Long-term carbon burial in European lakes: Analysis and estimate



[1] Sediment accumulation in lakes provides a small but permanent carbon sink. To date, global estimates of the C cycle have barely considered variations in lake carbon burial. To improve the understanding of carbon storage in lakes this study analyzed the sedimentary record of 228 European lakes concerning long-term carbon burial and its correlation to lake and catchment properties. The results suggest that carbon mass accumulations in small lakes are significantly lower than those used for global estimates so far. On the other hand, the total surface area of small lakes has been severely underestimated. Results from calculations based on a Pareto distribution show that total lake surface is 240,000 km2 in Europe. We estimate total C burial in European lakes at 1.25 Mt yr−1. Half this storage takes place in boreal lakes of northern Europe, although they contribute up to 65% to the European lake surface. This is due to generally lower carbon burial rates in this region. Carbon mass accumulation rates increased in many lakes between 5000 to 2000 years BP. This coincides with increased clastic inputs due to land use change, i.e., increasing cropland coverage and soil erosion. On average, carbon accumulation rates are twice as high in younger sediments at 20 cm depth when compared to the long-term mean.

1. Introduction

[2] The freshwater system is an important transfer point within the global carbon cycle connecting atmosphere, hydrosphere, geosphere and biosphere. Carbon dioxide is withdrawn from the atmosphere by weathering and biomass production and subsequently transported by rivers in solid or dissolved form. Major fractions of carbon are rapidly returned to the atmosphere by degassing of carbon dioxide and methane via the water surface or are transported to the ocean. In contrast, carbon stored in lake sediments is removed from the short-term carbon cycle and presumably buried over a long period [Einsele et al., 2001].

[3] Although total lake area is small compared to other carbon sinks, the substantial amounts of carbon locked and permanently stored in their sediments justify further research. Previous global estimates [Dean and Gorham, 1998; Einsele et al., 2001; Mulholland and Elwood, 1982] are very vague and only differentiate between large and small lakes with low and high C sequestration rates, respectively. In particular, the large variations in small lake carbon accumulation were not considered. More recent and detailed regional studies [Campbell et al., 2000; Kortelainen et al., 2004] suggest that carbon accumulation in lakes of the boreal zone is generally lower, in particular for small lakes. Furthermore, the global studies considered the uppermost parts of the sediment, where degradation processes are still active, rendering these unsuitable for estimating long-term burial.

[4] In addition to the uncertainties concerning the amount of carbon locked in lake sediments, regional and global lake coverage also remains a subject of discussion. Recently published data suggest that global lake surface may be underestimated by a factor of two [Downing et al., 2006]. Particularly the fraction of small lakes is largely undervalued. This means that there are two reasons why the role of carbon burial in small lakes remains unresolved. Neither the variations in C accumulation nor the total lake area has been reliably quantified until now.

[5] To allow more robust estimates on a global scale it is necessary to identify the main factors controlling C accumulation and integrate these into extrapolation models. For this reason this study examines the relation between carbon accumulation and lake and catchment properties in Europe and uses the results to estimate total European carbon burial in lake sediments. Europe was chosen as a supraregional investigation area because of high variability of lake types, climatic conditions and anthropogenic influences and the existence of a relatively good database. Europe is in many aspects comparable to northern America and parts of Asia and therefore the results should be applicable for a significant portion of global lake area.

2. General Approach

[6] The main approach is to identify the relationship between C burial in lakes and lake and catchment characteristics, which can be assigned to other lakes with unknown C accumulation. The results are used for a supraregional estimate of long-term carbon burial in European lake sediments. The following procedure was used:

[7] 1. Estimation of total European lake area derived from analysis of existing lake data sets.

[8] 2. Compilation of C accumulation data for a significant number of lakes (n = 228) located in diverse environments in Europe. Where necessary, accumulation rates were calculated from carbon content and sedimentation rate using dry bulk density estimates.

[9] 3. Determination of lake and catchment characteristic with potential control on C burial for the investigated lakes. For this purpose, a GIS database comprising relevant high-resolution spatial data covering entire Europe was established.

[10] 4. Identification of main factors controlling C burial via statistical analysis and generation of a suitable model equation.

[11] 5. Transfer of main control factor parameters to lakes with unknown C accumulation and subsequent application of the model equation toward obtaining an estimate of total European C burial in lakes.

3. Total European Lake Area

[12] An accurate evaluation of C burial in European lakes requires precise estimates of the total European lake area. To date there is no existing complete register of European lakes. Especially the number and surface area of small lakes is largely unknown. Previous estimates suggest that large lakes dominate the global lake area. Assumed total lake surface estimates range from 2 to 2.8 × 106 km2 [Kalff, 2001; Meybeck, 1995; Shiklomanov and Rodda, 2003]. Based on several regional data sets, Downing et al. [2006] proposed that the relationship between the number of lakes and total lake area, including very small lakes, follows a Pareto distribution. They conclude that the fraction of small lakes is greatly underestimated and that global lake area is therefore approximately 2 times larger than previously assumed.

[13] In consequence this necessitates a closer look at the distribution of European lake area. The estimated number of European lakes is given at about 500,000 lakes > 0.01 km2 and only 16,000 lakes > 1 km2 [Kristensen and Hansen, 1995]. Total lake area is not specified. As far as known to the authors the most comprehensive lake data set is the Global Lakes and Wetlands Database (GLWD) by Lehner and Döll [2004]. The GLWD lists about 27,000 lakes > 0.1 km2 covering a total area of 170,000 km2 for Europe.

[14] Lehner and Döll [2004] show that lake distribution can be largely described with the size-frequency function

equation image

where Na is the number of lakes larger or equal than a specific lake size A, and a and b are fitted equation parameters. Following the approach of Downing et al. [2006], this corresponds very well with a Pareto distribution. By computing b and using a complete lake data set for a certain lake size range, lake numbers and respective lake areas can be calculated up to very small lakes.

[15] Five randomly distributed test regions were chosen to examine completeness of the GLWD. Lakes listed in the GLWD were compared to those identified on satellite images. In all cases, we found that lakes > 5 km2 are completely represented by the GLWD.

[16] The Parameter b describes the logarithmic rate of decline in number of lakes with lake area [Downing et al., 2006] or in other words the ratio of large to small lakes. Obviously, this ratio varies significantly within Europe. To calculate this ratio, regions must be defined where b is almost constant. In order to ensure statistical significance, these regions must also contain a sufficient number of lakes > 5 km2.

[17] The estimation is based on the assumption that lake area distribution is consistent in adjacent regions with constant lake density and comparable morphology. Lake density plots were generated from the GLWD using all lakes with a surface area exceeding 5 km2. The evaluation of these lake density plots and combination with geomorphologic attributes, enabled the definition of eight regions for which uniform lake size distribution is assumed (Figure 1). 60% of the European land surface (region 8) has very low lake densities. We did not further subdivide this region, although strong environmental heterogeneities exist. In terms of lake cover and carbon burial, however, the role of this region is almost negligible.

Figure 1.

Division of Europe in regions with assumed continuous lake area distribution. For these regions the appearances of lakes smaller than 5 km2 were calculated separately on the basis of the assumption that lake size-frequency can be described as a Pareto distribution.

[18] Parameter b was computed based on the GLWD lake data for lakes > 5 km2 using a nonlinear regression model (Levenberg-Marquardt algorithm) corresponding to equation (1). The results are shown in Table 1. Knowing b and extracting additional required attributes from the GLWD, the total number of lakes (Nt) exceeding a certain area can be calculated. Equation (2) specifies the fraction (fb) of Nt represented by the canonical data set, i.e., the GLWD for lakes > 5 km2.

equation image

Amax and Amin are the maximum and minimum lake size covered by the data set. In this case Amin is 5 km2 while Amax depends on the lake region in question. Parameter k represents the smallest lake size included in the calculation or rather the smallest lake size for which the Pareto distribution is valid. Because the Pareto distribution could be demonstrated for lakes exceeding 0.1 km2 for all regional data sets investigated by Downing et al. [2006], this threshold was chosen as the smallest reliable value. Knowing Nt, the number of lakes and their average lake area can be computed for every desired lake size class using equations (3) and (4).

equation image
equation image

Using these equations, we calculated the number of lakes and their total area for lakes ranging from 0.1 km2 to 5 km2 for each region and added the results to the GLWD database for lakes > 5 km2.

Table 1. Calculated Coefficients of Equation (1) and Coefficient of Determination for Different European Lake Regions (see Figure 1)a
  • a

    A low coefficient b as in Region 7 (Alps) indicates that the lake size-frequency distribution is dominated by large lakes.


[19] The results of this combined analysis indicate that the total European lake area for lakes > 0.1 km2 is 240,000 km2 consisting of about 300,000 individual lakes (Table 2). These results surpass the total lake area given in the GLWD by 40%. The number of lakes is exceeded by a factor of 10.

Table 2. Lake Number (N) and Total Area (A in km2) of Europe for Different Lake Regionsa
RegionN (GLWD)A (GLWD)N (0.1–5 km2)A (0.1–5 km2)N TotalA TotalA (%)
  • a

    Region 4 (Finland, Karelia, Kola Peninsula, plane parts of Sweden) covers only 10% of the European land surface but contains 50% of the European lake surface. Values for lakes > 5 km2 are taken from the GLWD. Results for smaller lakes are calculated using equations (3) and (4).


[20] As expected, the highest number of lakes and the largest total lake area is found in the northern, formally glaciated regions. Regions 3 and 4 (Scandinavia, northwestern Russia) together comprise only 15% percent of the European land surface but comprise 61% of the total lake area. High limnicity is also found in Region 6 (Eastern European lake districts) with about 3% land surface and almost 9% of the European lake area.

4. Determination of Carbon Mass Accumulation Rates

[21] Any assessment of carbon burial in lake sediments requires knowledge of the carbon mass accumulation rates (CMARs). Unfortunately, most publications report sedimentation rates in mm yr−1. In addition to this, they often lack data on dry bulk density (dbd) and/or porosity. These values could be used to calculate mass accumulation rates. Furthermore, we need a reliable age/depth model [Dean and Gorham, 1998]. Because dbd is commonly not specified, an indirect approach is necessary to estimate the CMAR from available data.

[22] To this end, the empirical relationships between carbon content and dbd from different authors (Figure 2) were examined for the present study. Menounos [1997] found a loge relationship (r2 = 0.89) between dbd and organic carbon content for samples from six alpine lakes. Avnimelech et al. [2001] extended this data set with global distributed samples from different flooded sediments to a total of n = 868 including lake, pond, river and ocean floor sediments. The relationship remained very similar while r2 decreased to 0.7. The organic carbon content of the studied samples was relatively small, reaching only a maximum of 10%. The equation obviously underestimates dbd at higher carbon contents.

Figure 2.

Empirical relationships between OC contents of lake sediments and dry bulk density found by different authors [Avnimelech et al., 2001; Campbell et al., 2000; Dean and Gorham, 1998; Menounos, 1997]. Larger discrepancies are due to the range of considered OC contents. While Avnimelech et al. [2001] and Menounos [1997] focused on examining sediments with low OC content, the relationships found by Campbell et al. [2000] and Dean and Gorham are reliable for sediments with OC content larger than 5%. The relationship given by Campbell et al. [2000] takes sediment depth into account. A fixed depth of 50 cm was selected because the compaction effect is only significant in the upper section of the core.

[23] Dean and Gorham [1998] and Campbell et al. [2000] describe relationships between OC content and dbd for Williams Lake (Minnesota) and Christiana Lake (Alberta), respectively. Both were very similar to those given by Menounos [1997] and Avnimelech et al. [2001]. Williams Lake contains a broad spectrum of OC values up to 38%. While the calculated relationship in the range of 2 to 10% OC corresponds very well to that presented by Avnimelech et al. [2001], the dbd values are overestimated in the very low range. Campbell et al. [2000] present another equation introducing the effect of compaction. Sediment compaction follows a power law and is of particular importance for the uppermost 20 to 30 cm [Campbell, 1996]. The empirical correlation between dbd, carbon content and depth found for Christiana Lake suggests that compaction is insignificant for sediments buried deeper than 50 cm or containing less than 10% organic carbon. With a premised depth of more than 50 cm and more than 5% organic carbon content the presented relationship corresponds accurately to that proposed by Dean and Gorham [1998]. The dbd for organic carbon contents less than 5% is obviously overestimated because samples with low carbon content were not included.

[24] All authors reach very similar results with respect to the extent of validity for the investigated OC range. As a consequence, it seems safe to presume that lake sediment dbd can be generated from measured organic carbon contents. In case of low organic carbon content (<6% OC) we used the relationship

equation image

given by Dean and Gorham [1998]. The dbd for sediments with higher OC content was calculated using the formula

equation image

presented by Avnimelech et al. [2001]. The difference in dbd between both equations at 6% OC of 0.05 g cm−3 is negligible.

[25] This allowed us to estimate dbd from OC data of lake sediments, which are extensively reported in the literature. The corresponding dbd was calculated for every available OC value. Together with published age depth models we were able to calculate CMAR. Plots of CMAR/age were generated to evaluate temporal development of individual lakes. Because ongoing C degradation must be assumed for the uppermost 20 cm, we excluded this part of the sediment section from further analysis of long-term carbon burial.

[26] Dating of Holocene lakes sediments is usually carried out using radiocarbon methods. Typically about five radiocarbon ages per core are available covering approximately 10,000 years. In combination with varved sections, a high accuracy can be achieved. 210Pb chronology is useful for gauging modern lake history and covers only about 150 years. Because mass accumulation values are very sensitive to changes in sedimentation rates, their accuracy is limited. To ensure comparability of the results, mean values were chosen in the following analysis. Although some lake records date back to the late glacial, only the Holocene sediment section was considered. For a more quantitative review of temporal CMAR trends, the most recent value below the degradation zone, i.e., the CMAR at 20 cm depth (CMAR20cm), was extracted. Due to different sedimentation rates this point is not isochronal. In most lakes the CMAR20cm marks an age of around 200 years BP. Hence, CMAR20cm only enables a general view of subrecent C accumulation. For some lakes, only the time of lake formation was used to calculate CMAR long-term means. In this case, changes in sedimentation rates could not be distinguished and no CMAR20cm was applied.

5. Evaluation of Lake and Catchment Properties

[27] In order to characterize lakes with a calculated CMAR value, general lake properties (lakes area, depth etc.) were collected from the literature. In many cases the geographic coordinates required specification using Google Earth®. To avoid confusion, lake shape was compared to maps or sketches in the publication. Thereby all lakes could be clearly identified. To evaluate catchment properties, a European spatial database was established, which includes all data potentially relevant for carbon accumulation in lakes (climate, land use, population, geology etc.). We used ArcGis® Software with extensions for spatial analysis. To achieve comparability and applicability for later extrapolation, the spatial data covered the entire area of Europe with a minimum resolution of one square kilometer.

[28] We calculated lake catchment areas using a DEM with a resolution of about 90 m from the Shuttle Radar Topography Mission (SRTM). In some cases, catchment areas were adopted from the literature. Because of the limited resolution of the spatial data only lakes > 5 km2 were chosen for catchment analysis. The collected information concerning CMAR, lake and catchment properties was subsequently used for statistical analysis using Statistica® Software. To identify the main controlling factors of CMAR we applied single and stepwise multiple linear regression analysis. Some variables were loge transformed to improve the normality of their distribution.

[29] Matrix plots were analyzed to avoid multicollinearity and to evaluate single variables. In addition to this, we tested model equations for linearity, homoscedasticity, normal distribution of residuals and significance. Model equations that did not fulfill the requirements were rejected.

6. Distribution of Lake C Burial and Correlation to Lake and Catchment Properties

[30] After a thorough review of published data and a quality check, 228 lakes were selected for further investigation. They are widely distributed over Europe covering a wide range of environmental conditions (Figure 3). The lake areas range from 0.006 km2 to Europe's largest lake, Lake Ladoga, with an entire surface of over 17,000 km2. Long-term C accumulation values were available for 183, CMAR20cm for 103 and both values were available for 66 of the 228 lakes. The data set comprises 36% of all European lakes larger than 100 km2 and 99 lakes with a surface area less than 1 km2 (Figure 4).

Figure 3.

Distribution of analyzed lakes and available data for CMARs. Most considered lakes are located in the boreal zone and the Alps, which roughly correspond to the natural appearance of lakes.

Figure 4.

Number of investigated lakes per surface area class. The numbers above the bars indicate the estimated total number of European lakes belonging to the respective size class. Although small lakes account for the bulk of investigated lakes (44 lakes smaller than 0.1 km2), they only represent a minimal portion of the naturally occurring number of lakes in this size class.

[31] Long-term C accumulation rates range from 0.1 (Lake Lampi) to 57.8 g C m−2 yr−1 (Lake Dudinghausen) with an average of 5.6 g C m−2 yr−1. However, 75% have rates less than 6.5 g C m−2 yr−1 (Figure 5). These values exclusively represent OC carbon burial. Only 13% of the lakes have significant carbonate accumulation. Most of these are located in regions with carbonate bedrocks, which means that the precipitated carbon does not contribute to atmospheric carbon withdrawal [Einsele et al., 2001]. Carbonate sedimentation is rare in the boreal zone, which comprises the main part of European lake area. Carbonate accumulation may be large within individual lakes but is of minor importance on a European scale.

Figure 5.

Histogram of the analyzed long-term CMARs and percentage of the total investigated lakes. Long-term accumulation of carbon is generally low. Only 14% of the investigated lakes have a CMAR > 10 g m−2 yr−1.

[32] Lakes with a large surface area generally have a low CMAR. While long-term C accumulation of lakes > 100 km2 does not exceed 5 g m−2 yr−1, it varies largely among small lakes (Figure 6a). North of 62° the values are generally below 10 g C m−2 yr−1 (Figure 6b). This observation is not exclusively due to an influence of temperature. A considerable number of lakes in regions with annual mean air temperature below 5°C have CMARs exceeding 10 g C m−2 yr−1. The highest CMARs can be found in regions with annual mean temperature between 5° and 10°C.

Figure 6.

Long-term CMAR plotted against (a) lake area and (b) latitude. Lakes larger than 100 km2 have relatively uniform low CMAR. Variations in CMAR are largest in small lakes situated in moderate climatic regions.

[33] Catchment properties were determined for 86 lakes. Single correlations are relatively low (Table 3). Temperature, latitude, ln(catchment area), mean slope of the catchment area and standard deviation of the catchment altitudes show highest Pearson coefficients.

Table 3. Single Correlations Between Long-Term CMAR in Lakes > 5 km2 and Lake and Catchment Parameters
VariableCMARln (CMAR)
ln (lake area)−0.45<0.01−0.50<0.01
ln (catchment area)−0.54<0.01−0.53<0.01
SD altitude0.45<0.010.36<0.01
Mean slope0.58<0.010.45<0.01
Bare area percent0.47<
Cropland percent0.30<
Population density0.44<0.010.32<0.01
Tree cover percent−0.32<0.01−0.250.03

[34] The best model for calculating ln(CMAR) by stepwise linear regression is given in equation (5). Nonlogarithmized CMAR by stepwise linear regression were rejected due to violation of the normality assumption of residuals.

equation image

Because catchment area and standard deviation of altitude are not appropriate parameters, we replaced them with the strongly correlating parameters lake area and slope. This modified model equation (equation (6)) is also highly significant with a slightly lower r2 of 0.42. Beta coefficients, i.e., contribution to the model equation, for ln(lake area), slope and percentage cropland are 0.39, 0.36 and 0.25, respectively.

equation image

Catchment characteristics could not be determined for lakes <5 km2. Hence, a statistical analysis for the whole lake data set (n = 228) including small lakes was restricted to a few general parameters (Table 4). Best correlation with CMAR was found for latitude and temperature with Pearson's correlation coefficients 0.49 and 0.42, respectively. Lower but also significant correlations result for ln(lake area), altitude and maximum depth.

Table 4. Single Correlations Between Long-Term CMARs and Parameters That Could be Determined for All Lake Sizes (n = 183)
VariableCMARln (CMAR)
  • a

    Maximum depth was only available for 97 lakes.

ln (temperature)0.42<0.010.42<0.01
ln (lake area)−0.39<0.010.55<0.01
ln (max. depth)a−<0.01

[35] As an outcome of the stepwise linear regression, the best predictors for CMAR are latitude and ln(lake area) (r2 = 0.49; p < 0.01). Because latitude and temperature are highly correlated they were analyzed separately. Analysis of the total data set supports a correlation between lake size and CMAR also for lakes < 5 km2. However, the large variability in the CMARs of small lakes cannot be explained by latitude or lake area. Furthermore, latitude is correlated with temperature, and especially in Europe, land use change and population.

7. Temporal Variations in Lake C Accumulation

[36] Although temporal variations in C accumulation vary significantly between lakes, some general trends can be observed. C accumulation increased in most lakes during the Holocene. The CMAR20cm are on average 5 g C m−2 yr−1 higher than the long-term C accumulation of about 5.6 g C m−2 yr−1. This is true for the 66 lakes for which both values were available. Adding lakes with only CMAR20cm values and no long-term record increases the mean value another 11.3 g to 21.9 g C m−2 yr−1. This is due to an overrepresentation of Central European lakes in areas characterized by high human impact and is well in agreement with statistical results. The geographical distribution of CMAR20cm is similar to the long-term CMAR but with enhanced values.

[37] Interestingly, we observed in the depth profiles of the investigated lakes that a strong increase of CMAR is always linked to a significant enhancement of sedimentation rates. Changes in OC content are of minor importance. Although absolute sedimentation rates should be regarded with caution, because of uncertainty of dating and age models, trends are significant.

[38] The onset of increased sedimentation rates varies between lakes. Most common is a starting point between 5000 and 2000 years BP. Several authors have linked this increase of the sedimentation rates to changing human activity in the catchment [e.g., Niessen et al., 1992; Lami et al., 1997; Schmidt et al., 2002]. This is true for example, for the three lakes with the highest calculated CMAR in 20 cm depth (>100 g C m−2 yr−1): Schwarzsee (Switzerland) [Dapples et al., 2002], Barton Broad (UK)[Bennion et al., 2001] and Lake Dudinghausen (Germany) [Dressler et al., 2006]. The timing of increased sedimentation rates roughly corresponds to a general change in human activities in Europe. During the Neolithic, agricultural societies increasingly replaced the hunter-gatherers [e.g., Kalis et al., 2003]. In some regions, such as the alpine foreland, lakes seem to be favorite settling areas. The close timing of increased sedimentation rates and the agricultural revolution in Europe supports the hypothesis of human impact although climate deterioration has been also proposed in some cases.

[39] Schwarzsee, Barton Broad and Lake Dudinghausen are small lakes with a surface area of 0.5, 0.6 and 0.2 km2, respectively. This means that the general relation between agricultural area in the catchment and enhanced CMAR in the lake can also be observed in lakes smaller than 5 km2. Due to different settlement histories, this impact varies strongly between lakes.

[40] Some lakes also show high sedimentation rates at their base. Dry matter accumulation is usually fastest immediately after lake initiation [Pajunen, 2004]. Although OC content is generally lower at this time due to sparse vegetation this can lead to enhanced CMAR. In addition, independent events such as flooding, fires, debris flows etc. can cause short time changes in the CMAR mostly on a local scale and effecting single or adjacent lakes.

8. Estimate of Carbon Burial in European Lake Sediments

[41] To obtain a more sophisticated estimate of long-term carbon burial in European lakes we considered main controlling factors, i.e., lake size, proportion of cropland and mean slope of the catchment area, in more detail. For practical reasons it was not possible to apply the model equation directly, in particular for calculating lake catchments. The approach used in this case involved defining geographic regions, according to their potential effect on lake CMAR. This enabled the assessment of the probable slope and fraction of cropland within the lake catchment without having to calculate their actual catchment areas. The assessment of the potential effect for these regions, i.e., the impact of land use and slope on carbon burial, required the reclassification of land use and slope maps. These simplified maps show continuous areas with high, medium or low agricultural land use and accordingly high, medium or low slope. The moving windows method was used to generalize given thematic maps and recalculate each pixel within a square of 20 km edge length. Finally, the newly generated thematic map was categorized into three classes of slope and agricultural land use (Figures 7 and 8) or in combination 9 classes of potential effect on lake carbon accumulation.

Figure 7.

Simplification of (left) the pixel-based agricultural land use map by (middle) a moving windows method results in (right) a regionalized map with three categories (example, Brittany). The simplification was necessary to distinguish regions with different intensities of agricultural land use.

Figure 8.

Merged classified map of land use and slope. This generalized map was subsequently used for pan-European extrapolation of carbon burial in small lakes. Basic colors (green, brown, red) indicate cropland categories (1, no/low use; 2, moderate use; 3, intensive use). Gradation of the colors indicate different slopes (1, flat; 2, moderate; 3, steep) within the specific cropland category.

[42] Potential increase of carbon accumulation (potential effect) was calculated by substituting the mean values of the thematic classes into equation (5) without considering lake area. The latter was directly taken from the GLWD for lakes > 5 km2. Potential CMAR and potential effect classes were computed for every lake. The estimated CMAR for a lake with a surface area of 1 km2 in a plane landscape and without surrounding croplands is 3.4 g C m−2 yr−1. The calculated CMAR for the same lake in an intensively utilized agricultural area is 22 g C m−2 yr−1. In combination with a steep catchment area, estimated CMAR for this lake increases to 80 g C m−2 yr−1. Lakes with known CMAR were excluded from this calculation. These were later included in the total estimate of carbon burial. Carbon burial of each individual lake was estimated by multiplying CMAR with lake area.

[43] The surface cover of lakes < 5 km2 was calculated for eight lake regions without knowing the exact location of single lakes. Based on the Pareto distribution, the number of lakes between 0.1 and 5 km2 was evaluated in 0.1 km2 steps for each region. Potential effect was calculated using the average fraction of land use and the mean slope within the specific regions. This information was used to estimate total carbon burial for lakes < 5 km2 in all eight regions. Results are presented in Table 5.

Table 5. Extrapolated C Accumulation (CAC) in Tons Per Year for the Different Lake Regions
CAC (<5 km2)13,23814,916115,383156,03945,667107,79911,578172,142636,762
CAC (>5 km2)16,8562,56995,519244,21139,01928,91631,388160,944619,422
CAC (total)30,09417,485210,902400,25084,686136,71542,966333,0861,256,184

[44] The results of this approach reveal a total C burial in European lake sediments of 1.25 Mt yr−1. Approximately 50% can be attributed to lakes < 5 km2. The high limnicity in the northern regions 3 and 4 accounts for 48% of Europe's lacustrine carbon burial. This is relatively low, when taking into account that 65% of the total European lake area of is found in these regions.

9. Discussion

[45] This study identifies catchment area, slope and proportion of cropland in the catchment as the main factors controlling carbon burial in lakes. Other parameters theoretically affecting sedimentation rates or primary production such as temperature, evapotranspiration, percentage bare area or population density do not contribute toward improving the model equation although they have significant individual correlations with CMAR. Agricultural land within a lake catchment affects carbon burial in several ways. First, clearing of natural vegetation (mostly forest) releases large amounts of carbon in a short time [Gurtz et al., 1980]. Second, the newly established cropland is more susceptible to erosion and weathering. This leads to a higher input of detritus [Stumm, 2005] and probably dissolved organic carbon into the lake [Mattsson et al., 2005]. The enhanced sedimentation rate enhances carbon preservation within the lake sediment [Einsele et al., 2001], because of the restricted time available for diagenetic alteration, e.g., by microbial consumption. Third, fertilizer use affects the trophic state of the lake and in turn its primary productivity.

[46] The observed correlation between high sedimentation rates and high C accumulation suggests that enhanced erosion is the main factor causing CMAR increases observed since the Neolithic agricultural revolution. Because most agricultural areas in Europe have been cultivated for several hundreds and thousands of years, this effect is well reflected in our data. It is less clear, however, whether CMAR in 20 cm depth has also been affected by the increased anthropogenic impact since the 19th century. Widespread use of industrial fertilizers and increased wastewater discharge caused widespread eutrophication of European lakes as of this time period [Asman et al., 1988]. While this effect is apparent at 20 cm depth in lakes with high sedimentation rates, it may be absent in others. Therefore, we presume that a comparison of CMAR at 20 cm depth with the long-term average can provide valuable clues about recent trends, but cannot yield a quantitative estimate.

[47] The mean CMAR at 20 cm depth for all lakes with long-term data (n = 66) exceeded the long-term mean by a factor of about 2. This doubling is attributable to a strong increase of lacustrine sedimentation rates in agricultural areas, especially in the temperate climate zone. Boreal lakes are less affected so that the risk of underestimating the mean CMAR at 20 cm depth based on the long-term means is rather small.

[48] In this study, the upper 20 cm of the sedimentary record were excluded from the large-scale analysis. Hence, the results do not cover the more recent impacts on lake systems, in particular the widespread eutrophication in agricultural and densely populated areas. The increased primary productivity caused by lake eutrophication probably results in an enhanced C accumulation as well as long-term mineralization in the water column and the upper sediment. Several lakes, however, developed an oxygen deficiency [Thomas, 1969], decreasing organic consumption and increasing organic burial.

[49] On the other hand, the effect of eutrophication may be restricted to a particular time period depending on the treatment of wastewater and the nature of land use. For instance, Kappler et al. [2001] observed continuous TOC contents in the uppermost 8 cm (corresponding to about 20–40 years) of the sediment of Lake Constance. However, the organic matter content decreases strongly between 8 and 20 cm depth due to ongoing microbial activity. Organic matter contents are expected to be lowest in the upper few millimeters of the sediment column, where oxygen penetration causes high respiration activity [Brune et al., 2000]. Since constant TOC is probably due to a decrease in eutrophication during the last 40 years, there will be no permanent enhancement of carbon burial in this instance. Other formerly eutrophic water bodies in Europe will presumably display a similar TOC distribution due to the implementation of different environmental measures, e.g., the EU Water Framework Directive.

[50] An additional recent factor, which could lead to enhanced carbon burial in lakes, is an increased DOC input into the lakes. Monteith et al. [2007] noticed increased DOC concentrations in the surface waters of North America, northern and central Europe. DOC can be transformed into POC in the lake water column by flocculation and deposited as lake sediment. This process is of major importance in the boreal zone [von Wachenfeldt and Tranvik, 2008]. Changing climate and runoff is likely to affect carbon inputs to aquatic systems from the watershed [Tranvik et al., 2009]. Freeman et al. [2001] suggested that rising temperature could increase the transport of DOC from peatlands to the oceans, while Tranvik and Jansson [2002] stated that precipitation is the main cause of changing DOC fluxes. This is particularly true for boreal regions with high soil C contents and extended wetlands. Here the increased precipitation predicted by Intergovernmental Panel on Climate Change [2007] may significantly increase water runoff and change the influx of organic matter into lakes. However, a contrary trend has been observed in parts of boreal Canada, where runoff and DOC concentrations in the surface waters have decreased [Benoy et al., 2007].

[51] There is some difficulty with quantifying the extent of the effects of these recent changes in the lake systems on permanent C storage in lake sediments. Large quantities of organic matter can be degraded, even under anoxic conditions [Wetzel, 2001]. However, little is known about the actual quantities of organic material affected by these processes. In addition to this, modern eutrophication as well as global warming are changing lakes into nonsteady state systems. Each lake system has its specific sensitivity and response time, a fact that cannot be taken into consideration in a continent-wide study like this. It is noteworthy, that eutrophication is decreasing in some lakes and that the impact on C burial is probably not permanent. On the other hand, there remain many uncertainties with respect to global warming that have an affect on the entire hydrological cycle as well as on temperature-dependent microbial activities. Gudasz et al. [2010] found a strong positive relationship of OC mineralization in lake sediments and temperature. They conclude based on Intergovernmental Panel on Climate Change scenarios that future OC burial in boreal lakes could decrease by 4–27%. Although our approach is not suitable for calculating the actual carbon burial rates for European lakes, it does provide a critical threshold value for their long-term burial capacities. This value should be taken into consideration in modeling approaches.

[52] The observed low C accumulation in large lakes is in good agreement with previous estimates of carbon burial. Mulholland and Elwood [1982] reported CMAR between 2 and 10 g m−2 yr−1 for lakes larger than 500 km2 with the exception of Lake Erie and the Caspian Sea, which accumulate 30 and 19 g C m−2 yr−1, respectively. For their global estimate, Dean and Gorham [1998] used a CMAR of 5 g C m−2 yr−1 for lakes larger than 5000 km2. The present study indicates that C accumulation in large European lakes lies within the same range. However, unlike previous authors, this study shows that small, relatively uniform CMAR are also found in lakes with an area of 100 km2. Since the database used in this study comprises more than one third of all European lakes larger than 100 km2, it can be assumed that the boundary size between large and small lakes chosen by previous authors is not representative on a global scale, in any case not for Europe.

[53] In spite of their low CMAR, the contribution of large lakes to total C burial in lakes is significant due to their wide surface coverage. The largest European lake, Lake Ladoga, accounts for 4.4% of European lacustrine C burial, although its CMAR is only about 3.1 g C m−2 yr−1.

[54] The present study indicates that CMARs of small lakes in Europe are significantly lower than those taken for global estimates so far. Mulholland and Elwood [1982] used a value of 50 g C m−2 yr−1 for lakes smaller than 4000 km2, while Dean and Gorham [1998] assume an accumulation of 72 g C m−2 yr−1 in lakes smaller than 5000 km2. The discrepancy between these values and those presented in this study is probably due to the large contribution of boreal lakes to the European lake area. Regional estimates by Kortelainen et al. [2004] for Finland (data included in this study) and Campbell et al. [2000] for Alberta already introduced low CMARs for the boreal zone. These agree with our results for Europe and reflect the fact that the boreal zone is less modified by human activity. As a result, boreal lakes have a lower CMAR than, for example, the lakes of Minnesota studied by Dean and Gorham [1998]. Mulholland and Elwood [1982] also reported a strong increase of sedimentation rates due to human activities in a large portion of their considered lakes.

[55] Furthermore this study presents Holocene mean accumulation rates, which deviate from recent data used in other studies. The CMAR used by Dean and Gorham [1998] assumes an average postsettlement sedimentation rate of 3 mm yr−1 and an average OC content of surface sediments of 12%. This results in a dbd of about 0.2 g cm−3. This assumption neglects the observation that the uppermost centimeters of the sediment are less compacted, which results in a lower dbd and hence lower accumulation rates. The average sedimentation rate was applied by Webb and Webb [1988] who explicitly stated that the chosen midlatitude rates are significantly higher than those found in lakes north of 50° or south of 40° latitude. Degradation of organic material continues in the uppermost part of the sediment and can significantly deplete the OC contents. This means that the C content of surface sediments does not represent permanent carbon burial. Even our subrecent data at 20 cm depth does not support such high CMAR for the bulk of European lakes as reported by Dean and Gorham [1998] or Mulholland and Elwood [1982] for small lakes.

[56] In conclusion, previous global estimates of C burial in lakes are not based on representative CMAR data. A large proportion of the European lakes are located on the Fennoscandian Shield, which resembles the natural landscape conditions of the Laurentian Shield in many ways [Simola and Arvola, 2005]. Thus a significant proportion of the global lakes bury much less carbon than assumed even in high limnicity regions strongly impacted by human activities, particularly in the northern United States and southern Canada.

[57] On the other hand total lake area, particular the surface coverage of small lakes, has been underestimated in previous studies. We calculated a European lake area of 240,000 km2 for lakes > 0.1 km2 which is 40% more than the GLWD comprises and about 5.7% of the global lake area estimated by Downing et al. [2006]. Combining our new CMAR model with our new estimate of lake cover in Europe, results in a total C sequestration of 1.25 Mt yr−1. This is rather low compared with other terrestrial C sinks such as the European forest, which accumulated an estimated 140 Mt C yr−1 during the 1990s [Nabuurs et al., 2003].

[58] Assuming European lakes bury carbon in the same order of magnitude than in other regions, global C sequestration rates would be around 22 Mt yr−1 or 0.022 Gt yr−1. This value is valid if the distribution and properties of European lakes are representative on a global scale. Although Europe does not include tropical regions, this assumption is reasonable for major lake regions of the Earth from the boreal to the subtropical climate zone. Therefore, we propose this as a reasonable tentative global estimate, based on a thorough analysis of different lake types over a whole continent. This also suggests that an underestimate of lake area and an overestimate of CMAR could cancel each other out on a global scale.

[59] There are further factors limiting the accuracy of the presented estimate. Sediment focusing especially affects lakes with steep basin slopes [Blais and Kalff, 1995]. Hence, the C accumulation of some lakes is probably overestimated. Other inaccuracies result from the limited resolution and quality of spatial data. The parameter cropland, for example, merges miscellaneous agricultural land use types, which presumably affect lakes in different ways.

[60] Uncertainties also exist with respect to the estimated European lake area. In particular the largest lake region (region 8) is certainly not as uniform as assumed. Because this area has a relatively low limnicity, the absolute error for European lake area remains low, but the effect on C burial estimate is rather higher because of stronger anthropogenic impact in this region.

[61] For the extrapolation of CMAR in lakes < 5 km2 we assumed that carbon burial is controlled by the same parameters identified for lakes > 5 km2. This assumption could only be tested qualitatively. The correlations between CMAR and lake size for lakes > 5 km2 resembled that found for the whole data set. For lakes with a notably high CMAR, the enhanced C accumulation can be explained by the presence of agricultural land uses and/or steep hillslopes in the catchment area. The extrapolated CMAR values for small lakes were in the same range as the observed data, although an independent approach was chosen. We concluded that the model equation is also suitable for small lakes.

[62] The role of very small lakes (<0.1 km2) still remains uncertain. Downing et al. [2006] suggest that the Pareto distribution can also be used for very small lakes except in the case of some arid to semiarid regions. A test to find out whether this also applies to Europe could not be performed within the scope of this study. Very little is known about the number of very small lakes in Europe and their surface area, however their importance is presumably only small in view of the relatively low CMAR. Kortelainen et al. [2004] noted less carbon burial in such lakes of the boreal zone due to absent river inflow. Because the fraction of very small lakes is probably negligible in other regions, the total contribution of these to the European lake carbon accumulation is rather low.

[63] This study only deals with natural lakes. The role of artificial lakes is still a matter of speculation. Estimates suggest that while C content in artificial lakes is generally smaller than in natural lakes, C accumulation is several times larger due to the very high sedimentation rates [Dean and Gorham, 1998; Mulholland and Elwood, 1982]. Reservoir management measures such as sediment flushing or dredging to avoid aggradation also make it difficult to obtain reasonable estimates. The World Register of Dams [International Commission on Large Dams, 2003] lists over 5600 large reservoirs in Europe with a total surface area of about 48,000 km2. Although total C burial of reservoirs may well considerably exceed C accumulation in natural lakes, this does not also imply long-term storage of carbon. The role of small impoundments such as farm ponds also seems particularly interesting. Downing et al. [2008] studied 40 eutrophied small impoundments in Iowa with CMAR ranging from 148 g C m−2 yr−1 to 17,000 g C m−2 yr−1. In view of these extraordinary high C accumulations, small impounds are an important factor in the C Cycle, despite their small estimated global surface cover of 77 000 km2 [Downing et al., 2006] compared to lakes.

10. Conclusions

[64] According to this study, the estimated permanent long-term C burial in European lakes amounts to 1.25 Mt yr−1. Half of this storage takes place in boreal lakes of northern Europe, although they contribute up to 65% to the European lake surface. This is due to generally lower carbon burial rates in this region. C accumulation rates in larger lakes are generally low. In small lakes these rates vary considerably both spatially and in time. Agricultural land use enhances C accumulation particularly in small lakes with steep catchment areas. The observed increase in C burial recorded between 5000 and 2000 years BP was caused by Neolithic land use change. From a European perspective, C accumulation rates for small lakes have been overestimated in previous global estimates while their appearance and total surface area has probably been underestimated. A tentative extrapolation of European data on a global scale yields a global C burial rate of 0.022 Gt yr−1. The total European lake area >0.1 km2 was estimated to 240,000 km2 based on a Pareto distribution. Lakes are a minor, but permanent C sink.


[65] The study was funded by the Deutsche Forschungsgemeinschaft (DFG), project HI 643/4-1.