Consequences of different sample drying temperatures for accuracy of biomass inventories in forest ecosystems

Biomass estimation is one of the crucial tasks of forest ecology. Drying tree material is a crucial stage of preparing biomass estimation tools. However, at this step researchers use different drying temperatures, but we do not know how this influences accuracy of models. We aimed to assess differences in dry biomass between two drying temperatures (75 °C and 105 °C) in tree biomass components and to provide coefficients allowing for recalculation between the given temperatures. We used a set of 1440 samples from bark, branches, foliage and wood of eight European tree species: Abies alba Mill., Alnus glutinosa (L.) Gaertn., Betula pendula Roth., Fagus sylvatica L., Larix decidua Mill., Picea abies (L.) H. Karst., Pinus sylvestris L. and Quercus robur L. The differences between drying temperatures were 1.67%, 1.76%, 2.20% and 0.96% of sample dry masses of bark, branches, foliage and stem wood, respectively. Tree species influenced these differences. Our study provided coefficients allowing for recalculation of masses between the two temperatures, to unify results from different studies. However, the difference in dry mass between the two temperatures studied is lower than the range of uncertainty of biomass models, thus its influence on results of large-scale biomass assessments is low.


Results
The mean difference between dry masses obtained at different drying temperatures were highest for foliage (2.20 ± 0.06%) and lowest for wood (0.96 ± 0.01%). For bark and branches the mean differences were 1.67 ± 0.03% and 1.76 ± 0.01%, respectively. Models of differences between drying temperatures revealed statistically significant impacts of tree species for each component studied ( Table 2, S3). In the cases of bark and foliage we also found statistically significant influences of sample mass (Fig. 1). However, for foliage higher values of sample mass occurred only for two species (A. alba and P. abies). For bark samples fixed effects explained 20.7% of the variability in D and both random and fixed effects explained 26.7%. The highest D was in P. abies and P. sylvestris and the lowest in B. pendula (Table 3). Fixed effects in the model of D for branches explained 61.0% of variability whereas both random and fixed effects explained 64.4%. The highest D was found in L. decidua and the lowest in B. pendula. In the case of foliage, fixed effects explained 88.8% of the variability whereas both random and fixed effects explained 89.8%. We found the highest D in A. glutinosa while the lowest was in P. abies, A. alba and P. sylvestris. Fixed effects in the model of D in wood explained 88.7% of the variability whereas both random and fixed effects explained 89.8%. The highest D was in A. alba and Q. robur and the lowest in A. glutinosa, B. pendula and F. sylvatica.
Difference in tree stem biomass in samples of calculated biomass was highest for L. decidua (0.34-15.36 kg, with an average of 5.74 ± 0.71 kg), while the lowest-for F. sylvatica (0.19-12.25 kg, with an average of 4.05 ± 0.60 kg; Fig. 2). At the stand level we found the highest difference for F. sylvatica (0.05-1.42 Mg ha −1 , with an average of 0.73 ± 0.08 Mg ha −1 ) and the lowest-for L. decidua (0.03-1.24 Mg ha −1 , with an average of 0.61 ± 0.07 Mg ha −1 ; Fig. 2). At the country scale the difference was the highest for P. sylvestris (6.3 × 10 6 Mg), while the lowest-for Alnus spp. (0.4 × 10 6 Mg; Table 4). In total, for species studied the sum of species-specific differences was 11.2 × 10 6 Mg at the country level (Table 4). www.nature.com/scientificreports/

Discussion
Similar to Samuelsson et al. 37 , our study revealed the differences between tree species studied in reference to masses obtained by drying at different temperatures. For foliage, we found that all broadleaved tree species had higher D than coniferous tree species (Table 3). This may be connected with higher leaf area and higher moisture in broadleaved species. The highest D was for leaves of A. glutinosa, which is a nitrogen-fixing species and has one of the highest nitrogen concentrations in leaves [40][41][42] . As lower drying temperature is recommended for analysis of nitrogen content, this difference may be connected with the presence of nitrogen-based compounds. In the case of bark we found the lowest differences between drying temperatures in B. pendula, which has the thinnest bark. However, the species with thick bark, such as Q. robur and A. alba did not differ statistically significantly from B. pendula. A similar pattern was found for branch samples. For wood samples we found the lowest D values in three diffuse porous species and the highest in ring-porous Q. rubra and coniferous species, especially in A. alba. Two species in which wood contains numerous resin ducts-P. sylvestris and L. decidua-had intermediate D values. This pattern was connected with anatomical build of vessels in diffuse porous wood. In diffuse porous wood vessels in both earlywood and latewood have even lumen diameters, while in ring porous and coniferous wood, latewood has much tighter vessels. These vessels may be opened during drying at higher temperatures. www.nature.com/scientificreports/ In the cases of foliage and bark we found impacts of sample mass on differences in masses obtained after drying at two temperatures. In contrast to our expectations, sample mass did not affect wood and branches. This might be connected with the decrease of wood density along stem height 43 , increasing sample sensitivity to difference in drying temperature. Significant influence in the case of foliage was connected with different sample masses for A. alba and P. abies, connected with length and thickness of their needles. Thus, these differences were likely an effect of species and have no other biological meaning. For bark the trend was connected with higher variability of D in smaller samples. Variance estimated by random factors was low and did not amount to more than 6.2% (in bark). This indicates a low level of site-specificity and good transferability of our results.
Despite the low values of the differences, accounting for about 1%, our study provided values which allowed recalculation of biomasses using models developed from data obtained after different drying temperatures Comparing the differences between drying temperatures with errors of biomass estimation models we can state that the bias connected with drying temperature is lower than biases of tree-and stand-level biomass models. For example, mean difference of tree-level stem biomass was 4.05, 5.74 and 5.67 kg, while RMSE of tree-level models (Table S4) (Table S4) was 1.30, 0.71 and 6.34, respectively. For that reason we may assume that this source of differences has a low impact on compilation of data sources, such as large biomass databases [44][45][46] or generalized allometric equations based on published models 16,47 . Nevertheless, application of D coefficients proposed in our study would surely decrease the uncertainty. www.nature.com/scientificreports/ Our study revealed small differences between dry masses of biomass samples dried at the two most commonly used temperatures (i.e. 75 °C and 105 °C). We explored differences between biomass components and tree species studied. The differences among species may result from different morphology of the species studied. We also developed a set of coefficients which may be used to recalculate dry masses between the two temperatures studied. The differences between temperatures are lower than the range of uncertainty of models used in forest biomass assessments. For that reason this source of bias in large-scale forest carbon assessments is low. However, application of our coefficients may decrease these biases.
Although our study covered the eight most economically important tree species in temperate and boreal European forests, the results are not representative for all tree species from this area. Further studies should focus on more tree species rather than on sample size. We believe that D coefficients will be correlated with plant functional traits, especially with wood density and leaf dry matter content. Also, it can be phylogenetically-dependent, similarly to functional traits of wood 48 and leaves 49,50 . Therefore expansion of data about impacts of drying temperature on a wide set of species and correlation with traits and phylogeny can yield in models providing species-specific D coefficients transferable across understudied taxa. Such models would decrease uncertainty of the global carbon budget connected with uneven drying temperatures in previous studies. Also, this lead to the question whether we should use higher drying temperature, connected with higher carbon footprint, for studies on climate change mitigation capacity of forest ecosystems.

Methods
Study material. To assess the differences between drying temperatures we prepared random sets of samples collected during a large biomass assessment in Western Poland; for details based on the example of one species analysis see Jagodziński et al. 14 . We analyzed 1440 (15.6%; Table S1) of 9216 samples of biomass components from eight tree species (Table 1) from 124 study plots (Table S2). These samples consisted of four biomass components (stem bark, stem wood, fine branches, i.e. branches with diameter < 7 cm, and foliage) and eight tree species: Abies alba Mill., Alnus glutinosa (L.) Gaertn., Betula pendula Roth., Fagus sylvatica L., Larix decidua Mill., Picea abies (L.) H. Karst., Pinus sylvestris L. and Quercus robur L. We assumed that wood and bark of coarse branches (> 7 cm diameter) were similar to stem wood and stem bark, respectively. These species are the most frequent forest-forming trees in Central Europe (Table 1). From each species and each component we randomly selected 30 samples, with the exception of stem wood. For the latter we selected 90 samples, among which 30 were small discs (from the higher parts of stems), 30 intermediate and 30 large discs (from the lower parts of trees), to account for the variability in sample mass. As the research project was focused on aboveground biomass estimation, we did not harvest belowground organs of sample trees. For that reason our study did not cover this important and variable part of tree biomass.
After harvesting, each sample was dried at 75 °C for biomass assessment for other studies 14 and then stored in our sample warehouse, as study material was harvested from 2015 to 2017. After all samples were collected, we randomly selected our samples and dried them in ovens with forced air circulation (UN 750 and ULE 600, Memmert GmbH + Co. KG, Germany) at 75 °C to constant mass. Then, we weighed the samples with an accuracy of 0.1 g, dried them at 105 °C to constant mass and weighed them again. We determined mass of a few samples representing different dimensions every day and we assumed constant mass when a given sample did not change on two consecutive days.

Data analysis
All analyses were performed using R software 38 . All mean values are followed by standard error (SE). For each species and component we calculated the relative difference between dry weight of samples dried at both temperatures: where D-difference in masses, m 75 -dry mass of sample at 75 °C and m 105 -dry mass of sample at 105 °C. Calculated D allowed us to recalculate dry mass for each temperatures using the following formulas: www.nature.com/scientificreports/ To assess the differences among species studied and impacts on sample mass we used mixed effects analysis of covariance (ANCOVA), assuming sample mass and species as fixed effects and study plot as a random effect. The latter allowed us to evaluate and exclude site-specific effects. We did not account for sample tree as a random effect, as preliminary analyses revealed that this effect on variance was lower than 0.0000001 g. Mixed effects ANCOVA were implemented in the lmerTest::lmer() function 54,55 . To assess the differences among species we used Tukey post hoc tests, implemented in the multcomp::glht() function. We also used two types of determination coefficients to assess the proportion of variance explained by random and fixed effects. Marginal coefficients of determination (R 2 m ) express the amount of variance explained only by fixed effects and conditional coefficients of determination (R 2 c ) express the amount of variance explained by both random and fixed effects. These coefficients were calculated using the MuMIn::r.squaredGLMM() function 56 .
We showed impact of drying temperature differences using published biomass models at tree and stand levels and we used published biomass estimation at the country level. The difference also showed the level of uncertainty when drying temperature is unknown (for country level). We extracted tree and plot level biomasses from our previous studies for three species: P. sylvestris 57 , L. decidua 14 and F. sylvatica 58 (Table S4). We calculated biomass of tree stem and plot stem biomass across gradients of tree DBH and stand volume, for tree and plot level analyses (Table S5). We provided estimated biomass for DBH ranging from 10 to 50 cm and volume ranging from 10 to 300 m 3 ha −1 . For F. sylvatica we also used height ranging from 10 to 30 m, as models were based on both DBH and height. For calculations we assumed that bark comprised 15% of stem biomass, similarly as Jabłoński and Budniak 39 in country-level biomass assessment, as some of models provided only stem biomass.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information files).