.48	0.52
	Early/late	0.44	0.60	0.44	0.53
	Full	0.51/0.56^a				0.77	0.039	2.48E–26
ΔDensity	Late/early	0.55	0.66	0.53	0.47
	Early/late	0.47	0.61	0.46	0.51
	Full	0.55/0.65^a				0.822	−0.223	4.93E–29
ΔBIadj	Late/early	0.64	0.65	0.52	0.48
	Early/late	0.62	0.69	0.57	0.41
	Full	0.65/0.72^a				0.785	−0.220	1.68E–37

Explained variance from first differenced data.

The fact that the higher frequencies in the target data are reproduced with more skill than the lower ones in the tree-ring reconstructions is clearly shown in the coherency analysis (Figure 4d–f). The results also suggest that the ΔBI_adj data have a slightly higher performance as a summer temperature predictor across most frequencies than both MXD and ΔDensity. Similarly, ΔDensity has slightly higher performance than MXD in most frequencies (Figure 4d–f).

The ΔBI_adj composite chronology is almost identical to the ΔDensity composite (r = 0.95 unfiltered, r = 0.96 first differenced, n = 708), seen also in the coherence analysis (Figure 4c). Comparing the ΔBI_adj composite chronology against the McCarroll et al. (2013) NFT and the Esper et al. (2014) N-Eur reconstructions shows distinct similarities (Figure 5c; rNFT = 0.64 and rN-Eur = 0.69 unfiltered, and rNFT = 0.64 and rN-Eur = 0.73 first differenced, n = 705).

Figure 5.

(a and b) Central Scandinavian summer temperatures (JJA) reconstructed with ΔBI_adj (CST ΔBI_adj) against observed temperatures (CST_Obs.). (c) CST MXD, CST ΔDensity and CST ΔBI_adj as well as reconstructed Northern Fennoscandian multi-proxy summer temperature reconstruction (NFT; McCarroll et al., 2013) as well as a Northern European summer temperature reconstruction (N-Eur; Esper et al., 2014). Bold black lines are the reconstructions smoothed with a cubic smoothing spline with a 50% cut-off at 25 years. Instrumental data are an average of the target temperature data CRUTEM.4.2.0.0. (Jones et al., 2012), where the grid points in the area Longitude 10–20° and Latitude 60–65° were used (Central Scandinavia). The common overlap between the chronologies and observational data was 1850–2008. See Figure S3 (available online) supplementary information for results with an alternative standardisation procedure.

The reconstructions suggest that cold summers occurred in the 15th and 17th centuries and at the end of the 19th century. Warm summers are suggested in the 16th and 20th centuries. The most pronounced differences among the three records are seen in the 18th to early 19th century, where the variability is high in NFT and low in the N-Eur with CST in between the two other records. Overall, all three records are very similar. However, comparing the ΔBI_adj CST reconstruction against instrumental observations reveals that the warmest years on record, 1997, 2002 and 2006, are underestimated.

Looking at the spatial representations of the reconstructions as compared with the observed JJA temperatures over the period 1901–2005, the MXD reconstruction does not reach r = 0.71 (corresponding to an explained variance of the observed JJA temperature of 50%) in the Fennoscandian domain, while the ΔDensity reconstruction displays >0.71 correlations in patches across the central parts of the domain mainly adjacent to the Baltic Sea (Figure 6). The spatial field correlations of the ΔBI_adj CST are above 0.71 for most of Fennoscandia, except eastern Finland and parts of southern Norway and Sweden, peaking in the central parts of the Scandinavian Mountains (Figure 6), complementing the correlation field of NFT and N-Eur that both have more northerly confined correlation field with r > 0.71 (Figure 6).

Figure 6.

Field-correlation between CRU TS 3.1 (Harris et al., 2014) and the Central Scandinavian summer temperature reconstructions (CST) derived from MXD, ΔDensity, ΔBI_adj, the Northern Fennoscandian multi-proxy summer temperature reconstruction (NFT; McCarroll et al., 2013) and the Northern European summer temperature reconstruction (N-Eur; Esper et al., 2014). Period of correlation overlap is 1901–2005 and target season is JJA. Grey areas outside the (blue band) r = 0.71 isoline are correlations significant at (p = 0.001). Correlation fields were produced using KNMI Climate explorer (http://climexp.knmi.nl; van Oldenborgh et al., 2009).

Discussion

One of the major issues when developing BI dendroclimatology has been the inability of MXBI to provide information on long (centennial) timescales as reliable as MXD (Björklund et al., 2014). Here, we show that by using the contrast adjustment and combining data (Arjeplog and Jämtland), ΔBI_adj can match ΔDensity in multi-centennial-scale temperature reconstructions, and possibly surpass them at annual scale. For this dataset, ΔBI_adj is actually a significantly better predictor of JJA temperatures than MXD on the annual scale, but if all frequencies are included, this separation cannot be done. Since BI has been viewed as a cost-efficient complement to density analyses, this is a significant advance.

The problem of biased centennial times scale information in BI was first addressed in Björklund et al. (2014) when they introduced ΔBI. The new parameter showed much improvement in the ability to express the low-frequency variability found in MXD. However, as the ΔBI chronologies from Jämtland and Arjeplog still exhibited differences in long-term trends compared with their corresponding ΔDensity chronologies (where we assumed the trend in the density data to be more correct), it was clear that additional adjustments were needed. The positive trend offset indicated that it was related to a contrast offset among trees in the BI data, where darker samples exhibited lower contrast between earlywood and latewood than lighter samples. The fact that the ΔBI reconstruction has slightly better statistics than the ΔDensity one does not change this assumption; it would likewise be strange to use MXBI instead of MXD even if MXBI has slightly better statistics than MXD (results not presented), showing that a high correlation during the calibration period does not guarantee high performance throughout a record. Introducing a contrast-adjusted ΔBI_adj, where the differential contrast between the BI and density methods was dealt with, provided a tool to address this bias. The two ΔBI_adj chronologies have only minor offset in trend compared with the ΔDensity ones (Figure 2e and f). The still existing slight trend mismatch between ΔDensity and ΔBI_adj was of different signs in the two datasets, suggesting that the bias from a differential contrast is now largely eliminated. Trying a different detrending method, that also can preserve low-frequency variability, only marginally changed results (Figures S1–S3 and Table S1 (available online)). However, small errors still exist, but if they are random, combining several records can match density since the data-production is so much more inexpensive.

The Jämtland dataset has, overall, a smaller spread in the degree of staining among samples, and also a lower number of samples (about half of Arjeplog). This contributes to making the Jämtland contrast adjustment less robust when using it on the Arjeplog material. Merging the Arjeplog and Jämtland datasets to produce a global contrast adjustment improves the match with the ΔDensity chronologies slightly more, but this was not done since it would have compromised the tests’ independence. Adjusting the Arjeplog and Jämtland chronologies with the dependent adjustment (contrast adjustment derived on the same material that is adjusted) improves the match yet again, but the mismatch is still of opposite sign in the different chronologies and is further suggesting that the error now is random.

Had it not been for the Arjeplog ΔBI chronology having such a large spread in degree of staining, the contrast issue might have gone undetected. The discovery of the staining problem, and this potential solution, now makes it possible to create high-quality BI chronologies without needing density data for reference. However, as a note of caution, other yet undiscovered biases may still be unresolved, which highlights the need for further tests where density and BI are measured in parallel. For example, only a small amount of data in this study was derived from lake material. Lake material can potentially be used to extend chronologies substantially back in time. However, potential sources of uncertainty with this type of data could be that the contrast between earlywood and latewood may behave differently from dry deadwood because these logs have under a long time been ‘extracted’ for resin in the water and their darker appearance may instead arise from rapid oxidation when stored in a lab environment.

Nevertheless, in future work on Scots pine in Fennoscandia, the combined relationship derived from both Jämtland and Arjeplog can likely be used to contrast adjust BI data from other sites. Further studies are of course needed to more firmly establish this, also including lake material but also for other species and for other areas. As an example, BI from Norway spruce (Picea abies L. Karst) may be a less complicated species to work with since Norway spruce does not have the pronounced heartwood sapwood division, and because resin and staining generally are more limited. Pooling parallel measurements of density and BI data from several sites (e.g. along the Scandinavian Mountains) would likely also provide a better and more robust estimate for the contrast adjustment.

We presented the first attempt to perform and evaluate the novel contrast-adjustment technique, using data from Jämtland and Arjeplog independently. A significant improvement was achieved compared with the unadjusted data when records were combined into a composite reconstruction. We recommend that future work on this new proxy focuses on bulk BI measurements, where density analysis on a few parallel samples are made to evaluate how the contrast offset of new samples relate to the contrast offset in previous measurements. These parallel density and BI measurements should ideally encompass the entire spectra of degrees of staining in new datasets to achieve increased leverage when mean earlywood BI is regressed against slopes and intercepts from the regression between P_BI and P_Density. By adding sites with a small number of parallel density measurements to the pool of BI data that was started here, the robustness of the contrast adjustment may be continuously improved. Thus, we propose that in addition to BI (both earlywood and latewood) data, complementary density data should be uploaded to the International Tree-Ring Data Bank (ITRDB; http://www.ncdc.noaa.gov/data-access/paleoclimatology-data/datasets/tree-ring) to make the possibility to adjust the BI data, using the above-mentioned protocol, freely available to researchers.

The Central Scandinavian summer temperature reconstruction from ΔBI_adj is almost identical to the one based on ΔDensity, again showing the merit of the adjusted BI data and indicates that BI likely can be viewed as a complement to density data. The good agreement between the ΔBI_adj CST and the NFT and N-Eur indicates a temporally coherent summer temperature pattern over Fennoscandia. The weakness in the reconstruction – underestimating a few record warm years around the 2000 CE – is a delicate issue. Since these features also can be observed in larger datasets where more advanced detrending options have been employed (Linderholm et al., 2014) and also in the supplementary material (available online) where alternative detrending options was used, it may be related to that summer temperature as target being too simple to accurately describe the complex relationship between tree growth and climate. Further studies are needed to evaluate this. From the spatial analyses, it was clear that the three reconstructions (CST, NFT and N-Eur) actually represent different regions, where the NFT and N-Eur represented the northernmost part of Fennoscandia, while the CST represents a more southerly oriented part of the region. So if data from the two regions is merged together, it will likely provide a more coherent view of all-Fennoscandian summer temperature history. However, the different spatial patterns of the CST and NFT/N-Eur suggest that there are slightly different sub-regional temperature characteristics in this otherwise quite homogenous region. Possibly more information on sub-regional scales can be gained if datasets are kept separately to reconstruct more sub-regional summer temperatures, as opposed to pooling them together in grand means to provide reconstructions of regional scale as in McCarroll et al. (2013). The bimodal appearance in the NFT field correlation perhaps further supports this argument.

The aim of this study was to address the lack of plausible centennial-scale variability in BI data. We have shown one possible method to deal with this problem, as indicated by the good agreement in long-term variability of the BI reconstruction and previous reconstructions. Therefore, a more in-depth analysis of BI-derived climate variability is put on hold until more and longer time-series are available. The climate reconstruction presented here was mostly made as a proof of concept. We anticipate that more extensive studies using existing density and BI records, exploiting adjusted ΔBI and ΔDensity, will provide new insights and more details into the climate history of the region than can be achieved by only using MXD and TRW.

Conclusion

By exploiting the systematic difference in contrast between earlywood and latewood in Scots pine from two sites in the Central Scandinavian Mountains, we have shown that RCS-based contrast-adjusted ΔBI chronologies have potential as excellent predictors of past summer temperatures in Fennoscandia. When combined into a regional composite chronology, they show the similar level of robustness and performance as the existing MXD records.

The new approach can likely be applied to Scots pine BI measurements from other sites in Fennoscandia, even when parallel density measurements are not available. However, we recommend that small subsets of samples are measured with regard to density, to contribute to a data pool used to further improve the contrast adjustment.

Footnotes

Acknowledgements

We thank one anonymous referee and Danny McCarroll for valuable comments on the manuscript. We thank Kenth Björk (Skogsstyrelsen Administrator Arjeplogs common land) for valuable advice and permission to collect the Arjeplog tree-ring chronology.

Funding

This work was supported by Vetenskapsrådet (grants to Hans W Linderholm). The paper contributes to the Swedish strategic research areas Modelling the Regional and Global Earth system (MERGE), and Biodiversity and Ecosystem services in a Changing Climate (BECC). This is contribution #29 from the Sino–Swedish Centre for Tree Ring Research (SISTRR).

References

Alexandersson

Karlström

Larsson-McCann

(1991) Temperature and precipitation in Sweden 1961–90: Reference normals. SMHI Reports Meteorology and Climatology 81: 1–88.

Björklund

Gunnarson

Krusic

. (2013) Advances towards improved low-frequency tree-ring reconstructions, using an updated Pinus sylvestris L. MXD network from the Scandinavian Mountains. Theoretical and Applied Climatology 113: 697–710.

Björklund

Gunnarson

Seftigen

. (2014) Blue intensity and density from Northern Fennoscandian tree rings – Using earlywood information to improve reconstructions of summer temperature. Climate of the Past 10: 877–885.

Briffa

Jones

Bartholin

. (1992) Fennoscandian summers from AD 500: Temperature changes on short and long timescales. Climate Dynamics 7: 111–119.

Campbell

McCarroll

Loader

. (2007) Blue reflectance in Pinus sylvestris: Application, validation and climatic sensitivity of a new palaeo-climate proxy for tree ring research. The Holocene 17: 821–828.

Campbell

McCarroll

Robertson Loader

. (2011) Blue intensity in Pinus sylvestris tree-rings: A manual for a new palaeoclimate proxy. Tree Ring Research 67: 127–134.

Cook

Krusic

(2005) Program ARSTAN: A Tree-Ring Standardization Program Based on Detrending and Autoregressive Time Series Modeling with Interactive Graphics. Palisades, NY: Tree-Ring Laboratory, Lamont-Doherty Earth Observatory, Columbia University.

Cybis Elektronik (2010) CDendro and CooRecorder. Available at: http://www.cybis.se/forfun/dendro/index.htm.

Esper

Duthorn

Krusic

. (2014) Northern European summer temperature variations over the Common Era from integrated tree-ring density records. Journal of Quaternary Science 29: 487–494.

10.

Gordon

(1982) Verification of dendroclimatic reconstructions. In: Hughes

Kelly

Pilcher

. (eds) Climate from Tree Rings. Cambridge: Cambridge University Press, pp. 58–61.

11.

Grudd

(2008) Torneträsk tree-ring width and density AD 500 2004: A test of climatic sensitivity and a new 1500-year reconstruction of north Fennoscandian summers. Climate Dynamics 31: 843–857.

12.

Gunnarson

Linderholm

Moberg

(2011) Improving a tree-ring reconstruction from west-central Scandinavia: 900 years of warm-season temperatures. Climate Dynamics 36: 97–108.

13.

Harris

Jones

Osborn

. (2014) Updated high-resolution grids of monthly climatic observations. International Journal of Climatology 34: 623–642.

14.

Jones

Lister

Osborn

. (2012) Hemispheric and large-scale land-surface air temperature variations: An extensive revision and an update to 2010. Journal of Geophysical Research 117: D05127.

15.

Linderholm

Gunnarson

Liu

(2010) A comparison of two Scots pine tree-ring proxies and standardization methods – Examples from the central Scandinavian Mountains. Dendrochronologia 28: 239–249.

16.

Linderholm

Björklund

Seftigen

. (2014) Fennoscandia revisited: A spatially improved tree-ring reconstruction of summer temperatures for the last 900 years. Climate Dynamics. Epub ahead of print 1 October. DOI: 10.1007/s00382-014-2328-9.

17.

McCarroll

Loader

Jalkanen

. (2013) A 1200-year multiproxy record of tree growth and summer temperature at the northern pine forest limit of Europe. The Holocene 23: 471–484.

18.

McCarroll

Pettigrew

Luckman

. (2002) Blue reflectance provides a surrogate for latewood density of high-latitude pine tree rings. Arctic, Antarctic, and Alpine Research 34: 450–453.

19.

Melvin

Briffa

(2008) A “Signal-Free” approach to Dendroclimatic Standardisation. Dendrochronologia 26: 71–86.

20.

Melvin

Grudd

Briffa

(2013) Potential bias in ‘updating’ tree-ring chronologies using regional curve standardisation: Re-processing 1500 years of Torneträsk density and ring-width data. The Holocene 23: 364–373.

21.

National Research Council (2006) Surface Temperature Reconstructions for the Last 2,000 Years. Washington, DC: National Academies Press, 196 pp.

22.

Raven

(2004) Biology of Plants. 6th Edition. New York: W. H. Freeman and Company/Worth Publishers.

23.

Regent Instruments Canada Inc. (2009) WinDendro for Tree-Ring Analysis. Available at: http://www.regentinstruments.com.

24.

Schweingruber

Fritts

Bräker

. (1978) The x-ray technique as applied to dendrochronology. Tree-Ring Bulletin 38: 61–91.

25.

Štěpánek

(2008) AnClim – Software for Time Series Analysis. Brno: Department of Geography, Faculty of Natural Sciences, Masaryk University.

26.

Stokes

Smiley

(1968) An Introduction to Tree-Ring Dating. Chicago, IL: University of Chicago Press.

27.

van Oldenborgh

te Raa

Dijkstra

. (2009) Frequency- or amplitude-dependent effects of the Atlantic meridional overturning on the tropical Pacific Ocean. Ocean Science 5: 293–301.

28.

Von Storch

Zwiers

(2004) Statistical Analysis in Climate Research. Cambridge: Cambridge University Press.

29.

Wigley

TML

Briffa

Jones

(1984) On the average value of correlated time series, with applications in dendroclimatology and hydrometeorology. Journal of Climate and Applied Meteorology 23: 201–213.

30.

Wilson

RJS

Luckman

(2003) Dendroclimatic reconstruction of maximum summer temperatures from upper treeline sites in Interior British Columbia, Canada. The Holocene 13: 851–861.