Assessing scale effects on modelled soil organic carbon contents as a result of land use change in Belgium

Authors


N. Dendoncker. E-mail: n.dendoncker@ed.ac.uk

Abstract

This paper explores the influence of spatial scale on modelled projections of soil organic carbon (SOC) content. The effect of land use change (LUC) on future SOC stocks was estimated using the Rothamsted Carbon model for a small area of southern Belgium. The study assumed no management change and used a single climate change scenario. Three model experiments were used to identify how data scale affects predicted SOC stocks: (i) using European LUC datasets at a resolution of 10′ and assuming equal distribution of change within the study area, (ii) using more accurate regional data aggregated to the 10’ resolution, and (iii) using the regional data at a spatial resolution of 250 m. The results show that using coarse resolution (10′) data is inappropriate when modelling SOC changes in the study area as only the methods using precise data predict a change in SOC stocks similar to those reported in the literature. This is largely because of differences in model parameterisation. However, precisely locating LUC does not significantly affect the results. The model, using either pan-European or region-specific precise data predicts an average SOC increase of 1 t C ha−1 (1990–2050), mainly resulting from afforestation of 13% of agricultural land.

Introduction

Land use change (LUC) is known to be a major factor influencing the global carbon cycle and in particular soil organic carbon (SOC) content (Houghton et al., 1999). However, only a few studies explicitly account for the effect of LUC on SOC changes at the regional scale. Most of these studies focus on a single type of land use (LU) conversion, are not spatially explicit, focus on past LUC, or make short-term projections (e.g. Batjes, 1999; Houghton et al., 1999; Priess et al., 2001; Kerr et al., 2003; Grünzweig et al., 2004). Future LUC is highly uncertain, but it will have a large impact on the future terrestrial carbon balance (Levy et al., 2004).

Smith et al. (2005) were the first to use spatially explicit LUC scenarios to present a comprehensive pan-European assessment of future agricultural SOC changes. To undertake a study over such a large spatial extent, several simplifying assumptions had to be made. For example, by working at a spatial resolution of 10′, the mean SOC stocks and clay contents for each 10′ grid cell were selected from the European Soils Bureau 1 km2 database (Jones et al., 2005). Similarly, conversions from one LU type to another were estimated using a simple formula to match the predicted changes in LU. Whilst such simplifications are appropriate for continental scale model applications, it is reasonable to ask how representative the results will be for small land areas. Datasets with finer spatial resolution, and arguably better intrinsic quality might be expected to improve the precision of the results for small areas, but no previous studies have compared SOC estimates obtained by working at different spatial scales.

Spatial scale can be defined as the dimension used by scientists to measure and study objects and processes (Gibson et al., 2000). All scales have extent and resolution (Verburg et al., 2004). In this study, scale varies in terms of the actual resolution (and quality) of the datasets used and the level of aggregation at which the analysis is performed. The Belgian soil associations database (Tavernier & Maréchal, 1962) and a database of soil profiles (Aardewerk database, Van Orshoven et al., 1993) were used together with the European soil database (Jones et al., 2005) to estimate SOC for a part of Belgium. Belgium has comprehensive SOC databases (Lettens et al., 2004), so it is an ideal case study to test between European and good quality regional soil data. The LU data used here were based on the downscaled future scenarios of the Advanced Terrestrial Ecosystem Assessment and Modelling project (ATEAM; Ewert et al., 2005; Rounsevell et al., 2005, 2006; Schröter et al., 2005) following the method proposed by Dendoncker et al. (2006, 2007), allowing for a precise location of the LU conversions within the study area for the years 2020 and 2050.

The analysis was then performed in three ways: (1) working at the ATEAM 10′ resolution by averaging the available European data in a similar way to that used by Smith et al. (2005); (2) working at the ATEAM 10′ resolution, but using national databases and more accurate plant input based on independent studies; and (3) using the same available information as in (2) but working at a resolution of 250 m: i.e. taking into account the appropriate clay content and initial SOC stock of the soil association of the CORINE cells within which the various LU conversions occur. The main objective of this paper is to examine the impacts of working at a coarse level of aggregation and using coarse resolution datasets when projecting changes in SOC for a specific study area. The effect of using accurate datasets was estimated by comparing methods (1) and (2), while the effect of changing the level of aggregation of the analysis was estimated by comparing methods (2) and (3). This analysis is intended to provide insight into the appropriateness of using coarse resolution datasets at the European scale when making estimates of SOC change in response to scenarios of LU change.

Material and methods

The Rothamsted carbon model

The Rothamsted carbon model (RothC) 26-3 model (Coleman et al., 1997) is one of the most widely used SOC models and has been applied at the regional (e.g. Falloon et al., 1998a), continental (Smith et al., 2005, 2006) and global scale (Tate et al., 2000). It has been validated for a variety of ecosystems including croplands, grasslands and forests (Smith et al., 1997). The RothC model was chosen in this study because of its simplicity and, hence, because of the availability of data to run the model for long time series at the scale of a small region. RothC requires three types of data: (1) weather data: monthly rainfall (mm), monthly potential evapotranspiration (mm) and average monthly mean air temperature (°C); (2) soil data: clay content (%), inert organic carbon (IOM), initial SOC stock and depth of the soil layer considered (cm); and (3) LU and management data: soil cover, monthly input of plant residues (t C ha−1), monthly input of farmyard manure (FYM) (t C ha−1) and finally a residue quality factor (DPM/RPM ratio).

Study area.  The study area consists of a single 10′ by 10′ grid cell used within the ATEAM project (Schröter et al., 2005; Rounsevell et al., 2006) located in southern Belgium (Figure 1). The particular ATEAM cell was chosen because it contains contrasting soil textures, as clay content is an important parameter in RothC. This was important to meet the objectives of this study to examine the effects of different levels of aggregation. Focusing on a single 10′ cell also limits the extent of the analysis and the number of simulations.

Figure 1.

 Study area. The Advanced Terrestrial Ecosystem Assessment and Modelling (ATEAM) grid is overlaid on Belgium (left) and the selected ATEAM cell is shown on the right, the thinner lines delineate the soil associations as defined in Tavernier & Maréchal (1962).

The chosen cell is located at the border of two bio-geographic regions with distinct characteristics, and with large differences in soil types (Tavernier & Maréchal, 1962): the ‘Ardennes’ in the north and the ‘Jura’ in the south. The main soils of the Ardennes are poor stony loam soils (with schists and phyllites) generally unfavourable to arable agriculture. Mean temperature ranges from 7 to 8.5 °C while precipitation varies from 1100 to 1400 mm year−1 (Maréchal & Ameryckx, 1992). The Jura comprises a mix of soil types including sandy soils, sandy loams, clayey soils and alluvial soils in the valleys. Mean temperature reaches 8.6 °C in Virton and precipitation varies from 800 to 900 mm year−1 (Maréchal & Ameryckx, 1992). This mix of soil types associated with a warmer and less humid climate makes the region more favourable to agriculture. The various soil associations are described in Table 1. Dealing with such a range of soil types allows for a better estimation of the effect of soil type and initial SOC content on projected SOC changes, as this will lead to larger differences between the three model experiments described below.

Table 1.   Description of the soil associations present in the study area
Association numberaDescription Clay content (%)SOC content in 1960 cropland (t C ha−1)SOC content in 1960 grassland (t C ha−1)
  1. aBelgian soil associations as described by Tavernier & Maréchal (1962). SOC, soil organic carbon.

40Stony loams with gravel18.249.748.1
45Stony loams with conglomerate18.633.438.2
47Stony loams with limestone and shale23.469.776.1
50Stony loams with shale and phyllite18.971.175.8
52Dry loamy soils16.548.751.5
55Sandy to sandy-loam soils11.895.387.7
58Clay soils23.954.957.2
59Dry alluvial soils19.157.367.3
60Humid alluvial soils17.275.891.1
62Steep slopes17.362.366.4

The three model experiments

The first model experiment used coarse datasets at a spatial resolution of 10′. This results in one modelled SOC change value per LU type for the entire cell. This approach will subsequently be referred to as experiment ‘da’, where ‘d’ stands for coarse datasets and ‘a’ stands for coarse level of aggregation. The level of aggregation does not change in the second experiment (referred to as ‘Da’, where ‘D’ stands for regional data), but regional datasets are used instead of the European data. The third experiment (referred to as ‘DA’, where ‘A’ stands for fine level of aggregation) uses the regional datasets at a spatial resolution of 250 m. The input data and calibration procedures used in each experiment are described below and in Tables 2 and 3.

Table 2.   Monthly repartition of plant input
Experiment da – plant input (t C ha−1 year−1)Experiment Da and DA – plant input (t C ha−1 year−1)
MonthCroplandGrasslandForestMonthCroplandGrasslandForest
January00.270.12January00.200.09
February00.270.12February00.200.09
March0.740.270.12March0.290.200.09
April0.740.270.12April0.290.200.09
May0.740.540.24May0.290.390.19
June0.740.810.24June0.290.590.19
July2.940.810.24July1.150.590.19
August00.540.24August00.390.19
September00.540.95September00.390.74
October00.540.95October00.390.74
November00.270.95November00.200.74
December00.270.47December00.200.37
Total5.95.44.7Total2.33.93.7
Table 3.   Synthesis of the parameters used in the three model experiments
 Experiment daExperiment DaExperiment DA
  1. LU, land use; SOC, soil organic carbon.

Baseline SOCFrom the European soil database, a single value for the entire cell and for all LU typesFrom Lettens et al. (2004), one surface weighted average value for the entire cell and for each LUFrom Lettens et al. (2004), one value per soil association and for each LU
Clay contentFrom the European soil database, a single value for the entire cellFrom the Aardewerk database, one surface weighted average value for the entire cellFrom the Aardewerk database, one value per soil association (cf. Table 1)
LU conversionsNo spatial informationNo spatial informationSpatially explicit within each soil association
Plant inputIteratively fitted to match observed SOC contents. One value per land use typeCropland and grassland: calibrated and tested on Belgian agricultural areas by van Wesemael et al. (2005). Forest: Iteratively fitted to match observed SOC contentsCropland and grassland: calibrated and tested on Belgian agricultural areas by van Wesemael et al. (2005). Forest: Iteratively fitted to match observed SOC contents
Climate dataA1FI scenarioA1FI scenarioA1FI scenario

Input data

Mean monthly temperature and precipitation values for the selected cell were taken from a single climate change scenario – i.e. the A1FI HadCM3 ATEAM climate dataset (Mitchell et al., 2004), which has the largest temperature increases of the IPCC special report on emissions scenarios (SRES) (Nakicenovic et al., 2000). Potential evapotranspiration was estimated using the empirical formula of Thornthwaite (Shaw, 1994). To focus only on the effect of LUC, changes in NPP and technology changes were not considered, although they have been shown to significantly affect predictions of SOC stocks (Smith et al., 2005).

For the experiment da, clay content was extracted from the European soil database (Jones et al., 2005) and a single clay content of 26.5% and initial SOC content of 98 t C ha−1 (upper 30 cm) were applied to the entire cell to be consistent with the method of Smith et al. (2005). For the experiments Da and DA, clay content and initial SOC content were extracted from the Aardewerk database (Van Orshoven et al., 1993) for each soil association (and combination of soil association and LU) in the study area (Table 1). In experiment Da, a single surface-weighted average value was kept for the entire cell. In experiment DA, the models were run at a resolution of 250 m, so the original clay and SOC content values were used.

The CORINE database (version 8/2005, available at: http://dataservice.eea.eu.int/dataservice/) was extracted for the study area and thematically generalized into four LU classes (i.e. Urban, Cropland, Grassland and Forest) to match the categories of the ATEAM scenarios of LUC. Only cells under cropland and grassland in 1990 were retained to focus on baseline agricultural land. This gives a total of 1757 cells covering an area of 10 981 ha. The A1FI ATEAM scenario of LUC was downscaled to the spatial resolution of CORINE (250 m) following the method proposed by Dendoncker et al. (2006). The LU category ‘liquid biofuels’ (e.g. oilseed rape) present in the ATEAM scenario was grouped with arable land to match the CORINE LU categories. This was performed for both time slices: 2020 and 2050. LU trajectories were stored for the 1757 cells of the baseline agricultural area. Eight trajectories are possible, cropland to cropland (no change), cropland to grassland, cropland to forest, cropland to urban; grassland to grassland (no change), grassland to cropland, grassland to forest and grassland to urban.

Running the model

Model calibration.  The model calibration procedure differs between experiment da and experiments Da and DA. The simulations in experiments Da and DA used the most accurate data available for Belgium. SOC stocks for 1960 per landscape units (i.e. the intersection between CORINE LU classes and soil associations) from Lettens et al. (2004) were used to initialise the model (Table 1). van Wesemael et al. (2005) obtained calibrated plant input for arable soils in Belgium by using RothC to simulate the change in SOC stocks for long-term experiments representing the dominant Belgian crop rotation (i.e. sugar beet – winter wheat – winter barley). The long-term experiments were conducted by the Centre de Recherche Agronomique de Gembloux from 1959 onwards (Frankinet et al., 1993). Climate data [i.e. monthly precipitation (mm) and average monthly air temperature (°C)] were available from a spatial climate database (Mitchell et al., 2004) and potential evapotranspiration was calculated from the temperature data using the empirical formula of Thornthwaite (Shaw, 1994). Inert organic matter (IOM) was estimated from the initial SOC stock using the equation proposed by Falloon et al. (1998b). RothC was run to equilibrium under constant climate conditions, i.e. average monthly climate data for 1900–1959 and iteratively fitting carbon inputs to match the initial SOC stock and thus the distribution in the four active compartments: decomposable plant material (DPM), resistant plant material (RPM), microbial biomass (BIO) and humified organic matter (HUM; Coleman et al., 1997). The default DPM/RPM ratio of 1.44 was used for cropland and grassland. By optimising the total SOC stock predicted by the model to the measured data of the control experiment, a value of 2.3 t C ha−1 year−1 was derived for annual carbon input for a soil depth of 30 cm. van Wesemael et al. (2007) also showed that farmyard manure (FYM) production remained constant throughout the second half of the 20th century and that calibrated plant input for RothC included baseline FYM production for the Walloon region. Because of a lack of long-term experiments for grassland in Belgium, a standard plant input value of 3.9 t C ha−1 year−1 was used for a soil depth of 30 cm from the Rothamsted experiment (Coleman et al., 1997). These parameter values allowed the agricultural SOC change from 1960 to 2000 to be reproduced reasonably well with a root mean square error (RMSE) of 1.1 t C ha−1 for the Belgian soil associations (van Wesemael et al., 2007).

As one of the objectives was to examine the effect of LU conversions on SOC stocks, the model was also calibrated for urban and forest LU. The RothC model has never been validated for urban ecosystems and only a few studies deal with urban carbon stocks (e.g. Jo, 2002; Pouyat et al., 2002). However, the region considered is essentially rural and the few new urban settlements are likely to be composed of single family dwellings with relatively large gardens and so, extensive non-built areas. Urban LU was assumed, therefore, to have similar SOC stocks to grassland and the grassland SOC value of 3.9 t C ha−1 year−1 was allocated to urban cells. This probably results in a slight overestimation of urban SOC stocks but, given the low number of urban cells, the assumption is not likely to significantly affect the results.

As no independent value was available, plant input for forest soils were calibrated in a similar way as for cropland (see above): a default value of 7.09 t C ha−1 year−1 given by Falloon et al. (1998b) was used to initiate a series of equilibrium simulation runs to adjust the annual plant addition to observed SOC values for each soil association. Observed SOC values per soil association and for the years 1960 and 2000 were available from Lettens et al. (2005). The SOC stock of forest soils has increased on average by 18.3 t C ha−1 over a 40-year period for the soil associations considered. This is a combined effect of the increased productivity of forests and the change in forest management as Belgian foresters currently leave behind more dead wood than in the 1960s (Lettens et al., 2005). To take account of this observed increase, plant inputs were calibrated using the observed SOC values for the year 2000. An area weighted mean was computed giving a final value of 3.66 t C ha−1 year−1. The monthly partitioning of C inputs throughout the year for cropland, grassland and forest is given in Table 2 (bottom). RothC is known to be relatively insensitive to the distribution of C inputs during the year (Smith et al., 2005).

Experiment da applied the same data and methods as Smith et al. (2005). The model was thus initialised in 1900 with a single starting SOC value for all LUs and a single clay content for the entire cell. Plant inputs obtained by using this starting SOC value and clay content from the European Soils Bureau database are higher than those used in experiments Da and DA (Table 2). All methods allow for the minimizing of potential initialisation effects (Smith et al., 2005): experiment da has a spin-up period from 1900 to 1990 and experiments Da and DA from 1960 to 1990.

Simulations for experiment da.  The applied method was similar to that used by Smith et al. (2005). RothC was run to the year 2050 for each LU type (arable, grassland and forest) and for the entire ATEAM cell, assuming no LUC. The carbon stocks were stored for each LU in the last month of the years for which LU data were provided (i.e. 1990, 2020 and 2050). Values are given in number of cells rather than in hectares. The LU data were used to determine changes in LU occurring during 30-year time slices: i.e. the number of cells changing LU in each year was obtained by a linear interpolation between the calculated LUC between 1990 and 2020, and between 2020 and 2050. The SOC simulations were then run imposing annual LUC, using the stored SOC values for unchanged LU to seed the simulations. The main difference between this method and that used by Smith et al. (2005) is that LU conversions (in number of cells) within the study area are known and a fortiori do not have to be estimated.

Simulations for experiment Da.  The same method described for experiment da was applied in experiment Da except that input data were based on Belgian national sources. Clay content and initial SOC were extracted from the Aardewerk database, and plant input was calibrated as described in section Input data.

Simulations for experiment DA.  The same method described for experiment da was applied in experiment DA, but using Belgian input data as in experiment Da. The main difference with experiment Da is that the simulations were made for each LU type at the soil association level. This means that clay contents and initial SOC stocks of each soil association were used, and that LUC was calculated (in number of cells) for each soil association present in the study area. In this way, LUC is directly linked to the soil type on which it occurs. Such a methodology implies a large number of simulations and would therefore be difficult to implement over a large geographic area such as Europe. Table 3 summarizes the information used.

Results

The A1FI scenario of land use change

Figures 2 and 3 summarize the LUC scenarios in the study area: 23% of all CORINE cells change LU between 1990 and 2050 (i.e. 403 cells). About 64% of this change occurs in grassland cells, and the remaining 36% occurs in cropland cells. The majority of these changes are conversion to forest. Only 2.7% (47 cells) of the baseline agricultural area becomes urbanized. Some grassland cells are converted to cropland (mostly biofuel) and a few cropland cells change into grassland. Generally speaking, most change occurs at the 1990 agricultural frontier (i.e. at the interface between agricultural land and urban and forested land). Moreover, areas dominated by one LU type in 1990 tend to further specialize into that LU type while most isolated cells are lost (i.e. a cropland cell surrounded by grassland cells will no longer be classified as cropland in 2050). This is a consequence of the downscaling methodology, which is strongly based on neighbourhood relationships (Dendoncker et al., 2006, 2007).

Figure 2.

 Agricultural land within the study area in 1990, extracted from the CORINE land cover map (a); evolution of this agricultural area as projected by the Advanced Terrestrial Ecosystem Assessment and Modelling (ATEAM) A1FI scenario of land use change (b); and location of changing CORINE cells (c).

Figure 3.

 Number of CORINE cells changing land use between 1990 and 2050 per soil association, as projected by the Advanced Terrestrial Ecosystem Assessment and Modelling (ATEAM) A1FI scenario of land use change (the various soil associations are described in Table 1).

RothC simulations

Without LUC.  Changes in SOC stocks for cropland and grassland under the A1FI scenario of climate change are presented in Figure 4. Results are presented in percent change relative to the 1960 value (assuming that the 1960 value = 100%). In experiment da, cropland and grassland SOC stocks increase slightly. Cropland increases by a mean of 0.04 t C ha−1year−1, or 4.01 t C ha−1 over the entire period. This is equivalent to an increase of 3.8% of the 1960 value. Grassland SOC stocks increase by a mean of 0.07 t C ha−1year−1, or 6.37 t C ha−1 over the whole period; equivalent to an increase of 5.9% of the 1960 value.

Figure 4.

 Change in soil organic carbon (SOC) stock of (a) croplands and (b) grasslands without land use change. SOC stocks are expressed in percent change compared with the 1960 stocks (=100%).

The results of the simulations for experiments Da and DA are different from those of experiment da as, in the former case, cropland SOC stocks decrease quite strongly, while grassland SOC stocks increase strongly. For both experiments, cropland SOC stocks are expected to decrease by a mean of about 0.12 t C ha−1 year−1, or 10 t C ha−1 over the whole period; equivalent to a decrease of about 18% of the 1960 value. In contrast, grassland SOC stocks are expected to rise by a mean of about 0.12 t C ha−1 year−1, or 12 t C ha−1 over the whole period; equivalent to a rise of about 18% of the 1960 value. The slight difference between experiment Da and DA is due to the level of aggregation at which the simulations were performed. For experiment Da, the runs were performed at the ATEAM cell level while for experiment DA the runs were performed at the soil association level (using different clay content values and different baseline SOC values), then aggregated to obtain a single value for the entire ATEAM cell. Only results from experiments Da and DA are similar to those observed for the Walloon region: an increase of SOC in grassland (+10%) and a decrease of SOC in cropland (−16%) with the largest changes in grassland occurring between 1960 and 1990 (Lettens et al., 2005).

Incorporating land use conversions.  The number of cells that change LU varies greatly depending on the soil association, and this was taken into account in the DA experiment (Figure 3). For the results of the three experiments to be comparable, the area weighted mean SOC stocks for the 10′ cell were computed for each type of LU conversion and for each year of the period considered (1960–2050) (Figure 5).

Figure 5.

 Change in soil organic carbon (SOC) stock taking into account land use change (starting in 1990) from (a) grassland to cropland, (b) cropland to grassland, (c) grassland to forest and (d) cropland to forest. SOC stocks are expressed in percent change compared with the 1960 stocks (=100%).

Soil organic carbon stocks were averaged over a 5-year period to compensate for the variation caused by the inter-annual climate variability (Table 4). Estimated changes in SOC range from a decrease of −0.22 t C ha−1 year−1 (or 18% of the mean 1985–1990 value) following conversion from grassland to cropland (experiment Da) to an increase of 0.33 t C ha−1 year−1 (or 28% of the mean 1985–1990 value) following conversion from cropland to forest (experiment Da). In general, estimated changes are larger when experiments Da or DA are used, and the results of experiments Da and DA are fairly similar.

Table 4.   Summary of modelled soil organic carbon (SOC) changes
 Mean SOC stockChange (t C ha−1 year−1)
1985–1990 (t C ha−1)2045–2050 (t C ha−1)
Grassland to cropland
 Experiment da111.5110.6−0.02
 Experiment Da73.260.1−0.22
 Experiment DA70.760.2−0.17
Grassland to forest
 Experiment da111.5119.80.14
 Experiment Da73.282.60.16
 Experiment DA70.283.10.22
Cropland to grassland
 Experiment da109.9113.90.07
 Experiment Da57.367.80.17
 Experiment DA61.272.40.19
Cropland to forest
 Experiment da109.9119.50.16
 Experiment Da57.373.60.27
 Experiment DA55.174.20.33

Taking the whole study area into account, including cells that do and do not change LU, average SOC increases from 110.73 to 113.06 t C ha−1 for experiment (da) while it increases from 66.20 (1990) to 68.21 (2050) t C ha−1 for experiment DA. In the absence of LUC, SOC stocks would increase to 112.1 t C ha−1 for experiment da and to 67.0 t C ha−1 for experiment DA. This means that the contribution to the increases in SOC stocks because of LUC is equal to about 1.2 t C ha−1 or 60% of the change for experiment DA and 1 t C ha−1 or 40% of the change for experiment da. A fortiori, this means that the contribution to the increases in SOC stocks because of the effect of climate alone is equal to about 40% of the change in experiment DA and 60% of the change in experiment da.

Discussion

Without LUC

Results for the change in SOC under cropland modelled in experiment da and for change in grassland SOC modelled in all three experiments show increases in SOC over time. However, plant inputs used to model the change in SOC under cropland in experiments Da and DA were calibrated using long-term experiments representing the dominant Belgian crop rotation while plant inputs used in experiment da were not calibrated and are higher than those estimated from region-specific experiments. These high plant inputs, and the fact that the European soil database has substantially higher clay content, give much higher SOC values in experiment da. This occurs in spite of using the same climate data as for the other two experiments. In experiment da, results for cropland are very similar to those for grassland. This can be explained by the initialisation of the model in 1900 using the same SOC value for both cropland and grassland.

Results for experiments Da and DA are fairly similar with the slight difference because of the level of aggregation at which the simulations were performed. Results obtained in experiments Da and DA are comparable with those given by Lettens et al. (2005) for the SOC change in southern Belgium from 1960 to 2000. Despite the large uncertainty associated with the observed SOC stocks, this suggests that experiments Da and DA provide a more realistic representation of the change in SOC stocks in this part of Belgium than simulations with lower resolution pan-European data.

Incorporating land use conversions

Overall, SOC changes from 1990 to 2050 are quite low for the ATEAM grid-cell at 2.33 t C ha−1 (da) to 2.01 t C ha−1 (DA). In general, the projected SOC changes arise from both climate change and LUC. A net gain in SOC stock as a result of LUC is expected in 2050. This gain is mainly because of the afforestation of 13% of the baseline agricultural area and is estimated to be about 1 t C ha−1. However, compensation between increases and decreases of SOC stocks following individual LUCs result in similar overall figures from the different experiments. If one focuses on the four types of LU conversion, it can be seen that the direction of change (i.e. increases or decreases) is the same for all three experiments (Table 4, Figure 5). However, larger changes in SOC stocks because of LUC are expected in experiments Da or DA. This is due to the difference between input data for the different LU types (Tables 1–3). Differences between the two LUs in experiment da are minimized by using the same initial SOC value in 1900 for both cropland and grassland and using this value to derive a plant input. In fact, the modelled decrease of 0.02 t C ha−1 year−1 following conversion from grassland to cropland and increase of 0.07 t C ha−1 year−1 following conversion from cropland to grassland (Table 4) are at the low end of the range reported in the literature. For example, in a meta-analysis, Guo & Gifford (2002) report equilibrium decreases in the SOC stocks of about 60% of the initial value following conversion from grassland to cropland and increases of about 20% following conversion from cropland to grassland. Both values have a fairly small confidence interval despite the range of studies considered. In a later application to European agricultural land, Freibauer et al. (2004) report potential soil carbon sequestration rates of −1.0 to −1.7 t C ha−1 year−1 following conversion from grassland to cropland and 0.6–3.1 t C ha−1 year−1 following conversion from cropland to grassland, based on analysis of long-term experiments reported in Smith et al. (1997).

In contrast to these plot studies, LU conversions were not implemented simultaneously for the entire study area. Instead a linear interpolation was used to calculate LUC at each time step (2020 and 2050) because no information was available from the scenario database about the timing of LUC and it is unlikely that all conversions will take place at the same time. As shown in Figure 6, the SOC stock of a single cell evolves quite rapidly when LU conversion occurs. However, the sum of changes for all cells results in a smoother linear relationship and SOC stocks do not reach equilibrium within the period considered. This also explains why the changes reported in Table 4 are lower than those cited by Freibauer et al. (2004).

Figure 6.

 Modelled change in soil organic carbon (SOC) for soil association 45, assuming climate change and conversion from cropland to forest.

In addition to time, the location of change is also important. In spite of using the same data sources, working at the 10′ level of aggregation or at a spatial resolution of 250 m by precisely locating LU conversions within the various soil associations does not lead to important differences in modelled SOC stocks. However, it is expected that when LUC is concentrated on certain soil types, the effect caused by the difference in clay content and initial SOC stock of each soil type will persist in the modelled SOC stock over time. Therefore, the ‘fine level of aggregation’ experiment (DA) is likely to provide more accurate results than the other experiments. Such a concentration of LUC might be expected to occur in areas of Belgium where there is a correlation between soil type and LU (e.g. valley bottoms soils are largely dominated by grassland). This issue needs further investigation.

Conclusion

The objective of this study was to examine the influence on estimates of SOC change of using coarse, pan-European databases compared with detailed, region-specific soil, LU and plant input information for a specific study area. Two levels of aggregation were also compared. The study showed that the simulations with regional datasets, with or without specifying the soil associations on which LU conversions occur, gave more pronounced changes in SOC stocks following LUC for the agricultural area of a small region of Belgium. Overall, the changes in SOC are limited as a result of the opposing effects of individual LU conversions, the limited extent of these conversions and their gradual occurrence over the 1990–2050 time period. However, the simulations using regional data project larger changes (−0.17–0.33 t C ha−1) than those based on pan-European data (−0.02–0.16 t C ha−1). The average SOC stock of the study area increased from 1.0 (experiment da) to 1.2 t C ha−1 (experiment DA), and the contribution of LUC was estimated to be between 40 (experiment da) and 60% (experiment DA) of the total change. These results indicate that the use of coarse resolution datasets may lead to estimates of SOC change that overall are relatively good, but that the effects of specific LU conversions are masked. However, a balance can be struck between using local input data and attributing these data to the specific CORINE cells undergoing LUC. Although this method did not improve the prediction of SOC in the work presented here, differences could be observed if the methods were to be applied to other grid cells with contrasting LU patterns. Other factors that were omitted in this study that could affect SOC stocks include technology and management change, and these effects need to be considered before generalizing the results of this study to wider areas.

Acknowledgements

This work contributes to the project ‘Carbon sinks in the dominant Belgian terrestrial ecosystems’ financed by the Federal Belgian Science Policy (OA/00/01). The support is gratefully acknowledged. The contribution of PS was supported by CarboEurope-IP (FP6 Integrated Project: Assessment of the European terrestrial carbon balance; http://www.carboeurope.org; GOCE-CT2001-00125).

Ancillary