Technical Comments

Comment on “The whole-soil carbon flux in response to warming”

See allHide authors and affiliations

Science  23 Feb 2018:
Vol. 359, Issue 6378, eaao0218
DOI: 10.1126/science.aao0218

Abstract

In a compelling study, Hicks Pries et al. (Reports, 31 March 2017, p. 1420) showed that 4°C warming enhanced soil CO2 production in the 1-meter soil profile, with all soil depths displaying similar temperature sensitivity (Q10). We argue that some caveats can be identified in their experimental approach and analysis, and that these critically undermine their conclusions and hence their claim that the strength of feedback between the whole-soil carbon and climate has been underestimated in terrestrial models.

Hicks Pries et al. (1) used a deep soil warming experiment to examine CO2 production in response to warming across the soil profile in a coniferous temperate forest. They found that a 2-year in situ warming of 4°C significantly enhanced the whole-soil CO2 production by 34 to 37%, and that all soil layers exhibited similar temperature sensitivity with a mean apparent Q10 of 2.7 ± 0.3. We argue that although the idea of subsoil carbon dynamics in response to warming is worth testing (2, 3), their conclusions are critically undermined by the experimental approach and analysis.

Hicks Pries et al. buried heating rods at a depth of 2.4 m (with additional circular heating cables near the surface at radii of 0.5 and 1 m) to warm the 1-m soil profile evenly by 4°C. This method differed from the majority of previous studies, in which soil temperature was elevated through heating surface air and/or topsoil layers to simulate the magnitude of rise in surface air temperature (2, 3). Although it is expected that ongoing and future air warming would increase the temperature of the soil profile, the magnitude of temperature elevation likely attenuates with depth. This happens because net energy inputs from air and subsequent heat conduction driven by a thermal gradient would inevitably invoke temperature variation in deep soils lagging behind that of surface soils under a dynamic climate system. Such a thermal lag effect is also ascribed to low thermal diffusivity and thickness of the soil profile (4). In an alpine meadow with clipping to simulate animal grazing under 4°C surface air warming, for instance, the magnitude of temperature elevation decreased by 88% and 77%, respectively, at 60- and 100-cm soil depths (5). In a boreal forest under 3.4°C soil warming at a depth of 10 cm, the magnitude of temperature elevation declined by 40% and 53%, respectively, at 75- and 100-cm soil depths (6). These previous findings imply that Hicks Pries et al.’s experimental warming approach may cause higher temperature elevations in deeper soil layers (e.g., <50 cm) than expected under air warming, thus overestimating the whole-soil CO2 production.

To examine temperature sensitivity of CO2 production across the soil profile, Hicks Pries et al. used the following equation to calculate the apparent Q10, a factor by which the CO2 production rate increases with a 10°C rise, of each soil layer:Embedded Image(1)where RC and RH are the CO2 production rates of each control and heated plot pair, respectively, and TC and TH are the soil temperatures of the corresponding control and heated plots, respectively. By directly comparing CO2 production between each treatment plot pair on every sampling date, however, a large number of anomalous Q10 values appeared (for example, 33% of them <1.0 and 20% of them >5). Hicks Pries et al. included Q10 values less than 1.0 but excluded those greater than 6.4 (45 of 281) as “unrealistic values” (Q10 > 30) or “outliers” (6.4 < Q10 < 30) in their analysis. At 7.5-cm soil depth, for instance, nearly one-third of the data points were not included in computing the mean apparent Q10. Although many Q10 values were ignored in Hicks Pries et al., they did not propose a different mechanism (other than the warming treatment) underlying the emergence of these anomalous data points. We believe that the analysis of Hicks Pries et al. is inappropriate, leading to the biased conclusion that all soil depths responded to warming with very similar temperature sensitivity.

In a seminal work, Hawkins defined an outlier as “an observation which deviates so much from other observations as to arouse suspicions that it was generated by a different mechanism” (7). Evidently, the apparent Q10 was not a direct observation but was calculated from the paired soil CO2 in their study. As such, the exclusion of any Q10 must be based on the statistical analysis of soil CO2 measurements rather than the Q10 itself. If any Q10 value could be considered an outlier or even unrealistic, the corresponding paired soil CO2 data should also be excluded in analyzing the mean soil CO2 production (figure 2A of Hicks Pries et al.). To reach a fair conclusion about the whole-soil response to warming, we reanalyzed the mean soil CO2 production using the original data sets by omitting those not included in the estimation of the mean apparent Q10 in figure 2B of Hicks Pries et al. In contrast to Hicks Pries et al., results from the current analysis show that CO2 flux from the whole-soil profile remained unchanged under warming (Fig. 1A, P = 0.15). This result indicates that the omission of many data points in the Hicks Pries et al. analysis could lead to inconsistent conclusions about the response of the whole soil to warming.

Fig. 1 Soil CO2 production of different soil depths in response to warming.

(A) Soil CO2 production (mean ± SE, n = 3) at different soil layers in control plots (blue circles) and heated plots (red circles). P < 0.05 denotes a significant effect of warming. Data are from soil CO2 measurements of Hicks Pries et al. but excluding the “outliers” and “unrealistic high values” in their Q10 analysis. (B) The mean apparent Q10 of each soil layer according to a linear regression model. Hicks Pries et al. estimated the mean apparent Q10 by averaging the Q10 values of each sampling point but excluded many data points because of high variation. Our method allows for a more accurate calculation of temperature sensitivity through an unbiased estimation of ln(Q10) with low deviation. Error bars were computed from SE of the fitted parameters. The apparent Q10 of the two upper depths Embedded Image is significantly higher than that of the three deeper depths Embedded Image, P < 0.001. Data are from the Hicks Pries et al. analysis of soil CO2 production. All paired soil CO2 data except for those equal to zero are included in the current analysis.

We argue that the exclusion of many data points in the Hicks Pries et al. analysis could be problematic because their data were collected from a time-series experiment. In their study, soil CO2 measurements had been conducted repeatedly on the same object (i.e., each soil layer) over 2 years; thus, measurements at different time points were not independent of each other (8). Actually, soil CO2 production decreased significantly over time in the whole-soil profile (P < 0.001). Moreover, two CO2 measurements taken at adjacent time points were more highly correlated than two taken several time points apart [the fittest covariance structure: first-order autoregressive with heterogeneous variance ARH(1), ϕ = 0.994]. If such repeated and self-correlated CO2 measurements were treated erroneously as if they were random samples, as in the analysis of Hicks Pries et al., the exclusion of a large number of data points could lead to serious power loss in statistical analysis (7).

To avoid the occurrence of many “abnormal” Q10 values, we propose a different method to estimate the mean apparent Q10 of each soil depth. In the Hicks Pries et al. analysis, many anomalous Q10 values appeared because any variation in soil CO2 data collection would be amplified exponentially by a power of 10/(THTC). To ameliorate error propagation, we use a least-squares regression model to estimate the apparent Q10 (9) by reformatting Eq. 1 as follows:Embedded Image(2)Using Eq. 2, the apparent Q10 for each soil layer can be estimated through performing a linear regression between ln(RH/RC) and (THTC). Meanwhile, the potentially confounding effects of seasonal changes, emphasized in Hicks Pries et al., could be minimized because the same paired data are used. In our analysis, only those data points with CO2 production equal to zero were not included. Tests of difference in Q10 between soil depths were conducted based on a generalized linear model by introducing a dumb variable of depth. Interestingly, we found that the mean apparent Q10 of the two surface soil layers (mean Q10 = 2.62) was significantly higher than that of the three deeper soil layers (mean Q10 = 1.20) (Fig. 1B, P < 0.001). These results are in conflict with those of Hicks Pries et al. and suggest that the surface soil layers were more responsive to warming, with a much higher temperature sensitivity relative to the deep layers.

Hicks Pries et al. concluded their analysis with a claim that the strength of the carbon-climate feedback in terrestrial models has been underestimated because these models have ignored the warming effects on subsoil (e.g., >50 cm) and usually include relatively low apparent Q10 values of soil CO2 production. Given the above caveats with respect to how this experiment was carried out and how the data were interpreted, we believe that their recommendation lacks a solid foundation.

References and Notes

Acknowledgments: We thank M. S. Torn for stimulating discussions. Supported by National Natural Science Foundation of China (NSFC) grants 31670501, 31422010, and 31370487 and by Zhejiang Provincial NSFC grant LR14C030001.
View Abstract

Navigate This Article