Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Multiple factors co-limit short-term in situ soil carbon dioxide emissions

  • James W. Raich ,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Visualization, Writing – original draft, Writing – review & editing

    jraich@iastate.edu

    Affiliation Department of Ecology, Evolution and Organismal Biology, Iowa State University, Ames, Iowa, United States of America

    ⨯
  • Mark S. Kaiser,

    Roles Formal analysis, Validation, Writing – review & editing

    Affiliation Department of Statistics, Iowa State University, Ames, Iowa, United States of America

    ⨯
  • Mathew E. Dornbush,

    Roles Data curation, Formal analysis, Investigation, Writing – review & editing

    Affiliation Cofrin School of Business, University of Wisconsin-Green Bay, Green Bay, Wisconsin, United States of America

    ⨯
  • Jonathan G. Martin,

    Roles Data curation, Investigation, Writing – review & editing

    Affiliation Natural Resources, Northland College, Ashland, Wisconsin, United States of America

    ⨯
  • O. J. Valverde-Barrantes

    Roles Data curation, Investigation, Writing – review & editing

    Affiliation Institute of Environment, Department of Biological Sciences, Florida International University, Miami, Florida, United States of America

    ⨯

Abstract

Soil respiration is a major source of atmospheric CO2. If it increases with warming, it will counteract efforts to minimize climate change. To improve understanding of environmental controls over soil CO2 emission, we applied generalized linear modeling to a large dataset of in situ measurements of short-term soil respiration rate, with associated environmental attributes, which was gathered over multiple years from four locations that varied in climate, soil type, and vegetation. Soil respiration includes many CO2-producing processes: we theorized that different environmental factors could limit each process distinctly, thereby diminishing overall CO2 emissions. A baseline model that included soil temperature, soil volumetric water content, and their interaction was effective in estimating soil respiration at all four locations (p < 0.0001). Model fits, based on model log likelihoods, improved continuously as additional covariates were added, including mean daily air temperature, enhanced vegetation index (EVI), and quadratic terms for soil temperature and water content, and their interactions. The addition of land cover and its direct interactions with environmental variables further improved model fits. Significant interactions between covariates were observed at each location and at every stage of analysis, but the interaction terms varied among sites and models, and did not consistently maintain importance in more complex models. A main-effects model was therefore tested, which included soil temperature and water content, their quadratic effects, EVI, and air temperature, but no interactions. In that case all six covariates were significant (p < 0.0001) when applied across sites. We infer that local-scale soil-CO2 emissions are commonly co-limited by EVI and air temperature, in addition to soil temperature and water content. Importantly, the quadratic soil temperature and moisture terms were significantly negative: estimated soil-CO2 emissions declined when soil temperature exceeded 22.5°C, and as soil moisture differed from the optimum of 0.27 m3 m-3.

Citation: Raich JW, Kaiser MS, Dornbush ME, Martin JG, Valverde-Barrantes OJ (2023) Multiple factors co-limit short-term in situ soil carbon dioxide emissions. PLoS ONE 18(2): e0279839. https://doi.org/10.1371/journal.pone.0279839

Editor: Chun Liu, Jinan University, CHINA

Received: November 18, 2021; Accepted: December 15, 2022; Published: February 15, 2023

Copyright: © 2023 Raich et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data are held in a permanent repository maintained by Iowa State University, and are freely available to use under Creative Commons Attribution (CC-BY-NC 4.0) International License. The data is available at https://doi.org/10.25380/iastate.20323674. The dataset includes 8039 rows of data and no missing values; it includes all the data that was used and analyzed within this publication.

Funding: This study was supported in the form of funding from the Center for Global and Regional Environmental Research (https://cgrer.uiowa.edu/), the U.S. National Science Foundation (NSF, [http://www.nsf.gov)]http://www.nsf.gov) (Grant Nos. EF-0343766, EF-0236502, and DEB-0703561), and by the USDA (https://www.sare.org/) (Grant No. SARE LNC00-171).

Competing interests: The authors have declared that no competing interests exist.

Introduction

Soils across the globe produce CO2 via respiration by the organisms they contain. The resulting CO2 emissions, soil respiration (Rsoil), comprise the second largest land–to–atmosphere CO2 flux, after total ecosystem respiration. Soil respiration varies temporally and spatially, often in parallel with abiotic factors such as temperature and water availability. Plants contribute substantial carbon to soils and, like Rsoil, are influenced by temperature and soil water content. The concomitant uptake, production, and release of CO2 from plants and soils confounds the interpretation of atmospheric CO2 measurements. For example, seasonally maximum soil emissions coincide with declining atmospheric CO2 concentrations during northern-hemisphere summers, when photosynthetic CO2 assimilation exceeds respiratory CO2 production [1], but the atmosphere records only their net effect. Changes in climate, land cover, plant growth, and their interactions are all likely to modify soil-CO2 emissions in the future, but the magnitude and timing of potential changes remain uncertain [2–4]. Given that soils release about 85 Pg of CO2-C into the atmosphere each year (range = 68–94 Pg y-1) [5–10], even relatively small changes in Rsoil can meaningfully alter the atmosphere’s warming potential.

Properly characterizing environmental influences on soil respiration is challenging, however, because Rsoil is a compound flux that derives from multiple organisms and processes [11]. Soils, for instance, produce CO2 via the decay of detritus within the litter layer (O horizon), the soil matrix, and dead wood, each of which forms a distinct environment. Carbon dioxide is produced by soil microbes, protozoans, fungi, worms, and arthropods; via the respiration of roots and mycorrhizae; and, in some cases, from abiotic weathering of carbonates. Each of those organisms and processes may respond to the environment in distinct ways. For instance, roots and soil organic matter respond differently to soil temperature [12]; different soil heterotrophs differ in their responses to low soil water potentials [13]; and soil microbial activity persists in soils that are drier than the plant wilting point [14, 15].

Photosynthetic CO2 assimilation directly sustains Rsoil by providing the carbon that fuels it. Global terrestrial net primary productivity averages ~56 Pg of C annually (range 46–65) [16–26]. That indicates that about 65% of global soil-CO2 emissions derive from the decomposition of above- and belowground detritus, and about 35% derive from the respiration of roots and mycorrhizae, or soil autotrophic respiration. Therefore, factors that influence plant photosynthesis, carbon transport to roots, root turnover, and root-rhizosphere respiration also affect Rsoil. Greater canopy photosynthesis, if accompanied by carbon sequestration in biomass or soils, provides a potentially important brake on increasing atmospheric CO2 levels [27–30]. Over 2009–2018, net C accumulation by global vegetation averaged an estimated 3.2 Pg y-1 of C, which more than offset net land use-change losses of 1.5 Pg y-1 [31], but whether C sequestration can continue to increase in parallel with warming-enhanced soil-CO2 emission rates is unclear [32, 33], especially with increasingly pervasive human land use [e.g., 34]. The complexity of soil carbon-cycling processes makes robust predictions of soil responses to environmental changes difficult.

Multiple environmental factors influence plant photosynthesis, the partitioning of photosynthates to above- and belowground processes, detritus production, and soil carbon cycling. Environmental change is similarly multifactorial, as it includes climatic and atmospheric changes, changes in the distribution and abundance of biota, and alterations in land cover and land use. Determining how environmental changes will influence terrestrial carbon cycling is a complex challenge. Soil respiration is central to that challenge, as a result of its magnitude, and because it is sensitive to environmental conditions. Importantly, soil respiration is measured directly, in contrast to canopy photosynthesis and ecosystem respiration. Soil respiration measurements across environmental gradients provide a valuable resource for model testing and improvement. We theorize that the multiple individual processes and organisms that contribute to Rsoil generate the potential for limitation of any single contributing process, by any environmental condition, to limit total CO2 emissions from soils. That is, in a manner analogous to multiple potential nutrient limitations to plant growth [35], we posit that environmental conditions that diminish soil-CO2 production by any important contributor to soil respiration, will lower total soil respiration.

To test this idea, we applied sequentially more complex mathematical models for estimating observed in situ soil CO2 emission rates, based on paired measurements of fluxes and environmental attributes, to identify statistically important covariates of Rsoil within and across sites, and to characterize the forms of their relationships to fluxes. Individual sites experience limited ranges of environmental variability. To broaden the robustness and usefulness of our findings, we gathered a multisite dataset that included only fully comparable Rsoil measurements. We were specifically interested in determining (Objective 1) whether Rsoil continued to increase at ever-warmer temperatures, when the joint influences of other environmental covariates were included. Completing that objective required that we (Objective 2) identify the cofactors that meaningfully improved fits of models that already included temperature, across ecosystem types and through time, and thus might influence the observed Rsoil × temperature relationship. Those objectives allowed us to (Objective 3) identify a common model form that included only consistently influential covariates, even if that was not the model of choice for any single location. We focused on identifying the covariates that consistently improved overall model fits, and the forms of those fits, rather than defining parameter values, which are specific to input datasets.

Methods

Study sites

We utilized data from four study sites that encompassed a range of ecosystem types, climatic conditions, and CO2 efflux rates (Table 1): La Selva, Costa Rica; Rhodes Farm, Iowa, USA; the Chequamegon National Forest in northern Wisconsin, USA; and Bear Creek, Iowa, USA. Replicated-block field experiments were conducted at the first two locations; observational comparisons were made at the other two. At each location, the primary experimental or observational treatment was land cover, as characterized by the plant community present: soil characteristics, climate and remotely sensed parameters varied among locations. Our research in Costa Rica was undertaken in full collaboration with the La Selva Biological Station of the Organization for Tropical Studies (https://tropicalstudies.org), which owns and manages the site and was collaboratively funded in support of this research. No biological samples or protected species were collected or exported. The research was approved and permitted by the Costa Rican Sistema Nacional de Áreas de Conservación, Ministerio del Ambiente y Energía, Pasaporte Científico number 084544671.

thumbnail
Download:
Table 1. Environmental characteristics of the four study sites.

https://doi.org/10.1371/journal.pone.0279839.t001

At Rhodes Farm we conducted a randomized-block field experiment that compared the effects of C3 versus C4 grass species on ecosystem-level carbon-cycling processes [38–40]. The original vegetation there was broad-leaved deciduous riparian forest that was converted to cropland early in the 20th century. The experimental site was cultivated approximately annually for at least 70 years before the experiment was established, usually for maize that was harvested for cattle fodder. The soils were Mollic Udifluvents derived from loess. The experiment contained four 15 m × 90 m blocks that paralleled Clear Creek. Blocks were bisected into C3 and C4 treatments. Within each treatment, three different species of perennial grass species were planted in monoculture in 15 m × 15 m plots. One of the C4 species did not establish, leaving each block with three C3 and two C4 grass species.

We also analyzed measurements from a randomized-block experiment that was established in 1988–1989 on newly abandoned pastures at the La Selva Biological Station in the Atlantic lowlands of Costa Rica. The experiment is described by [41, 42]. The soils were Typic Tropohumults in the Matabuey consociation [43]. The native vegetation was species-rich rainforest [44] that had been cleared and converted to pasture in the 1950s. We utilized data from plots of four planted, native, broad-leaved tree species that retained relatively full canopies in all four blocks: Hieronyma alchorneoides Allemão, Pentaclethra macroloba (Willd.) Kunth., Virola koschnyi Warb., and Vochysia guatemalensis Donn.Sm. The plantations were 16 years old at the beginning of this study, and each plot measured 50 m × 50 m. Seedlings were planted at a 3-m × 3-m spacing, with 256 seedlings per plot. The data used herein are a subset of a larger online dataset [45, 46].

The Chequamegon National Forest is dominated by northern hardwood forests but includes a variety of forest types and ages, which were revisited through time to characterize seasonal dynamics [47, 48]. The dataset we used included: temperate coniferous forests dominated by hemlock (Tsuga canadensis (L.) Carrière or red pine (Pinus resinosa Aiton); temperate deciduous forests dominated by deciduous broad-leaved tree species; mixed deciduous-coniferous forests; Populus tremuloides Michx. (aspen) stands of different ages; and recently clearcut sites. The study sites were within the airshed of the Willow Creek flux tower [49, 50] and were interspersed with wetlands and lakes. The soils were sandy loams on glacial till plains and moraines left from the Wisconsin glaciation.

The Bear Creek study was in Story County, Iowa, on several farms that participated in riparian-zone management activities [51]. The experiment was observational with replicated plots in each of multiple land-cover types [52, 53]. Most of the plots utilized in this current study were on poorly drained, fine-loamy, mesic Cumulic Endoaquolls. We measured Rsoil in three plots in each of five anthropogenic grassland types that were <1 to >50 years in age and were dominated by either C3 or C4 grasses, defining two land-cover treatments.

Soil fluxes and ancillary measurements

We measured soil-CO2 emission rates with LI-COR soil efflux measurement systems (LI-COR, Lincoln, Nebraska, USA) in parallel with concomitant and co-located measurements of surface-soil temperature and surface-soil volumetric water content (VWC). We did not include measurements from frozen soils or from snow-covered sites, which represent physically different conditions. Similar protocols were applied in the different studies. The fundamental measurement—one sample—consisted of a CO2 flux measurement from one chamber at one time in one plot, with concomitant measurements of surface-soil temperature and surface-soil volumetric water content. Each flux estimate was based on two-three sequential measurements that were averaged, following the recommendations of the manufacturer. All measurements were gathered between 06:00 and 18:00. Instruments were carried throughout the study sites to access the different blocks, treatments, and plots, which were sometimes distant from one another. The dataset analyzed for this study is openly available [54].

Field measurements of soil CO2 fluxes at La Selva are described in [41, 42, 45, 55]. Soil respiration was initially measured with an LI-6400 System attached to a 6400–09 Soil CO2 Flux Chamber that was placed onto 5-cm tall, 10-cm diameter soil collars. Later measurements were collected with an LI-8100A Automated Soil CO2 Flux System attached to an 8100–103 chamber, with 20-cm diameter soil collars. Soil temperature was measured with a soil temperature probe that was pushed 5–10 cm into the soil next to each collar at the time of flux measurement. Soil moisture was measured at the same time to 12-cm depth with time domain reflectometry (CS-615 Soil Moisture Probe, Campbell Scientific, Logan, Utah, USA). The entire collection of soil respiration data from that site is available online [46]: we herein used data collected in four plantation types over 45 across-site surveys and totaling 2497 individual flux measurements.

At Rhodes Farm, soil respiration was measured with an LI-6400 System attached to a 6400–09 flux chamber that was placed on 10-cm diameter soil collars that were randomly pre-positioned in each plot of each block. The study encompassed five plots in each of four blocks, and one or two soil collars were placed in each plot prior to flux measurements. Soil temperature was measured beside each chamber at the time of flux measurement. Soil moisture was monitored in the 0–15 cm surface soil horizon with MoisturePoint soil-water profiling probes (E.S.I. Environmental Sensors Inc., www.esica.com). We initially installed one probe in the center of each block, and later increased that to one probe per plot. We herein used data from 48 across-site measurement surveys, which included 1543 flux determinations.

Soil carbon dioxide fluxes in different forest types in the Chequamegon National Forest are described in [47, 48]. Soil respiration was measured with LI-COR 6400–09 Soil CO2 Flux Chamber placed on 5-cm tall, 10-cm diameter soil collars that were placed at eight new, random locations per plot for each of the measurement periods. An integrated temperature probe (steel-embedded copper-Constantan thermocouple, type T) was used to record soil temperature down to 10 cm by each soil respiration collar. Volumetric soil moisture down to 15 cm was determined by each collar by time domain reflectometry using a portable soil moisture probe (CS615 Water Content Reflectometer, Campbell Scientific, Logan, Utah, USA). We herein used data from 67 surveys that included 2667 individual measurements of soil-CO2 efflux.

At Bear Creek, Iowa, we used and LI-8100A Automated Soil CO2 Flux System attached to an 8100–102 automated flux chamber, which fit over 10-cm diameter soil collars. Soil temperature was measured next to each chamber at the time of flux measurement with a thermocouple probe. At Bear Creek, soil gravimetric (GWC, g water per g soil) rather than volumetric water content was measured. Gravimetric data were transformed to volumetric water contents (VWC, m m-1) based on soil bulk densities (BD, Mg m-3) that were measured using volumetric samplers with four samples per plot and corrected to standard units: (1)

The treatments at Bear Creek represented riparian buffer strips that were planted along Bear Creek on multiple active farms, and they had different ages and histories. We distinguished mixed C3 grasslands, which contained a mixture of cool-season grasses, from plantings of C4 prairie grasses, which included some prairie forbs. We utilized a total of 1332 flux measurement from three different plots in each of five different ages of grasslands for this study.

Our final cumulative Rsoil dataset included 8039 chamber-level determinations of soil-CO2 emission rates, each with paired measurements of: date, soil temperature (Tsoil), soil VWC, the location of the site, and the experimental or observational treatment (land cover). To that dataset we appended measurements of air temperature (Tair) and enhanced vegetation index (EVI). Mean daily air temperature data was downloaded from weather stations located near each site; daily minimum and maximum temperature were averaged to estimate the daily mean when necessary. To characterize the seasonal dynamics of canopy development we downloaded remotely sensed EVI from NASA’s MOD13Q1 product (Band 250m_16_days_EVI, https://daac.ornl.gov/) for the time periods covering soil flux measurements at each site [56]. Imagery for Rhodes Farm was centered on 41.890° N, 93.205° W and covered 2.25 × 2.25 km. Imagery for La Selva were centered at 10.4299° N, 84.0363° W, covered 2.25 x 2.25 km. Data for the Chequamegon National Forest were centered at 45.8º N, 90.0667º W and encompassed 6.25 x 6.25 km. Data for Bear Creek were centered over 42.191° N, 93.4897° W and covered 2.25 x 2.25 km. We used linear interpolation between imaging dates to estimate the EVI on dates when Rsoil was measured. In all cases multiple land cover types were included within the imaged areas, so EVI data refer to study-site locations, whereas vegetation- and soil-attribute data pertain to within-location plots.

Statistical analyses

We examined marginal pairwise associations among Rsoil and Tsoil, VWC, Tair, and EVI using traditional Pearson product-moment correlations (r). We used linear regression as the basic tool of statistical analysis, implemented using JMP (2019, SAS Institute Inc., Cary, NC, USA. www.jmp.com). Statistical analyses were based on the response variable log(Rsoil), with Rsoil expressed as micromol cm-2 s-1 of C, because variability in emission rates increased with temperature. We fit a sequence of increasingly complex models using log(Rsoil) as the response variable. Each step consisted of testing a null hypothesis that was evaluated four times, once at each location, in fulfillment of our objectives: “The addition of variable x does not improve the fit of the model.” At a given location (Rhodes Farm, La Selva, Chequamegon, or Bear Creek), let Yi denote log(Rsoil) for observation i = 1, …, n. Let x1,i denote Tsoil for this same observation, let x2,i denote VWC, and let x3,i denote EVI. Let ecosystem types at a given location be indexed by v = 1, …, V. Define a set of indicator variables as, for v = 1, …, V– 1:

The four basic models used for each location follow. Our baseline model included covariates of soil temperature and soil water content:

  1. Model i

In model (i) and all that follow, the error terms (ϵi) are assumed to be independent and identically distributed with N(0, sigma2) distributions. To test for possible effects of canopy photosynthesis and carbon supply to roots, we next incorporated EVI and Tair and compared model i to:

  1. Model ii

To test for non-linearity in the soil temperature and soil moisture relationships, we added quadratic terms for each:

  1. Model iii

Finally, to determine whether land cover influenced observed fluxes, we considered:

  1. Model iv

Note that in model iv we have grouped terms involving land cover as main effect terms in the first parentheses, interactions between land cover and soil temperature in the second parentheses, interactions between land cover and soil water content in the third parentheses, interactions between land cover and EVI in the fourth parentheses, and interactions between land cover and air temperature in the fifth parentheses. Likelihood ratio tests for the significance of effects in the results that follow were conducted in terms of these groups of parameters.

Our strategy was to begin with a basic model formulated around the major factors of soil temperature and soil moisture, and then examine the effect of adding additional covariates. Note that we are primarily interested in which terms should be deemed meaningful rather than the estimated regression coefficients of those terms. Particularly in models that contain polynomial terms, interpretation of regression coefficients can be difficult if not impossible. The models given in (i-iv) were fit to data from individual locations. They were also fit to a combined data set containing all the observations from all four locations. We were especially interested in any differences found in the sets of meaningful model terms between the combined and location-specific (individual) data sets.

Estimation and testing of regression models were conducted within the framework of generalized linear models (glm) using normal random components and identity link functions. Formal tests between models were conducted as likelihood ratio tests, and tests of individual model parameter significance were conducted as Wald tests, the asymptotic bases of these procedures being justified by the large sample sizes at each of our four individual locations (N = 1332–2667). The Akaike information criterion (AIC) was also used in model comparison. Plots of deviance residuals against fitted values were examined as a diagnostic for model performance.

Results

Correlation analyses

Soil-CO2 emission rates correlated significantly and positively with Tsoil, Tair, and EVI at all four locations, but negatively with VWC (Table 2). Significantly positive correlations were also observed between Tsoil and EVI, Tsoil and Tair, and between EVI and Tair. Negative correlations were consistently observed between Tsoil and VWC, VWC and EVI, and between VWC and Tair (Table 2). Given the patterns of correlation observed among the variables that were potential covariates, we expected regressions with multiple covariates to display collinearity of effects, which contributes to our focus on which terms appear meaningful in models, rather than estimation of the numerical coefficients associated with those terms.

thumbnail
Download:
Table 2. Pearson product-moment correlations (r) between individual measurements of Rsoil and paired measurements of environmental variables at each of the four study sites.

https://doi.org/10.1371/journal.pone.0279839.t002

Regression analyses

Models i-iv were fit to data from each location individually, and a summary of the results is contained in Table 3, with parameter p values shown in S1 Table. Throughout, we refer to comparisons with an associated significance level of p < 0.05 as “significant”, although actual p values will be reported wherever possible. Residual plots are commented on, largely without discussion.

thumbnail
Download:
Table 3. Summary results of models fitted to data from four locations.

https://doi.org/10.1371/journal.pone.0279839.t003

Our baseline model i confirmed expectations that surface-soil temperature (Tsoil, °C) and VWC significantly covaried with Rsoil (micromol cm-2 s-1): they and their interaction were significant at all locations other than La Selva, where their interaction was not informative (S1 Table). Deviance residual plots (Fig 1) did not identify major problems with the model.

thumbnail
Download:
Fig 1. Deviance residuals of the five tested models, based on all observed data.

The main-effects model (mem) included only the main effects, without interactions: Tsoil, Tsoil2, VWC, VWC2, EVI and Tair. Horizontal lines divide each plot into positive and negative residuals, for ease of interpretation: the sample size for each is 8039.

https://doi.org/10.1371/journal.pone.0279839.g001

Adding either EVI or Tair to the baseline model improved model fits at each site: adding them both resulted in model ii, which improved model fits at all four sites (Table 3). Both EVI and Tair significantly covaried with Rsoil at all sites except, again, La Selva, where in situ variability in both variables was low. Model ii also increased the number of interaction terms by 12. Three of those additional terms were significant at three of the four locations, but none were significant at all four (S1 Table). In addition, in this more complex model, Tsoil was rendered non-significant at Bear Creek, and VWC was rendered nonsignificant at Rhodes Farm, although interactions involving each were significant at some locations. Residual plots for fits of model (ii) were not obviously better than those of model (i).

To test for non-linearities, we individually added quadratic terms for VWC and Tsoil into model ii. Each generated larger model log-likelihoods at each site (Table 3). Model (iii) included both quadratic terms, and better fit the observed data at all four sites than did model ii (p < 0.0.0004, S1 Table). The important interaction terms differed from those of model ii. The linear and quadratic terms for both soil temperature and soil moisture were significant at all four locations. Interestingly, the presence of these squared terms shifted the p values for the interactions. For example, in comparison to model ii, the significance of the Tsoil × Tair interaction was lost at two locations in model iii but emerged at the other two (S1 Table). The deviance residuals from model iii were similar to those of model ii (Fig 1).

Model iv included all terms from model (iii) plus the categorical variable land cover, of which there were two at Rhodes Farm and Bear Creek, five at Chequamegon, and four at La Selva. Although the number of parameters was again increased, log likelihoods increased and AIC decreased, at all four locations (Table 3). Land cover was highly significant at all four locations, and interactions between land cover and environmental covariates also occurred at each location (S1 Table). The significance of other interactions was similar in models iii and iv. Deviance residuals (Fig 1) from model iv were similar to those from models ii and iii.

In proceeding from our baseline model i through model iv, there was a consistent pattern of lower AIC for more complex models despite increases in the number of parameters (Table 3). That finding was mirrored by likelihood ratio tests of the model sequence, which were entirely nested (S2 Table). Comparisons of model i versus model ii, model ii versus model iii, and model iii versus model iv had p values of less than 10−5 in 80% of cases and were always significant. The more complex model was favored in each case at each step.

In addition, each of models i-iv was fit to a data set that combined measurements from all four locations into one collection with 8039 observations (Table 3), which was analyzed without regard to location. The results of these models were used to conduct likelihood ratio tests of full models with location-specific parameters against reduced models with a common set of parameters across all four locations (S2 Table). The maximized log likelihoods for the full models are obtained as a sum across locations of those values. The maximized log likelihood for the reduced model is obtained by fitting the combined data. For each of the five models considered, these tests were highly significant (p-values less than 10−5), with the result that in each case we would reject the combined model and favor of the location-specific models.

Main-effects model

With Objective 3 we sought to identify a common model form that included only consistently influential independent variables, even if that was not the model of choice for each location based on AIC and likelihood ratio tests. The main-effects model that emerged (model mem, Table 4) contained only the consistently informative covariates Tsoil, VWC, EVI, and Tair. It did, notably, include quadratic terms for both soil temperature and soil moisture. All terms in this model were significant for three of the four locations, with only VWC and VWC2 at Bear Creek as exceptions. Given this consistency, we can examine the estimated parameter values for these covariates (Table 4). There is some consistency in estimated parameter values over locations, although a likelihood ratio test of separate models for individual locations versus a single model for combined locations was significant (p value less than 10−5, S2 Table), so that we would reject the combined model and conclude that separate parameter values for each location would be preferred. At all sites and in the combined model, all intercepts were < -1.0, at which Rsoil equaled 0.37 micromol cm-2 s-1. Perhaps the most important indication from the parameter values in Table 4 is that all significant squared terms for soil temperature and soil VWC were negative, whereas all other terms were positive, suggesting again that log(Rsoil) has concave relationships with both Tsoil and VWC. This inference was supported by application of the main-effects model (mem) to the combined dataset (column 6), in which all included covariates, as well as quadratic terms for soil moisture and temperature, were highly significant.

thumbnail
Download:
Table 4. Summary of the main-effects model.

https://doi.org/10.1371/journal.pone.0279839.t004

Discussion

Our over-arching goal was to identify environmental correlates of in situ soil respiration that meaningfully improved understanding of soil-CO2 emission variability within and across four locations, from which multiannual field measurements were made using comparable techniques. Each location was unique in terms of soils, vegetation, or climate (Table 1). We were specifically interested in determining (Objective 1) whether Rsoil continued to increase at ever-warmer temperatures, when the joint influences of other environmental covariates were included. By including multiple locations and years in our analyses, we sought to broaden the ranges of environments included, to better characterize which covariates were commonly important, and the forms of their effects. We summarize the models we applied in Table 3, and parameter significance levels for the main models (i, ii, iii, and iv) in S1 Table.

Multiple factors influenced CO2 emission rates

Our baseline model i proved consistent across locations, with soil temperature and soil water content being significant everywhere, as is widely observed, and their interaction being significant except at one location with uniformly warm and moist soils (La Selva, S1 Table). To address Objective 2, we incorporated the enhanced vegetation index (EVI) and mean daily air temperature (Tair) within model ii, as potential covariates of canopy processes that contribute to shoot-to-root carbon fluxes, thus broadening the model to potentially include the effects of plant C assimilation. Ambient temperature controls rates of net leaf photosynthesis across plant types [57], independently of leaf area. We found that including Tair improved model performance even when Tsoil was already included (Table 3) and despite that Tair and Tsoil correlated strongly (Table 2). The EVI is a widely available, remotely sensed optical measurement of canopy greenness that incorporates influences of total leaf chlorophyll, leaf area, canopy cover, and canopy structure. The use of spectral imagery to improve local-scale Rsoil models has previously been demonstrated [58–60]. We cannot conclude cause-and-effect from these results, but inclusion of both EVI and Tair improved model performance (S1 Table), possibly by reflecting shoot-to-root carbon fluxes. Our results demonstrate that including both EVI and Tair within models can improve local-scale estimations of Rsoil, as they do in larger-scale studies [e.g., 61–64].

We did not test the importance of all potentially important covariates of Rsoil, but we demonstrated that VWC, Tsoil, EVI, and Tair were consistently influential. We addressed Objective 2, the potential for non-linearity in the soil temperature-respiration relationship, when the effects of other covariates were included, by adding quadratic terms for both Tsoil and VWC to model ii, generating model iii, with six potential covariates. Again, model performance was improved. Both soil temperature and its quadratic term were significant at all four locations. Both soil moisture and its quadratic term were also significant at all four locations (S1 Table). Throughout these steps the importance of individual covariates and their interactions differed among locations and from model to model, and even their signs, + or -, sometimes switched.

Adding a categorical descriptor of the plot-scale land cover being studied, generating model iv, further improved model fits at all four locations (S1 Table). Land cover had the single lowest p value at Bear Creek, Chequamegon, and La Selva, and p = 0.006 at Rhodes Farm. Model iv demonstrated that land cover type significantly influenced rates of soil CO2 emission within locations, even after we incorporated multiple location-specific environmental attributes as covariates. We conclude that land cover is very important to include in analyses of emission rates, as reported by others [10, 60, 65, 66].

Our overall goal was to identify the joint influences of different environmental factors on the magnitudes and variability of total in situ CO2 emissions, which reflect both within-soil CO2 production and its release into the atmosphere. The results of our empirical modeling exercise are best interpreted in terms of patterns that emerged, or were absent, over the progression of models considered. It was not our objective to determine “the best” model for relating soil respiration to all the plausible covariates that have been identified in one or another site, but to characterize the effects of previously recognized covariates. We found the Tsoil, VWC, their squared terms, EVI, Tair, and land cover all improved estimations of Rsoil, in comparison to models containing fewer of those covariates (S2 Table).

Interactions

We included interaction terms in models not to estimate how one effect changed at different levels of another, the traditional interpretation of interactions, but rather as an indication of complexity in the underlying mechanisms that may be involved in determining CO2 emission from soils. Throughout, we sought to identify stable and meaningful effects of potential covariates across locations. Our data sets were large enough to support models with multiple covariates. We did not, however, expect measures of model fits, based on likelihood ratio tests and AIC, to continue to favor models containing more terms and more interactions. Model i contained four parameters with one interaction term; model iv contained an average of 29 parameters including 15 interaction terms; yet at each location and at each step, model log likelihoods increased and AIC decreased as model complexity increased (Table 3).

Throughout the progression of model i to model iv the number of interaction terms increased, from one to >22. The significance of these interaction terms was apparently haphazard, with most terms being flagged as meaningful at some location, but only two interactions being meaningful at all four study locations: Tsoil×EVI in model ii; and Tsoil×EVI×Tair in models iii and iv (S1 Table). For example, the Tsoil×VWC interaction was significant at Chequamegon National Forest in models i, ii and iv but not in model ii, where p > 0.2. Even the nature of its effect, positive or negative, varied (S1 Table). The Tsoil×VWC interaction was never significant at La Selva but was influential in all four models at Bear Creek. The indication is that land cover and interaction terms that were added to basic models were picking up some structure present in location-specific data sets, but that structure was not necessarily related in any broadly applicable mechanistic fashion to the physical covariates involved. Overall, the lack of a pattern in the significance of interaction terms suggests that there were complex relations among the primary variables of Rsoil, Tsoil, VWC, Tair, and EVI at individual study locations, and that those relationships were not necessarily consistent across land-cover types or geographic locations.

Marginal effects and maximal responses

The conclusion of the preceding paragraph suggested that it might be profitable to examine marginal models which, although they may not provide the “best fit” at any individual location, retain some stability in meaningful terms and estimated parameters (Objective 3). Our analysis indicated that, for the four locations considered in this work, that model included soil temperature and soil water content, along with their squared terms, and EVI and air temperature:

  1. Main-effects model

In this model, the parameter values were visually stable in sign and rough magnitudes over locations, although formal tests indicate we would still prefer to estimate parameters for this model separately for each location (S1 Table). Of particular interest for this marginal model, both soil temperature and soil water content appear to have marginal concave relations with soil respiration.

To visualize these marginal relations, we plotted the expected values of Rsoil (i.e., the regression predicated values) from the marginal regression fit to the combined data (Table 4) against the observed values of the individual covariates of Tsoil, VWC, EVI, and Tair (Fig 2). These plots indicate a type of limiting effect of each covariate on the expected response, as by [67]. Maximal expected responses (for observed values of the other covariates) appear to be related in a monotonic fashion for both EVI and Tair, but in an intermediate-optimum fashion for Tsoil and VWC. The exact estimated expectations are functions of all four covariates as a set, and most of the expected values fail to obtain the largest values possible, being scattered below the maximal or limiting curve. The estimated expectations (mu; hat) plotted on the vertical axes of the graphs in Fig 2 were computed as: (2) where lambda;hat represents the regression-based estimate of log(Rsoil), and sigma;hat2 represents its variance (N = 8039). The value of EVI at which an overall maximum occurs will simply be the largest value of EVI observed (i.e., 0.7214) and the same is true for Tair (31.01°C). We can determine the values of Tsoil and VWC at which overall maximums will occur by setting partial derivatives equal to zero, which gives 0.27 m3 m-3 for VWC and 22.5°C for Tsoil. These match well visually with Fig 2.

thumbnail
Download:
Fig 2. Expected soil-CO2 emissions across observed ranges of (A) surface-soil temperature; (B) surface-soil volumetric water content; (C) enhanced vegetation index (EVI); and (D) air temperature. Expected values are based on the main effects model applied to the observed environments (N = 8039).

Symbol colors represent locations: Bear Creek—orange; Chequamegon—purple; La Selva—green; and Rhodes Farm—yellow.

https://doi.org/10.1371/journal.pone.0279839.g002

In this main-effects model, both EVI and air temperature covaried positively with Rsoil (Fig 2C and 2D), to maxima defined by the upper ranges of the observed data. Given that caveat we are unable to guess what might happen if either was greater. We gathered data from multiple sites to enhance the breadth of environments included, but the use of observed data to forecast possible futures faces this problem, i.e., the inability to address unobserved conditions. However, including EVI and Tair into our models improved model fits to the observed data (S2 Table), indicating that they were meaningful covariates with Rsoil. Altogether we found that Rsoil was potentially limited by four environmental factors (Fig 2), and was affected by the land cover (Table 3). These findings are consistent with our understanding of Rsoil as a system-level flux that derives from multiple interacting organisms and processes, each of which may be influenced distinctly by environmental conditions.

Our analysis also indicated that soil-CO2 emissions varied quadratically with soil moisture. That is consistent with previous observations that maximum CO2 production occurs at intermediate soil moisture contents [47, 68–75]. Soil water content influences vegetation productivity, detritus decay rates, and soil-atmosphere gas exchanges. Dry conditions diminish the availability of dissolved substrates to soil organisms, and directly depress cellular expansion and replication [69, 70]. In wet soils, water-filled pores diminish soil-atmosphere gas exchange, whereas a lack of air-filled pore volume may limit oxygen availability to soil organisms and thereby inhibit aerobic respiration [71, 76–78]. In our study, relatively low soil water contents were observed at all three temperate locations, but they did not depress CO2 emissions to the same extent as high soil water contents did. The optimal soil water content for CO2 emission across the sites was 0.27 m3 m-3, with emission rates declining in both drier and wetter soils (Fig 2B). That compares well with the ~27% global optimum inferred by [79], but we expect that the relationship will vary among soil types, textures, and organic matter contents [80].

We specifically sought, with Objective 1, to determine if Rsoil increased monotonically with increasing soil temperature, when additional covariates were considered. We found that emissions did strongly increase as soils warmed, but only to a maximum at ca. 22.5°C across locations. As soils warmed beyond that, emission rates neither increased further nor plateaued—they declined (Fig 2A). Slightly warmer temperature optima were inferred by [6], i.e., ~25°C across nine biomes; and by [81], i.e., 24.1–27.4°C for subtropical ecosystems. In our case, expected emissions at 30°C were the same as those at 15°C. Why might soil respiration decline as soils warm? We did not generate any evidence that soil heterotrophic respiration declined as soils warmed, whereas many studies have demonstrated that temperature optima for soil microbial respiration are higher than the temperatures we observed [82–86]. Our finding that Rsoil declined beyond an optimum temperature was robust in all four locations in our study (S1 Table, models iii–iv).

To pursue this question further, we evaluated the form of the relationship between temperature and the Q10 of Rsoil, by applying the main-effects model to generate expected values of Rsoil at soil volumetric water contents (m3 m-3) of 0.13 (10th percentile), 0.27 (optimum), and 0.52 (90th percentile), across a soil temperature range of 0°C to 31°C, given maximum observed values of EVI and Tair. We found that the Q10 of Rsoil was not affected by soil water content, consistent with [87], although its magnitude was (Fig 3). Reported Q10 values for Rsoil vary among studies [88], but usually within the range of 1–5 [66, 89–91]. In our case, the expected Q10 of emissions varied from 0.57 to 4.27: it declined from an average of 4.3 at 0°C, to 2.0 at 12°C, to 0.6 at 30°C (Fig 3B). The expected across-site temperature optimum for Rsoil occurred at 22.5°C, where the expected Q10 equaled 1.0; emission rates thereafter declined at warmer temperatures. Similar results are presented elsewhere [62, 89, 92]. Based on this analysis, the Q10 of Rsoil varied broadly, even within sites, and its magnitude varied with temperature but not moisture. This finding results from the significantly negative Tsoil2 terms in our models, but potentially explains, in part, variability in observed Rsoil Q10 values within and among studies.

thumbnail
Download:
Fig 3. Expected relationships between soil-CO2 emissions and temperature at three levels of soil volumetric water content based on the main-effects model.

(A) Expected Rsoil in relation to soil temperature at low, medium and high soil volumetric water contents. The temperature at maximum flux was 22.5°C regardless of soil water content. (B) The Q10 of emissions declined as temperature increased, but did not vary with water content. The plotted relationships show the midpoints of 10°C intervals, i.e., the Q10 at X = 2°C is the ratio of the expected fluxes at 7 and -3°C. Note that the individual relationships for low, medium, and high water contents overlie one another. The horizontal dotted line identifies Q10 = 1, below which estimated fluxes were lower at warmer temperatures.

https://doi.org/10.1371/journal.pone.0279839.g003

Our results seem to highlight the importance of temperate-zone seasonality: air temperatures <30.1°C, the maximum observed temperature within our dataset, were associated with lower soil-CO2 emissions (Fig 2D). The effects of EVI (Fig 2C) were similar: both EVI and air temperature were low at the beginning of the growing seasons in the temperate sites and increased as growing seasons progressed. Unfortunately, our dataset included only 38 observations at air temperature of ≥ 30°C, and those were all at one location on one day. The upper limit for positive net CO2 uptake by woody and many temperate-zone plants typically ranges from ca. 35–50°C [57, 81], beyond the range of our observations. We therefore cannot address the effects of air temperatures higher than 30°C on Rsoil or its autotrophic component. Pertinently, the air-temperature optima of vegetation productivity (i.e., estimated gross primary productivity) were found to be consistently lower than those for leaf photosynthesis, with a peak at 23°C [63], or with peaks at 18 and 28°C for C3 and C4 vegetation [93]. Those findings are similar to our across-location estimate of 22.5°C for Rsoil, suggesting—or confirming—that plant productivity influences soil CO2 emissions.

Models and meaning

Our analyses were consistent in demonstrating that location-specific analyses provided better model fits to observed data than did models based on the same data but excluding location (Table 3, S1 and S2 Tables). Our results suggest that incorporating local-scale information about EVI and Tair, in addition to Tsoil and VWC, can improve local-scale analyses of Rsoil: quantitative data for both variables are widely available. Many other environmental variables have been shown to covary significantly with in situ Rsoil and might be useful at other locations: further studies of additional variables and other locations are needed. Our results do demonstrate that incorporation of Tsoil, Tair, VWC, and EVI can be useful for empirical modeling of in situ soil-CO2 emissions at local and broader scales, and that quadratic terms for Tsoil and VWC can improve model fits. We also found that land cover typically had the single-most highly significant influence on Rsoil, within sites, even after multiple bioclimatic variables were included (Table 3, S1 Table). That indicates that models should distinguish among land cover types and experimental treatments whenever possible, to improve analyses. Given these findings, we sought insights into broader-scale questions. We found that a main-effects model that included only consistently meaningful covariates, without interactions, consistently provided interpretable findings across sites, despite that it was never ‘the best’ model at any one site.

Soil-CO2 emission constitutes a major land-to-atmosphere carbon flux that is highly variable over land use and vegetation types and is responsive to climatic conditions. That soil-CO2 emissions can be measured directly, in contrast to processes such as gross and net primary productivity and ecosystem respiration, allows the development of empirically derived models for land-to-atmosphere flux that can be used to test model predictions, and perhaps suggest modifications to those models [94, 95]. In the development of purely empirical models of soil-CO2 emissions it is important to distinguish between models that seek to provide the best possible fit to a data set of limited spatial and temporal scope, and models that seek to provide insight into the most consistently important factors across the range of situations in which data are collected. These consistently important factors, if they exist, become candidates for ubiquitous mechanistic forces that drive the system. We show that soil-CO2 emissions at local levels are influenced by a variety of potentially interacting variables, and by many possible interactions that exist among those variables. We show that a much smaller number of covariates in a model containing only main effects is stable across situations, meaning that the significance of the variables and their estimated coefficients are consistent among regression models estimated from different data sets. We also demonstrate that soil moisture and temperature relationships are significantly non-linear: emissions may be expected to either increase or decline if temperatures and rainfall increase, depending upon their prior conditions. Recognizing that local soil-CO2 emissions are influenced by multiple situation-specific limiting factors, but that a smaller and consistently important set of widely limiting factors may exist, can benefit terrestrial carbon-cycling analyses across scales.

Supporting information

S1 Table. Significance of model parameter values at four locations.

The right column refers to the same model applied to the data from all four sites without reference to location. An NS indicates not significantly different from zero.

https://doi.org/10.1371/journal.pone.0279839.s001

(DOCX)

S2 Table. Likelihood Ratio tests of differences between models, where Model 1 is nested within Model 2.

The chi-squared and p-values refer to one-sided tests of the null hypothesis that the log-likelihood of Model 2 is not different from that of Model 1.

https://doi.org/10.1371/journal.pone.0279839.s002

(DOCX)

Acknowledgments

We thank Ricardo Bedoya, Eugenio González, and Marlón Hernández for their assistance with field measurements and data management.

References

  1. 1. Raich JW, Potter CS, Bhagawati D. Interannual variability in global soil respiration, 1980–1994. Glob Change Biol. 2002;8: 800–812.
  2. 2. Davidson EA, Janssens IA. Temperature sensitivity of soil carbon decomposition and feedbacks to climate change. Nature 2006;440: 165–173. pmid:16525463
  3. 3. Bradford MA, Wieder WR, Bonan GB, Fierer N, Raymond PA, Crowther TW. Managing uncertainty in soil carbon feedbacks to climate change. Nat Clim Change 2016;6: 751–758.
  4. 4. Bond-Lamberty B, Bailey VL, Chen M, Gough CM, Vargas R. (2018) Globally rising soil heterotrophic respiration over recent decades. Nature 2018;560: 80–83. pmid:30068952
  5. 5. Hashimoto S, Carvalhais N, Ito A, Migliavacca M, Nishina K, Reichstein M. Global spatiotemporal distribution of soil respiration modeled using a global database. Biogeosciences 2015;12: 4121–4132.
  6. 6. Carey JC, Tang J, Templer PH, Kroeger KD, Crowther TW, Burton AJ, et al. Temperature response of soil respiration largely unaltered with experimental warming. P Natl Acad Sci USA 2016;113: 13797–13802. pmid:27849609
  7. 7. Xu M, Shang H. Contribution of soil respiration to the global carbon equation. J. Plant Physiol 2016;203: 16–28. pmid:27615687
  8. 8. Zhao Z, Peng C, Yang Q, Meng F-R, Song X, Chen S, et al. Model prediction of biome-specific global soil respiration from 1960 to 2012. Earths Future 2017;5: 715–729.
  9. 9. Jian J, Bailey V, Dorheim K, Konings AG, Hao D, Shiklomanov AN, et al. Historically inconsistent productivity and respiration fluxes in the global terrestrial carbon cycle. Nat Commun 2022;13: 1733. pmid:35365658
  10. 10. Huang N, Wang L, Song X-P, Black A, Jassal RS, Myneni RB, et al. Spatial and temporal variations in global soil respiration and their relationships with climate and land cover. Sci Adv 2020;6: eabb8508. pmid:33028522
  11. 11. Kuzyakov Y. Sources of CO2 efflux from soil and review of partitioning methods. Soil Biol Biochem 2006;38: 425–448.
  12. 12. Boone RD, Nadelhoffer KJ, Canary JD, Kaye JP. Roots exert a strong influence on the temperature sensitivity of soil respiration. Nature 1998;396: 570–572.
  13. 13. Manzoni S, Schimel JP, Porporan A. Responses of soi microbial communities to water stress: results from a meta-analysis. Ecology 2012;93: 930–938.
  14. 14. Whitmore AP, Whalley WR. Physical effects of soil drying on roots and crop growth. J Exp Bot 2009;60: 2847–2857. pmid:19584120
  15. 15. Liang J, Chen K, Siqintqna , Huo T, Zhang Y, Jing J, et al. Towards improved modeling of SOC decomposition: soil water potential beyond the wilting point. Glob Change Biol 2022;28: 3665–3673. pmid:35152535
  16. 16. Field CB, Behrenfeld MJ, Randerson JT, Falkowski P. Primary production of the biosphere: integrating terrestrial and oceanic components. Science 1998;281: 237–240. pmid:9657713
  17. 17. Cramer W, Kicklighter DW, Bondeau A, Moore B III, Churkina G, Nemry B, et al. Comparing global models of terrestrial net primary productivity (NPP): overview and key results. Glob Change Biol 1999;5(Suppl. 1): 1–15.
  18. 18. Saugier B, Roy J, Mooney HA. Estimations of global terrestrial productivity: Converging toward a single number? In: Roy J, Saugier B, Mooney HA, editors. Terrestrial global productivity: Physiological ecology. New York, Academic Press; 2001. pp. 543–557.
    • 19. Nemani R, Keeling CD, Hashimoto H, Jolly WM, Piper SC, Tucker CJ, et al. Climate-driven increases in global terrestrial net primary production from 1982–1999. Science 2003;300: 1560–1563.
    • 20. Cao M, Prince SD, Small J, Goetz SJ. Remotely sensed interannual variations and trends in terrestrial net primary productivity 1981–2000. Ecosystems 2004;7: 233–242.
    • 21. Del grosso S, Parton W, Stohlgren T, Zheng D, Bachelet D, Prince S, et al. Global potential net primary production predicted from vegetation class, precipitation, and temperature. Ecology 2008;89: 2117–2126. pmid:18724722
    • 22. Ito A. A historical meta-analysis of global terrestrial net primary productivity: Are estimates converging? Glob Change Biol 2011;17: 3161–3175.
    • 23. Potter C, Klooster S, Genovese V. Net primary production of terrestrial ecosystems from 2000 to 2009. Climatic Change 2012;115: 365–378.
    • 24. Running SW. A measurable planetary boundary for the biosphere. Science 2012;337: 1458–1459.
    • 25. Pan S, Tian H, Dangal SRS, Zhang C, Yang J, Tao B, et al. Complex spatiotemporal responses of global terrestrial primary production to climate change and increasing atmospheric CO2 in the 21st Century. PLOS ONE 2014;9: e112810.
    • 26. Yu T, Sun R, Xiao Z, Zhang Q, Liu G, Cui T, et al. Estimation of global vegetation productivity from global land surface satellite data. Remote Sens 2018;10: 327.
    • 27. Yvon-Durocher G, Jones JI, Trimmer M, Woodward G, Montoya JM. Warming alters the metabolic balance of ecosystems. Philos T R Soc B 2010;365: 2117–2126.
    • 28. Myhre G, Shindell D, Bréon F-M, Collins W, Fuglestvedt J, Huang J, et al. Anthropogenic and natural radiative forcing. In: Stocker TF, Qin D, Plattner G-K, Tignor M, Allen SK, Boschung J, et al., editors. Climate change 2013: The physical science basis. Cambridge: Cambridge University Press; 2013. pp. 659–740.
      • 29. Naidu DGT, Bagchi S (2021) Greening of the earth does not compensate for rising soil heterotrophic respiration under climate change. Glob Change Biol 2021;27: 2029–2038.
      • 30. Yan Y, Wang J, Tian D, Luo Y, Xue X, Peng F, et al. Sustained increases in soil respiration accompany increased carbon input under long-term warming across global grasslands. Geoderma 2022;428: 116157.
      • 31. Friedlingstein P, Jones MW, O’Sullivan M, Andrew RM, Hauck J, Peters GP, et al. Global Carbon Budget 2019. Earth Syst Sci Data 2019;11: 1783–1838.
      • 32. Kirschbaum MUF. Soil respiration under prolonged soil warming: are rate reductions caused by acclimation or substrate loss? Glob Change Biol 2004;10: 1870–1877.
      • 33. Bond-Lamberty B, Bailey VL, Chen M., Gough CM, Vargas R. Globally rising soil heterotrophic respiration over recent decades. Nature 2018;560: 80–83.
      • 34. Song X-P, Hansen MC, Stehman SV, Potapov PV, Tyukavina A, Vermote EF, et al. Global land change from 1982 to 2016. Nature 2018;560: 639–643.
      • 35. Ågren GI, Wetterstedt M, Billberger FK. Nutrient limitation on terrestrial plant growth–modeling the interaction between nitrogen and phosphorus. New Phytol 2012;194: 953–960. pmid:22458659
      • 36. Kottek M, Grieser J, Beck C, Rudolf B, Rubel F. World Map of the Köppen-Geiger climate classification updated. Meteorol Z 2006;15: 259–263.
      • 37. Jiang Z, Huete AR, Didan K, Miura T. Development of a two-band enhanced vegetation index without a blue band. Remote Sens Environ 2008;112: 3833–3845.
      • 38. Dornbush ME, Raich JW. Soil temperature, not aboveground plant productivity, best predicts intra-annual variations of soil respiration in central Iowa grasslands. Ecosystems 2006;9: 909–920.
      • 39. Dornbush ME. Grasses, litter, and their interaction affect microbial biomass and soil enzyme activity. Soil Biol Biochem 2007;39: 2241–2249.
      • 40. Dornbush ME, Cambardella CA, Ingham ER, Raich JW. A comparison of soil food webs beneath C3- and C4-dominated grasslands. Biol Fert Soils 2008;45: 73–81.
      • 41. Fisher RF. Amelioration of degraded rain forest soils by plantations of native trees. Soil Sci Soc Am J 1995;59: 544–549.
      • 42. Russell AE, Raich JW, Bedoya R, Valverde-Barrantes O, González E. Impacts of individual tree species on carbon dynamics in a moist tropical forest environment. Ecol Appl 2010;20: 1087–1100. pmid:20597292
      • 43. Sollins PF, Sancho MR, Mata Ch, Sanford RL Jr. Soils and soil process research. In: McDade LA, Bawa KS, Hespenheide HA, Hartshorn GS, editors. La Selva: Ecology and natural history of a Neotropical rain forest. Chicago: University of Chicago Press; 1994. pp. 34–53.
        • 44. Hartshorn GS, Hammel BE. Vegetation types and floristic patterns. In: McDade LA, Bawa KS, Hespenheide HA, Hartshorn GS, editors. La Selva: Ecology and natural history of a Neotropical rain forest. Chicago: University of Chicago Press; 1994. pp. 73–89.
          • 45. Raich JW. Temporal variability of soil respiration in experimental tree plantations in lowland Costa Rica. Forests 2017;8: 40.
          • 46. Raich JW, Valverde-Barrantes OJ (2017) Soil CO2 flux, moisture, temperature, and litterfall at La Selva, Costa Rica, 2003–2010. ORNL DAAC 2017.
            • 47. Martin JG, Bolstad PV Annual soil respiration in broadleaf forests of northern Wisconsin: Influence of moisture and site biological, chemical, and physical characteristics. Biogeochemistry 2005;73: 149–182.
            • 48. Martin JG, Bolstad PV. Variation of soil respiration at three spatial scales: Components within measurements, intra-site variation and patterns on the landscape. Soil Biol Biochem 2009;41: 530–543.
            • 49. Curtis PS, Hanson PJ, Bolstad P, Barford C, Randolph JC, Schmid HP, et al. Biometric and eddy-covariance based estimates of annual carbon storage in five eastern North American deciduous forests. Agr Forest Meteorol 2002;113: 3–19.
            • 50. Desai A. AmeriFlux BASE US-WCr Willow Creek, Ver. 23–5, AmeriFlux AMP (Dataset) 2022: https://doi.org/10.17190/AMF/1246111.
              • 51. Schultz RC, Isenhart TM, Simpkins WW, Colletti JP. Riparian forest buffers in agroecosystems—lessons learned from the Bear Creek Watershed, central Iowa, USA. Agroforest Syst 2004;61: 35–50.
              • 52. Marquez CO, Cambardella CA, Isenhart TM, Schultz RC. Assessing soil quality in a riparian buffer strip system by testing organic matter fractions in Central Iowa, USA. Agroforest Syst 1999; 44:133–140.
              • 53. Tufekcioglu A, Raich JW, Isenhart TM, Schultz RC. Biomass, carbon and nitrogen dynamics of multi-species riparian buffer zones within an agricultural watershed in Iowa, U.S.A. Agroforest Syst 2003;57: 187–198.
              • 54. Raich JW, Kaiser MS, Dornbush ME, Martin J, Valverde-Barrantes O. Data for: Multiple factors co-limit short-term in situ soil carbon dioxide emissions. 2022; https://doi.org/10.25380/iastate.20323674
                • 55. Valverde-Barrantes OJ. Relationships among litterfall, fine root growth, and soil respiration for five tropical tree species. Can J Forest Res 2007;37: 1954–1965.
                • 56. Didan K. MOD13Q1 MODIS/Terra Vegetation Indices 16-Day L3 Global 250m SIN Grid V006. 2015, distributed by NASA EOSDIS Land Processes DAAC, https://doi.org/10.5067/MODIS/MOD13Q1.006.
                  • 57. Larcher W. Physiological plant ecology. 3rd ed. New York: Springer-Verlag; 1995.
                    • 58. Sánchez ML, Ozores MI, López MJ, Colle R, De Torre B, García MA, et al. Soil CO2 fluxes beneath barley on the central Spanish plateau. Agr For Meteorol 2003;118: 85–95.
                    • 59. Huang N, He J-H, Niu Z. Estimating the spatial pattern of soil respiration in Tibetan alpine grasslands using Landsat TM images and MODIS data. Ecol Indic 2013;26: 117–125.
                    • 60. Gy Gelybó, Barcza Z, Dencső M, Potyó I, Kása I, Horel Á, et al. Effect of tillage and crop type on soil respiration in a long-term field experiment on chernozem soil under temperate climate. Soil Till Res 2022;216: 105239.
                    • 61. Potter CS, Klooster SA. Interannual variability in soil trace gas (CO2, N2O, NO) fluxes and analysis of controllers on regional to global scales. Global Biogeochem Cy 1998;12: 621–635.
                    • 62. Wang X, Liu L, Ciais P, Janssens IA, Reichstein M, Peng S, et al. Are ecological gradients in seasonal Q10 of soil respiration explained by climate or by vegetation seasonality? Soil Biol Biochem 2010;42: 1728–1734.
                    • 63. Huang M, Piao S, Ciais P, Peñuelas J, Wang X, Keenan TF, et al. Air temperature optima of vegetation productivity across global biomes. Nat Ecol Evol 2019;3: 772–779. pmid:30858592
                    • 64. Reynolds LL, Johnson BR, Pfeifer-Meister L, Bridgham SD. Soil respiration response to climate change in Pacific Northwest prairies is mediated by a regional Mediterranean climate gradient. Glob Change Biol 2015;21: 487–500. pmid:25205511
                    • 65. Han G, Xing Q, Luo Y, Rafique R, Yu J, Mikle N. Vegetation types alter soil respiration and its temperature sensitivity at the field scale in an estuary wetland. PLOS ONE 2014;9: e91182. pmid:24608636
                    • 66. Chen S, Zhang M, Zou J, Hu Z. Relationship between basal soil respiration and the temperature sensitivity of soil respiration and their key controlling factors across terrestrial ecosystems. J Soil Sediment 2022;22: 769–781.
                    • 67. Kaiser MS, Speckman PL, Jones JR. Statistical models for limiting nutrient relations in inland waters. J Am Stat Assoc 1994;89: 410–423.
                    • 68. Ino Y, Monsi M. An experimental approach to the calculation of CO2 amount evolved from several soils. Jpn J Bot 1969;20: 153–188.
                    • 69. Harris RF. Effect of water potential on microbial growth and activity. In: Parr JF, Gardner WR, Elliott LF, editors. Water potential relations in soil microbiology. Madison: Soil Science Society of America; 1981. pp. 23–95.
                      • 70. Brown AD. Microbial water stress physiology: Principles and perspectives. New York: Chichester; 1990.
                        • 71. Skopp J, Jawson MD, Doran JW. Steady-state aerobic activity as a function of soil water content. Soil Sci Soc Am J 1990;54: 1619–1625.
                        • 72. Mielnick PC, Dugas WA. Soil CO2 flux in a tallgrass prairie. Soil Biol Biochem 2000;32: 221–228.
                        • 73. Wang Y, Amundson R, Niu X-F. Seasonal and altitudinal variation in decomposition of soil organic matter inferred from radiocarbon measurements of soil CO2 flux. Global Biogeochem Cy 2000;14: 199–211.
                        • 74. Falloon P, Jones CD, Ades M, Paul K. Direct soil moisture controls of future global soil carbon changes: An important source of uncertainty. Global Biogeochem Cy 2011;25: GB3010.
                        • 75. Moyano FE, Vasilyeva N, Bouckaert L, Cook F, Craine J, Curiel Yuste J, et al. The moisture response of soil heterotrophic respiration: interaction with soil properties. Biogeosciences 2012;9: 1173–1182.
                        • 76. Silver WL, Lugo AE, Keller M. Soil oxygen availability and biogeochemistry along rainfall and topographic gradients in upland wet tropical forest soils. Biogeochemistry 1999;44: 301–328.
                        • 77. Liptzin D, Silver WL, Detto M. Temporal dynamics in soil oxygen and greenhouse gases in two humid tropical forests. Ecosystems 2011;14: 171–182.
                        • 78. Yan Z, Bond-Lamberty B, Todd-Brown KE, Bailey VL, Li SL, Liu CQ, et al. (2018) A moisture function of soil heterotrophic respiration that incorporates microscale processes. Nat Commun 2018;9: 2562. pmid:29967415
                        • 79. Hursh A, Ballantyne A, Cooper L, Maneta M, Kimball J, Watts J. The sensitivity of soil respiration to soil temperature, moisture, and carbon supply at the global scale. Glob Change Biol 2017;23: 2090–2103. pmid:27594213
                        • 80. Rawls WJ, Pachepsky YA, Ritchie JC, Sobecki TM, Bloodworth H. Effect of soil organic carbon on soil water retention. Geoderma 2003;116: 61–76.
                        • 81. McGowan HA, Lowry AL, Gray MA. Identification of optimum temperatures for photosynthetic production in subtropical coastal ecosystems: Implications for CO2 sequestration in a warming world. J Geophys Res-Biogeo 2020;125: e2020JG005678.
                        • 82. Drobník J. The effect of temperature on soil respiration. Folia Microbiol 1962;7: 132–140.
                        • 83. Parker LW, Miller J, Steinberger Y, Whitford WG. Soil respiration in a Chihuahuan desert rangeland. Soil Biol Biochem 1983;15: 303–309.
                        • 84. Guntiñas ME, Gil-Sotres F Leirós MC, Trasar-Cepeda C. Sensitivity of soil respiration to moisture and temperature. J Soil Sci Plant Nutr 2013;13: 445–461.
                        • 85. Zhou W, Hui D, Shen W. (2014) Effects of soil moisture on the temperature sensitivity of soil heterotrophic respiration: A laboratory incubation study. PLOS ONE 2014;9(3): e92531. pmid:24647610
                        • 86. Liu Y, He N, Wen X, Xu L, Sun X, Yu G, et al. The optimum temperature of soil microbial respiration: Patterns and controls. Soil Biol Biochem 2018;121: 35–42.
                        • 87. Chen B, Liu S, Ge J, Chu J. Annual and seasonal variations of Q10 soil respiration in the sub-alpine forests of the Eastern Qinghai-Tibet Plateau, China. Soil Biol Biochem 2010;42: 1735–1742.
                        • 88. Davidson EA, Janssens IA, Luo Y. On the variability of respiration in terrestrial ecosystems: moving beyond Q10. Glob Change Biol 2006;12: 154–164.
                        • 89. Schleser GH. The response of CO2 evolution from soils to global temperature changes. Z Naturforsch 1982;37: 287–291.
                        • 90. Fierer N, Colman BP, Schimel JP, Jackson RB. Predicting the temperature dependence of microbial respiration in soil: A continental-scale analysis. Global Biogeochem Cy 2006;20: GB3026.
                        • 91. Xu Z, Tang S, Xiong L, Yang W, Yin H, Tu L, et al. Temperature sensitivity of soil respiration in China’s forest ecosystems: Patterns and controls. Appl Soil Ecol 2015;93: 105–110.
                        • 92. Kirschbaum MUF. The temperature dependence of soil organic matter decomposition, and the effect of global warming on soil organic C storage. Soil Biol Biochem 1995;27: 753–760.
                        • 93. Duffy KA, Schwalm CR, Arcus VL, Koch GW, Liang LL, Schipper LA. How close are we to the temperature tipping point of the terrestrial biosphere? Sci Adv 2021;7: eaay1052. pmid:33523891
                        • 94. Phillips CL, Bond-Lamberty B, Desai AR, Lavoie M, Risk D, Tang J, et al. The value of soil respiration measurements for interpreting and modeling terrestrial carbon cycling. Plant Soil 2017;413: 1–25.
                        • 95. Bond-Lamberty B. New techniques and data for understanding the global soil respiration flux. Earth’s Future 2018;6: 1176–1180.