A global inventory of the soil CH4 sink

Authors


Abstract

[1] Methane uptake by soils is a small but important flux in the global budget of atmospheric methane, and could be susceptible to changes in land use and climate. Estimates of this sink vary between 20 and 45 Tg yr−1. We propose to develop a better constrained estimate using a mechanistic understanding of the biogeochemical controls of soil CH4 uptake. We reviewed over 120 published papers reporting field measurements of CH4 uptake and made over 318 annual estimates of CH4 uptake for various types of ecosystems. We collected data from these papers for a number of parameters that are known to influence the magnitude of the sink including climatic zone, ecosystem, latitude, annual mean rainfall, annual mean temperature, and the soil texture. Regression analyses with the continuous variables (latitude, rainfall, and temperature) yielded results with poor predictive ability and no significant relationship. Stratification by class variables such as climatic zone, ecosystem type and soil texture provided better predictive ability (R2 = 0.29, P < 0.0001). The mean largest uptake rates were observed in temperate forests with coarse soil texture, but the variance within this stratum was also large. Without any stratification, we estimate that the global soil CH4 sink is 36 ± 23 Tg yr−1. With stratification, the best current estimate of the global soil uptake of CH4 is 22 ± 12 Tg yr−1. The ecosystem type accounted for the largest part of the variation in the global data set. This inventory showed that ecosystem type, geographic zone, and soil texture strongly control CH4 uptake. Inventory methods that take into account underlying factors that control the process provide better estimates of sink strength.

1. Introduction

[2] Understanding the sources and sinks of atmospheric trace gas constituents is important for understanding anthropogenic impacts on the radiative balance of the atmosphere. Many of the budgets of these gases are poorly constrained, which limits our ability to understand the dynamics of these gases. The atmospheric concentration of methane (CH4) has doubled since pre-industrial times. The current concentration is ∼1780 ppbv and, until recently, the concentration was increasing. The rate of increase in the concentration of atmospheric CH4 slowed from about 15 ppbv yr−1 in the 1980s to near zero in 1999 [Dlugokencky, 2001]. Since 1990, the annual rate of CH4 increase in the atmosphere has varied between less than zero and 15 ppbv. The reasons for this change are not known.

[3] Soils both produce and consume CH4. The net soil–atmosphere CH4 flux is the result of the balance between the two offsetting processes of methanogenesis (microbial production) and methanotrophy (microbial consumption). Methanotrophs and methyltrophs are all obligate aerobes; the biochemical process requires a monooxygenase enzyme and therefore, requires molecular O2. Methanogenesis is the process of microbial production of CH4 in anaerobic conditions. Methanogenesis is an important process in wetland soils and rice paddies and these systems are usually sources of CH4 for the atmosphere. However, methanogenesis can also occur in upland soils, inside soil aggregates where anaerobic ‘microsites’ occur. Methanotrophy is the dominant process in upland soils, where oxidation generally exceeds production and there is a net uptake by the soil of CH4 from the atmosphere. Methanotrophy is also an important process in wetland soils at the aerobic soil-water interface. One estimate of the importance of soil methanotrophy suggests that as much as 50% of the CH4 produced in soils and sediments is consumed therein [Reeburgh et al., 1993].

[4] In the global CH4 budget soils are the largest biotic sink for atmospheric CH4, consuming 15–45 Tg annually, a rate that is of similar magnitude as the rate of CH4 accumulation in the atmosphere during the 1990s [Potter et al., 1996; Ridgwell et al., 1999]. Thus any significant change made to the soil CH4 sink could alter the net biosphere–atmosphere flux and alter the atmospheric accumulation rate of this potent greenhouse gas. Ojima et al. [1993] suggested that there has been a significant increase in the soil sink as atmospheric mixing ratios of CH4 have increased. At the same time, land use change and expansion of agriculture has significantly reduced the strength of the soil sink [Potter et al., 1996; Smith et al., 2000]. Ojima et al. [1993] calculated that the temperate forest and grassland sink for atmospheric CH4 has been reduced by 1.5 to 6 Tg CH4 yr−1. The changes in sink strength associated with land use change are of an order of magnitude such that they are affecting the atmospheric accumulation rate of this important greenhouse gas.

[5] Several general reviews have recently discussed the role of soils as source and sink for atmospheric CH4 [Conrad, 1989; Nedwell, 1996; Topp and Pattey, 1997; Le Mer and Roger, 2001]. During the past fifteen years, the number of published papers reporting on soil CH4 uptake has grown. This work has shown that variability in the methane uptake results from the interactions of a number of factors that regulate methane oxidation in the soil, for example climate, soil physical properties like water content and porosity, and biological variables like the nature of the soil microbial community. Among these factors, soil physical properties are likely to have the greatest influence on the soil CH4 sink at the regional to global scales [Potter et al., 1996; Striegl et al., 1992; Verchot et al., 2000; Del Grosso et al., 2000; Borken and Beese, 2006]. Low CH4 oxidation rates are consistently associated with fine textured soils that have low porosity and high water retention, which combine to restrict diffusion of CH4 into the soil profile. Coarse texture soils tend to drain rapidly and have high porosity and therefore high diffusion rates of CH4 into the soil profile. Several authors [Verchot et al., 2000; Kiese et al., 2003] noted that in environments with seasonal rainfall patterns, CH4 uptake closely followed the pattern of rainfall, increasing rapidly during the dry season as soil water content declines. Other factors also affect uptake of CH4 by soils. For example, cold winter temperatures inhibit microbial activity over the large landmasses of the northern hemisphere. However, Mosier et al. [1997] have shown that frozen soils continue to consume CH4 at a reduced rate.

[6] The first objective of this paper is to use this new data to reevaluate the global inventory of CH4 uptake by soils. Extrapolation estimates of greenhouse gas fluxes are often done by surveying the literature, averaging means from a number of studies, and multiplying by the area covered by the biome or ecosystem [Potter et al., 1996; Mosier et al., 1998; Smith et al., 2000]. However, there is a tremendous amount of variation within a biome and we understand enough about the controlling factors that we ought to be able to use them reduce the variance of the means within the extrapolation units. Thus our second objective is to examine the potential for explaining variability across studies in the literature using parameters that are often measured and that have mechanistic significance.

[7] In this study, we looked at a number of factors that are likely to explain large-scale variability in CH4 uptake by soils. Latitude is a good proxy for the length of the growing season or of the frost-free season. Seasonal variations in precipitation also correspond to variations in temperature and these two factors work in concert to affect fluxes. Thus parameters such as mean annual precipitation and mean temperature can be used as indicators of climate controls on microbial processes as well as on diffusion across the soil—atmosphere boundary. Ecosystem type also affects the variability in the CH4 uptake. Boeckx and Van Cleemput [2001] noted that CH4 uptake was higher in forest soils than in grasslands and arable soils.

[8] We used the relationships developed from this analysis to explore different stratification schemes for estimating the global soil CH4 sink. We judged the success of this approach through the reduction in variation within strata compared to more aggregated stratification schemes.

2. Methods

[9] We compiled a database from 120 studies that reported CH4 fluxes from different terrestrial biomes (auxiliary material). We included data from studies that reported soil texture, annual rainfall, mean temperature and latitude coordinates. Few studies were designed to estimate annual uptake of methane by soils. Some reported fluxes during more than one season, but several reported fluxes only during a brief period, which was often during the summer in temperate regions. Whenever the authors provided their own extrapolation to an annual estimate, we use their estimate. Where authors provided no annual extrapolations, we used our best judgment to make an estimate of the annual total.

[10] The method for extrapolation varied with the climatic zone, as we took into account seasonal variations. Generally, these studies presented data that were complete enough to calculate the total flux from the site for the frost-free season. However, Mosier et al. [1997] working in alpine tundra ecosystems showed that even in frozen soils CH4 fluxes were maintained, albeit at lower rates than during frost-free seasons. Thus, when studies reported fluxes only for the frost-free part of the year, we assumed that these fluxes represented 70% of the annual flux. Therefore, depending on the latitude, we calculated the CH4 sink for the measurement period between 100 and 365 days, and for ecosystems where soils froze, we assumed that this value represented 70% of the annual flux.

[11] All CH4 uptake fluxes were expressed in terms of kg CH4 ha−1yr−1. When the annual mean temperature, annual mean rainfall or latitude data were missing, we used globally gridded data sets available on the worldwide web: www.weatherbase.com and www.mappoint.msn.com, according to the site description provided in the paper.

[12] We stratified the data into different biomes according to the Leemans' classification [Leemans, 1990]. Many papers did not give the type of forest, so we grouped several of Leemans' categories together: tundra forest with boreal forest, cold temperate forests with warm temperate forests, and tropical seasonal forests with rain forests. We also grouped cold and hot deserts together because of their low biological activity and because of the lack of data. The soil texture data were used to classify soils into in four texture classes: organic, fine, medium and coarse, according to the USDA classification system [Brady, 1974].

[13] We made over 310 estimates of annual CH4 uptake by soils (auxiliary material). This number differs from the number of published papers because several papers reported fluxes for more than one type of ecosystem or for more than one crop. In a few cases, we combined estimates from more than one published paper for the same study area to estimate one single annual flux for the site. We analyzed the data to explore the relationships between CH4 flux and environmental or physical variables through both regression analysis and analysis of variance using SAS statistical software [SAS Institute, 2005]. The data set was not normally distributed and even with a logarithmic transformation, the data set deviated significantly from a normal distribution, albeit to a lesser degree than untransformed data. Thus, to reduce heteroskedasticity and to improve the distribution, we conducted all statistical analyses with transformed data. All extrapolations were based upon the arithmetic mean rather than the geometric mean, as the arithmetic mean is the center of probability of a distribution and is therefore parameter of choice for extrapolation [Parkin and Robinson, 1994].

3. Results

3.1. Adequacy of the Data Set

[14] The data set was extensive and covered the major ecosystems of the world. We note however, that the distribution of the sites was uneven, with high concentration of sites in eastern North America, Western Europe and northern South America (Figure 1). Large areas of Africa, Asia and southern South America had no measurements. Additionally, measurement density in the taiga, tundra and deserts was low.

Figure 1.

Location of study sites used for this analysis (GIS Unit, ICRAF).

[15] According to classification scheme of Leemans [1990], which was also used by Potter et al. [1996] for their global extrapolation, the land area of the globe can be divided into 14 ecosystem types. We present these ecosystem types in Table 1 along with the number sites for which we made estimates of annual CH4 uptake and the mean uptake rates. We note that the sum of the area data in Table 1 is superior to those used in other CH4 sink estimates in this paper. In later estimates, we removed lithosols from the area as we expect that the CH4 uptake in these soils is nil. Additionally the FAO classification lists 30% of the soils as undefined for tundra, so stratification based upon texture is impossible. Therefore the tundra area that we used for our estimate was only 0.05 × 106 km2 rather than 9.41 × 106 km2.

Table 1. Summary of the Data Set Used for the Analyses in This Paper by Biome
Ecosystem TypeaTotal Land Area,b × 106 km2Number of Study SitesMean, kgCH4 ha−1 yr−1Variance
Tundra9.41111.495.52
Forest 512.648.06
  Forest tundra8.18   
  Boreal forest15.31   
Desert 51.101.03
  Cool desert2.63   
  Hot desert19.29   
Grasslands2.61292.322.92
Cultivationc24.79471.231.22
Chaparral2.1732.250.54
Forest 925.7031.50
  Cool temperate forest5.07   
  Warm temperate forest1.69   
Tropical steppe/savanna 181.491.54
  Tropical semiarid steppe7.98   
  Tropical savanna11.50   
Tropical forest 623.334.82
  Tropical seasonal forest12.83   
  Tropical rain forest7.17   
Total130.63183.2914.80

[16] Forests, with 205 sites, represented more than 65% of the study sites. Over 29% of these sites were from temperate forests; 19% were from tropical forest; and 16% were from boreal forests. Among the remaining sites, 42% were cultivated. Desert, tundra, tropical steppes, savannas, and chaparral were all under represented in the data set. The variance in the temperate forest was very high, but in the other ecosystems, they were of similar order of magnitude as the mean. Temperate forests were very well represented in the literature and very well distributed around the world, with measurements at 92 different sites by 40 different authors.

[17] CH4 uptake rates varied between −0.04 and −49.93 kg (CH4) ha−1yr−1. The highest consumption rates were measured by Singh et al. [1997] in savanna and tropical forest ecosystems. Excluding these very high rates, the observed fluxes range −0.04 to −27.74 kg (CH4) ha−1yr−1. There is no obvious explanation for such high average fluxes, as compared with other CH4 fluxes from same ecosystems (auxiliary material) even if they are representative of particular situations. Thus they were not considered in the statistical analysis.

3.2. Analysis of Drivers at the Global Scale

[18] Among the explanatory variables that we collected from the publications, several were continuous variables, which lend themselves to regression analysis; others were categorical variables, which lend themselves to analysis of variance. For regression analyses on the continuous variables, we found a marginally significant relationship between CH4 and latitude (P = 0.0679) and significant relationships with annual rainfall (P = 0.0016) and mean annual temperature (P = 0.0226). However, these relationships had low predictive power: rainfall explained about 3% of the variation in the global data set; annual temperature explained 2%; and latitude explained 1%.

[19] For the analysis of the categorical variables, we began by stratifying the global data set into three zones: boreal, temperate and tropical. The stratification by zone only explained 2% of the variation, but was significant (P = 0.0296) and the boreal, temperate and tropical soils had a mean uptake of 2.38, 3.96 and 2.74 kg (CH4) ha−1yr−1, respectively (Figure 2). The boreal zone was significantly different from the temperate and tropical zones (P < 0.05), while the tropical and temperate zones were not different from each other. The temperate zone had the highest number of observations and the highest variability. All of the zones have several outliers, but the temperate zone showed the greatest variability with the greatest number of outliers. The coefficient of variation for the temperate zone was 120%, compared to 115% for the boreal zone and 77% for the tropical zone.

Figure 2.

Data stratified by climatic zone. Box and whisker plot of CH4 emissions in different climatological zones. Boxes show 25th and 75th percentiles, and the median is indicated as a solid line within the boxes. The whiskers show the 10th and 90th percentiles. Solid circles indicate outliers; the dotted lines indicate the means.

[20] We then stratified the global data set by ecosystem type. Analysis of variance indicated that there was a major ecosystem effect, so we stratified the data set into forest and nonforest (Figure 3). This stratification scheme explained 12% of the global variation and was highly significant (P < 0.0001). The average uptake for forests in the global data set was 4.22 kg (CH4) ha−1yr−1, while that of the other ecosystems was 1.60 kg (CH4) ha−1yr−1. The forest had the highest number of observations (205) compared to the other ecosystems (113) and the highest variability. The coefficients of variation were 104% in the forest and 93% for all other ecosystems combined.

Figure 3.

Data stratified by ecosystem type. Box and whisker plot of CH4 emissions in different ecosystems. Boxes show 25th and 75th percentiles, and the median is indicated as a solid line within the boxes. The whiskers show the 10th and 90th percentiles. Solid circles indicate outliers; the dotted lines indicate the means.

[21] Finally, we stratified the global data set by soil texture. Coarse textured soils had greater uptake than the other texture classes (Figure 4). The stratification by texture explained 1.5% of the variation, but was not significant (P = 0.2043). Coarse texture soils had mean uptake of 4.19 kg (CH4) ha−1yr−1, while fine and medium texture soils had uptake of 1.98 and 3.31 kg (CH4) ha−1yr−1, respectively. Organic soils, which occurred primarily in the boreal zone, had a mean flux of 3.05 kg (CH4) ha−1yr−1. The coefficients of variation were between 69 and 128% for the texture classes and increased with the number of observations.

Figure 4.

Data stratified by soil texture class. Box and whisker plot of CH4 emissions in soils of different texture. Boxes show 25th and 75th percentiles, and the median is indicated as a solid line within the boxes. The whiskers show the 10th and 90th percentiles. Solid circles indicate outliers; the dotted lines indicate the means.

3.3. Estimates of the Global Sink

[22] We calculated the effects of using each of these variables to stratify the global data set on the estimated global soil CH4 sink (Table 2). For comparison with other stratification schemes, we also developed an estimate based upon the unstratified data set (Table 2). The unstratified estimate of 36 Tg yr−1 was greater than the stratified estimates. Stratification by climatic zone or soil texture gave a negligible reduction in the estimate, while stratification by ecosystem reduced the global estimated sink strength. In each one of the stratification schemes, the largest part of the variation was captured in one stratum. Stratification by climatic zone indicated that the highest variability was in the temperate zone, while stratification by ecosystem showed high variability in forest ecosystems. Medium texture soils had the highest variability in the texture class stratification. For each stratification scheme, the highest variability was associated with the category of the highest flux rate. A double stratification scheme shed further light on these results (Table 2). This scheme explained 17% of the variation (P < 0.0001) and further reduced the global estimate to 25 Tg yr−1. The temperate forest class captured the highest degree of variability.

Table 2. Estimates of CH4 Uptake by Soils With Different Stratification Schemesa
Stratification LevelNumber of MeasuresMean, kgCH4 ha−1 yr−1VarianceArea, × 106 km2Total Flux, Tg yr−1SE of Total Flux
  • a

    The first estimate shows mean and standard error estimates from the data set without stratification. The global data set is then stratified by zone or ecosystem or by soil texture individually, and then by zone + ecosystem.

None3183.314.8108.2035.623.3
Climatic zone
Boreal652.47.520.835.07.1
Temperate1634.022.650.0719.818.6
Tropical902.74.437.3010.28.3
Total    35.021.6
Ecosystem
Forest2054.219.342.0117.712.9
Other1131.62.266.1910.69.3
Total    28.315.9
Soil texture
Organic543.19.22.140.70.9
Coarse854.221.122.739.511.3
Medium1213.318.063.9421.124.6
Fine462.01.819.393.83.9
Total    35.227.4
Climatic zone + ecosystem
BorealForest512.68.016.874.56.7
 Other141.44.33.960.62.2
Subtotal (zone)    5.1 
TemperateForest925.731.55.823.33.4
 Other711.72.344.257.67.9
Subtotal (zone)    10.9 
TropicalForest623.34.819.326.45.4
 Other281.41.217.982.63.7
Subtotal (zone)    9.0 
Total    25.012.9

[23] We ran an ANOVA with a full model for these three variables with interaction terms and the only significant main effect was ecosystem (forest versus other). Two-way interactions terms were all significant: climatic zone × ecosystem type (P = 0.0707); ecosystem type × texture (P = 0.0397); and climatic zone × texture (P = 0.0350). The three-way interaction term was not significant. We then tested a number of reduced models and found that a model based on the ecosystem type plus the three two-way interaction terms, provided the best alternative model, where all of the effects in the model were significant at the P < 0.05 level. This model explained 29% of the variation and was highly significant (P < 0.0001). The presence of significant interaction terms suggested that the best global estimate from this data set could be obtained from a full stratification by the three main categorical variables. On the basis of this analysis, we developed a global estimate of the soil CH4 sink using a full stratification (Table 3).

Table 3. Estimates of CH4 Uptake by Soils With a Zone Plus Ecosystem Plus Soil Texture Stratification Scheme
Stratification LevelNumber of MeasuresMean, kgCH4 ha−1 yr−1VarianceArea, × 106 km2Total Flux, Tg yr−1SE of Total Flux
Climatic zone + ecosystem + soil texture
BorealForestOrganic283.610.01.760.61.1
  Coarse63.27.63.061.03.4
  Medium140.50.211.180.61.4
  Fine12.00.870.2
Sub-subtotal (zone + ecosystem)   2.4 
BorealOtherOrganic81.77.70.050.00.1
  Coarse31.00.11.080.10.2
  Medium31.00.22.770.30.7
  Fine0.06
Sub-subtotal (zone + ecosystem)     0.4 
Subtotal (zone)     2.8 
TemperateForestOrganic74.610.30.080.00.1
  Coarse307.534.01.000.71.1
  Medium435.637.13.602.03.4
  Fine122.31.51.140.30.4
Sub-subtotal (zone + ecosystem)     3.0 
TemperateOtherOrganic91.41.50.100.00.0
  Coarse181.71.710.351.83.2
  Medium361.72.128.244.96.8
  Fine61.24.65.560.74.9
Sub-subtotal (zone + ecosystem)     7.4 
Subtotal (zone)     10.4 
TropicalForestOrganic23.60.120.0
  Coarse164.36.82.431.01.6
  Medium193.83.89.323.64.2
  Fine212.11.57.451.62.0
Sub-subtotal (zone + ecosystem)     6.2 
TropicalOtherOrganic0  0.03  
  Coarse120.91.44.810.41.6
  Medium62.30.68.832.02.7
  Fine61.51.14.310.61.9
Sub-subtotal (zone + ecosystem)     3.0 
Subtotal (zone)     9.2 
Total for the full stratification     22.412.1

[24] Our best estimate of the global soil sink is 22 Tg yr−1. It is impossible to calculate properly the uncertainty around this value because we do not know the uncertainty associated with the area estimates. If we account only for the uncertainty associated with the flux to calculate a standard error, the range of uncertainty would be between 10 and 34 Tg yr−1 for the full stratification.

4. Discussion

4.1. Global Drivers of CH4 Sink Strength

[25] The single factor analyses indicated that the most important global driver of CH4 uptake rates is the type of ecosystem. Other authors have noted similar differences in temperate and tropical soils [Del Grosso et al., 2000; Verchot et al., 2000]. Uptake rates were consistently higher in forests compared to all other ecosystems across all climatic zones. Forest soils differ from the soil of other ecosystems by the presence of a forest floor. As a result, the upper layers of the soil profile have high organic matter content, high microbial biomass, low bulk density and high porosity. Thus one would expect high consumption rates in these soil profiles. Biological activity is high, and with low bulk density, diffusion of atmospheric CH4 into the forest soil is rapid. This data set is not adequate to provide more mechanistic answers, but perhaps the high variances in forest ecosystems are due to factors associated with the nature of the forest floor, which is largely determined by the floristic composition of the forest ecosystem and by other biotic factors such as earthworm activity [Finzi et al., 1998; Finzi and Schlesinger, 2002; Carreiro et al., 2000; Verchot et al., 2001].

[26] The fact that soil texture accounted for an important part of the variation in the global data set is consistent with current thinking on the biogeochemical controls of CH4 oxidation in soils [Striegl, 1993; Dörr et al., 1993; Del Grosso et al., 2000]. Verchot et al. [2000] summarized the data for forest sites in the neotropics and noted that fine-textured soils generally consumed between 1.5 and 2.0 kg (CH4) ha−1yr−1, while medium and coarse textured soils consumed >4.0 kg (CH4) ha−1yr−1. Other authors have shown that coarse textured soils have higher oxidizing capacity than fine textured soils [see, e.g., Boeckx et al., 1997]. As Striegl [1993] noted, because gas transport of O2 and CH4 are the most important factors affecting CH4 fluxes, and soil texture strongly affects diffusivity of gases within soils, Verchot et al. [2000] concluded that texture class was more important than other factors at the biome scale.

[27] Coarse texture soils make up about 21% of the world's soils and contribute to only 22 percent of the global sink. Thus, despite high CH4 uptake rates, their contribution to the global sink is proportionately low. Medium texture soils make up about 59% of the world's soils and contribute 60% of the global sink. Organic soils are currently under represented in the global data set. The available data suggest that these soils make an important contribution only in the boreal zone. Given the small area that these soils occupy in other climatic zones, it is unlikely that this view will change with more data from these zones.

[28] For the full stratification, temperate forests with coarse soils had the highest uptake CH4 values (7.5 kg CH4 ha−1 yr−1), and a high variance. This value is higher than the median range of 1.6−6.4 kg CH4 ha−1 yr−1 suggested by Smith et al. [2000], although our value was based upon the mean rather than the median. Interestingly, the soil texture effect seems to be important only in temperate forests, which explains the interaction terms in the full rank model. In other ecosystems, CH4 uptake on coarse texture soils is similar to uptake rates in other texture classes.

[29] Seasonal patterns of soil water content closely follow the pattern of rainfall and several authors have noted that CH4 uptake also follows this pattern [Verchot et al., 2000; Kiese et al., 2003]. A simplistic extension of this observation suggests that annual rainfall might be a global predictor of CH4 uptake because soils are drier for longer periods in low rainfall areas. This of course ignores the complex interaction between soil, vegetation type, and water availability on the size and nature of the soil microbial community. We did find a significant linear relationship between CH4 uptake rates and total annual rainfall, but the relationship had low predictive value and was positive. It appears that this relationship is not explaining the effects of physical restriction of water in soils on diffusion, but is an indicator of the biological controls of the soil microbial community on uptake.

4.2. Success of Stratification

[30] The question that remains is whether the stratification that we used was successful in providing a better estimate of the sink strength. Stratification reduced the variance in the majority of strata each time. In the stratification by climate zones, the variances in the boreal and tropical strata were lower than the global variance, but variance in the temperate stratum was larger than the global variance. In the stratification by climatic zone + ecosystem, the variances were lower in all strata except for the temperate forest.

[31] Thus stratification had a significant effect on the estimate of the magnitude of the terrestrial sink strength by allowing us to compartmentalize the largest source of variation. More importantly, stratification compartmentalized the variance and provided narrower standard error ranges. In this sense, the stratification has been successful in producing a more realistic estimate by accounting for some of the variation among biomes. In Table 3, for example, subtotals for the total temperate zone flux was 10.4 Tg yr−1, which was much lower than the estimate of 19.8 Tg yr−1 derived from the simple climatic zone stratification (Table 2). When there was no large source of variation within a climatic stratum, the nature of the substratification had little bearing on the estimate. For example, the total flux estimated for the tropical zone in the simple climatic zone stratification is similar to that estimated by the climatic zone + ecosystem + soil texture stratification.

4.3. Comparison With Other Estimates

[32] Estimates of the global soil sink for atmospheric CH4 have been made by a number of authors and fall between 17 and 44 Tg yr−1 (Table 4). These efforts have involved both extrapolation of field studies and modeling. The modeling efforts have attempted to capture the effects of key biogeochemical drivers on soil CH4 consumption, while most of the stratify-and-multiply approaches have focused on climatic zone and ecosystem effects. The one exception to this is the estimate of Dörr et al. [1993], who based their extrapolation on soil texture stratification. These authors used the results of intensive measurements made at five sites in southern Germany and spot measurements at a wide range of sites in Central Europe and in different parts of the world to calibrate the relationship between texture and CH4 consumption. They produced an estimate of the global soil sink assuming that this parameterization was valid for soils across the globe.

Table 4. Estimates of the Global Soil Methane Sink From Different Sources Compared With Our Estimation Obtained Without and With Stratification
ReferenceCH4 Sink, Tg CH4 yr−1
From modeled estimates
Potter et al. [1996]17
Ridgwell et al. [1999]38
From extrapolation estimates
Dörr et al. [1993]29
Mosier et al. [1998]44
Potter et al. [1996]21
Smith et al. [2000]29
Intergovernmental Panel on Climate Change [2001]30
Our estimate without stratification36
Our estimate with full stratification22

[33] Other authors have produced estimates of the global soil sink. Mosier et al. [1998] derived global CH4 uptake from the references given by Reeburgh et al. [1993] and the land use data of Bouwman [1990]. They estimated CH4 uptake rates for 10 land cover types. A major difference between their estimates and ours is the difference in land cover definitions between the sources of land cover data. Smith et al. [2000] derived an estimate based upon data from 109 sites and estimated that while the mean was 29 Tg yr1, the range of uncertainty was very large, between 7 and 109 Tg yr−1. These authors used the standard deviation of the transformed data to estimate the range of uncertainty associated with the estimate rather than standard error of the mean, which explains the wide range in their reported uncertainty compared to our narrower range. The standard error would have provided a more realistic uncertainty estimate.

[34] Potter et al. [1996] made both a modeling estimate and an extrapolation estimate. For the extrapolation estimate, forest tundra was not included because it had been characterized in the literature as a variably wet environment where high methane production can occur during thaw season, resulting in net annual CH4 emissions to the atmosphere. Data availability was also a problem for this extrapolation estimate. For six of the 14 major ecosystems where they produced an estimate, three or fewer site data sets could be used for the extrapolation to annual fluxes. Over one half of the measured flux data sets came from temperate sites. Making an estimate for tropical grasslands and savannas was also problematic, owing to variability in soil condition associated with fires and human management, and the fact that several authors reported that CH4 emission was present at some times during the year.

[35] The model described by Ridgwell et al. [1999] also takes a mechanistic approach to determine the global consumption of CH4 by soils incorporating descriptions of the most important controlling factors related to both gaseous diffusion and microbial activity. Ridgwell's model predicted that the highest uptake rates would occur in dry tropical ecosystems, whereas the experimental data summarized here suggests that forest ecosystems have the highest uptake rates.

5. Conclusion

[36] The increase in available data from around the world over the past several years has allowed us to produce a more well constrained estimate of the global soil CH4 sink than has been possible previously. We used 318 data representative of a wide variety of ecosystems, climates and soil types, and produced a number of estimates based upon different stratification schemes that cover the lower to middle of the range of the other estimates. Our estimate of 22 ± 12 Tg yr−1 CH4 is toward the lower end of the accepted range for the soil sink.

[37] Just as for process based modeling estimates, this study shows that incorporation of mechanistic elements into the stratification scheme improves the global estimate and reduces uncertainty in extrapolations from field data. Without stratification, the range of the global estimate was between 12 and 60 and the best simple stratification scheme produced a range of 14 to 57. The standard errors of the different strata point to the opportunities to improve further the estimates of the soil sink. In the boreal zone, the standard error is highest in forests with coarse texture soils and the number of observations on these soils was low. In the temperate and tropical forests, highest standard errors were associated with medium texture soils. Both of these strata have high flux rates and there were a large number of observations, which points to the need for more targeted mechanistic work. Nonforest ecosystems are not well represented in the data set and, at least in the temperate zone, the standard errors associated with these estimates are relatively high. Clearly more observations in these situations are warranted. Thus efforts to better account for the variability within these strata will make the greatest contribution to improving the global sink estimate.

[38] This study points to a number of gaps in the global data set. One of the key findings is that a number of important ecosystems continue to be underrepresented in the global data set. In particular, grasslands in both the temperate and tropical zones are underrepresented, as are chaparral ecosystems. Desert ecosystems are also underrepresented, and although fluxes are low in these ecosystems, they represent a large portion of the earth's surface. Thus efforts to fill these gaps should be a high priority. Perhaps highest priority should be given to grasslands. Any assessment of the global soil sink may be seriously in error until more representative data for the whole land surface are available.

[39] Finally, this study raises questions about the sources of variability in temperate forests soils. The effects of soil texture were most pronounced in temperate forests and high uptake rates were associated with high variances. We do not understand the source of the large degree of variability in these soils, and more mechanistic work to elucidate the underlying processes will be helpful.

Acknowledgments

[40] Support for this research was provided by the International Centre for Research in Agroforestry. We would also like to thank two anonymous reviewers who provided comments on an earlier draft of this paper.

Ancillary