JEQ Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 1 May 2008
Published in J Environ Qual 37:1209-1219 (2008)
DOI: 10.2134/jeq2007.0438
© 2008 American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Agricola
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Related Collections
Right arrow Nitrogen
Right arrow Global Change
Right arrow Spatial Variability
Right arrow Upscaling
Right arrow Landscape-Atmosphere Interactions

TECHNICAL REPORTS

Landscape and Watershed Processes

Effect of Land Cover Data on Nitrous Oxide Inventory in Fen Meadows

Linda Nol*, Peter H. Verburg, Gerard B. M. Heuvelink and Karin Molenaar

Land Dynamics, Environmental Science Group, Wageningen Univ., P.O. Box 47, 6700 AA Wageningen, The Netherlands

* Corresponding author (linda.nol{at}wur.nl).

Received for publication August 21, 2007.

   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 Materials and Methods
 Results and Discussion
 Conclusions
 REFERENCES
 
Landscape representations based on land cover databases differ significantly from the real landscape. Using a land cover database with high uncertainty as input for emission inventory analyses can cause propagation of systematic and random errors. The objective of this study was to analyze how different land cover representations introduce systematic errors into the results of regional N2O emission inventories. Surface areas of grassland, ditches, and ditch banks were estimated for two polders in the Dutch fen meadow landscape using five land cover representations: four commonly used databases and a detailed field map, which most closely resembles the real landscape. These estimated surface areas were scaled up to the Dutch western fen meadow landscape. Based on the estimated surface areas agricultural N2O emissions were estimated using different inventory techniques. All four common databases overestimated the grassland area when compared to the field map. This caused a considerable overestimation of agricultural N2O emissions, ranging from 9% for more detailed databases to 11% for the coarsest database. The effect of poor land cover representation was larger for an inventory method based on a process model than for inventory methods based on simple emission factors. Although the effect of errors in land cover representations may be small compared to the effect of uncertainties in emission factors, these effects are systematic (i.e., cause bias) and do not cancel out by spatial upscaling. Moreover, bias in land cover representations can be quantified or reduced by careful selection of the land cover database.


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
Materials and Methods
 Results and Discussion
 Conclusions
 REFERENCES
 
EVERY land cover map or database is a simplification of the complexity of a real landscape (Arbia et al., 1998; Regnauld, 2001; Schmit et al., 2006). However, the scale and mapping technique are a source of variation when comparing different land cover maps (Ellis, 2004; Bach et al., 2006; Schmit et al., 2006; Verburg et al., 2006). Differences between a land cover database and a real landscape are a source of error when the database is utilized (Foody, 2002; Fang et al., 2006; Fassnacht et al., 2006). The large dependence of greenhouse gas emissions on land use makes land cover data an essential input in greenhouse gas inventories (Kern et al., 1997; Plant, 1999; Denier van der Gon et al., 2000; Matthews et al., 2000). Recently Huffman et al. (2006) acknowledged the need for highly accurate, high-resolution, and nationally consistent land cover data, while others have argued for statistically rigorous and accurate assessment of thematic maps (Stehman and Czaplewski, 1998; Heuvelink and Burrough, 2002). A lot of research has been performed to improve greenhouse gas inventories (Li et al., 1992; Kroeze, 1994; Denier van der Gon and Bleeker, 2005; Stacey et al., 2006), but careful analysis of how systematic errors in land cover data affect these inventories has received little attention. Often considerable emphasis is given to the provision of the most exact input data possible for soil and climate while little thought is given to the quality and accuracy of land cover or land use data (Jansen, 1998; Bach et al., 2006). Bareth et al. (2001) noted that the accuracy of spatial data should be regarded with more importance in the estimation of N2O emissions.

Signatories to the Kyoto Protocol (UNFCCC, 1997) must annually report emissions of their greenhouse gases CO2, CH4, and N2O. The Intergovernmental Panel on Climate Change (IPCC) has established Good Practice Guidelines for reporting and upscaling national greenhouse gas emissions. The inventory methods are divided into three levels of increasing complexity and classified as: Tier 1, Tier 2, and Tier 3 (IPCC, 1997; IPCC, 2000). The Tier 1 and Tier 2 methods are based on the assumption that the total N2O emission is a summation of emissions from different sources, which can be estimated by activity data (data on production and consumption of N) multiplied by an emission factor (the fraction of soil N emitted as N2O). The Tier 1 method is the most basic method and uses default emission factors and spatially coarse activity data. The Tier 2 method consists of the same basic relationships but is based on emission factors and activity data which are country-specific for the most important land uses and activities. The Tier 3 method on the other hand requires nonlinear process modeling. Examples of frequently used process models, as part of the Tier 3 method, are DNDC (Li et al., 1992; Li, 2000) and CENTURY (Parton, 1996).

Many countries are still striving to fulfill the Kyoto reporting requirements (Brown et al., 2002; Bolan et al., 2004; Saggar et al., 2004). Especially problematic are inventory methods for N2O emissions from agricultural soils (Lokupitiya and Paustian, 2006). For the Netherlands Kuikman et al. (2004) stated that current reporting to the Kyoto protocol is incomplete or inaccurate: several sources may not have been identified and others may well be reported incompletely. Accordingly, it is important to focus on decreasing the uncertainty and improving data quality of N2O emissions from agricultural soils. An important source of N2O emissions from agricultural soils is the emission from ‘cultivation of histosols’, which differs from estimation of other agricultural N2O sources because it requires spatial input data. Cultivation of histosols leads to oxidation of organic matter from peat soils due to the lowering of ground water tables in cultivated areas. Emission of N2O from cultivated histosols in the Netherlands has been estimated to contribute 10% of the direct N2O emissions from soils and 5% of the total N2O emissions from agriculture (Klein Goldewijk et al., 2005). Histosols cover a significant area (approximately 9% of the land surface) in the Netherlands (Kuikman et al., 2005; CBS, 2007) and are predominantly situated in the fen meadow landscape in the western part of the Netherlands. The main landscape elements of Dutch fen meadow systems are grassland parcels, ditches, and ditch banks, each with specific emission characteristics (Best and Jacobs, 1997; Van Beek et al., 2004).

The estimation of land surface area occupied by histosols and the main landscape elements depend on the available spatial input information and associated resolution. The scale of analysis or kind of information an investigator desires also influences the outcomes of the inventory. For example, if an investigator can choose between different land cover databases, each with a different resolution, then the choice for a certain database depends on the element of interest (Woodcock and Strahler, 1987). The optimum scale of analysis is usually the scale at which processes, in this case N2O emission, occur (Allen et al., 1984). Denitrification and nitrification are the most important processes in converting N into N2O in soils (Firestone and Davidson, 1989). These processes take place at the microbial scale, whereas national inventories require emissions to be reported on a national scale. These inventories are often based on emission factors derived from small-scale chamber measurements (0.03–6 m2). The chamber measurements in fen meadow landscapes have mainly taken place on grassland parcels, preferably not too close to the ditch (Ambus and Christensen, 1994). Since different landscape elements have different emission characteristics, it is worthwhile estimating the surface area of the different landscape elements using land cover databases and investigating the effect of using these land cover databases on the N2O emission inventory. The objective of this paper was to analyze how different land cover representations potentially introduce systematic errors into the results of regional N2O emission inventories. To this end five different land cover databases with differences in spatial resolution and accuracy were used in combination with four emission inventory methods. Understanding the influence of land cover databases on regional emission may help in the further refinement of reporting protocols.


   Materials and Methods
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results and Discussion
 Conclusions
 REFERENCES
 
Nitrous oxide emissions were calculated using different upscaling methods based on alternative land cover databases for two representative landscapes in the Dutch fen meadow system. Regional implications of using alternative land cover databases were analyzed by scaling the results up to the fen meadow landscape of the western Netherlands.

Study Area
The Dutch fen meadow landscape (Fig. 1 ) was formed in the Holocene, when the sea level was rising and large areas of the Netherlands were covered by swamps. In medieval times, this land was reclaimed for agricultural use. A popular way of reclaiming the land was by ‘cope agreements’. In these agreements, the length of the parcel was usually prescribed to be about ten times the width of the parcel. This pattern of parcels with the same shape is still recognizable in the fen meadows. However, a common strategy was not applied everywhere resulting in areas with more irregular reclamations. Between the sixteenth and the nineteenth century oligotrophic peat was excavated and used as fuel. Today, lakes and grassland intersected by wide ditches are located in these areas. The western fen meadow landscape exists primarily of grassland on peat soils and is intensively managed and owned by dairy farmers; however, more and more grassland is being extensively managed by nature organizations. We assumed the fen meadow landscape to be located on peat soils according to the Dutch soil map 1:50,000 (De Vries et al., 2003a). The borders of the western fen meadow landscape, located on peat soils, were assumed to be the ‘IJsselmeer’ at the north eastern boundary and the line dividing land below and above sea level as the south eastern boundary of the landscape.


Figure 1
View larger version (30K):
[in this window]
[in a new window]

  		Fig. 1. Location of the research polders in the Netherlands.

 
The Zegveld and Oukoop polders (Fig. 1) were chosen as case studies of the two most dominant reclamation types within the Dutch fen meadow landscape: cope (regular) and irregular reclamation.

The Zegveld polder (52°08' N 4°48' E) has a surface area of 670 ha (Fig. 1) and is representative of the ‘cope’ reclamation type. The area was reclaimed in the eleventh century. The village of Zegveld (Fig. 1; south corner) was the reclamation base from where the reclamation of the area started. The parcels stretch from Zegveld to the peat river Meije bordering the north of the polder. Many farms have settled in the center of the parcels. The polder was one of the latest reclamations in the area, which gave the polder its peculiar shape. At the reclamation base in the south the parcels are narrow becoming wider toward the north. The polder is predominantly drained 60-cm depth but an area of natural vegetation (25 ha) in the northwest is drained at approximately 30 cm below surface level.

The Oukoop polder (52°03' N 4°43' E) is smaller than the Zegveld polder and has a surface area of 168 ha (Fig. 1). The area was reclaimed in the eleventh or twelfth century and is representative of the reclamation type with wide ditches. The polder was enclosed by reclamations from the Hollandse IJssel river in the south, the Oude Rijn river in the north, and an old stream (the ‘Oude Wetering’) in the east. Both peat soils are classified as Terric Histosol and originate from wood and reeds.

Land Cover Databases
Surface areas occupied by grassland parcels, ditches, and ditch banks were estimated for the two research polders and for the entire western fen meadow landscape based on five land cover databases. The emissions of nitrous oxide from the Dutch fen meadows differ with elements in the landscape. Therefore, these landscape elements were separately accounted for in the analysis. The five land cover databases used, more or less ranked in order of decreasing resolution and accuracy, were: a detailed field map unit of the distinguished landscape elements, Top10Vector, LGN4, CBS soil use, and CLC2000.

Fang et al. (2006) pinpointed the importance of taking the uncertainty of land cover databases into consideration when using these for landscape studies. The five databases used in this study differ in uncertainty. Uncertainties in vector databases can be subdivided into geometric uncertainty and thematic uncertainty (Heuvelink et al., 2007). Geometric or positional uncertainty is uncertainty about the shape and the location of an object. Thematic uncertainty is uncertainty about the attribute values of an object and occurs in both vector and raster data. It is mainly caused by interpolation errors and wrong classification of pixels or mapping units (Bolstad and Smith, 1992; Foody, 2002; Steele et al., 1998; Van Oort, 2005). The resolution or minimum mapping unit of the land cover database is a source of geometric uncertainty (Hengl, 2006). This problem, the modifiable areal unit problem, is especially problematic when there are discrete changes within landscapes. Depending on the resolution and shape of data elements, almost any result may be obtained (Openshaw, 1983). In this paper the effect of the differences in geometry and resolution on the estimation of the prevalence of the different landscape elements, important to nitrous oxide emission, was evaluated. Details about the used land cover databases are given in Table 1 .


View this table:
[in this window]
[in a new window]

  		Table 1. Characteristics of land cover databases.

 
The goal of the field map was to accurately delineate ditches and ditch banks (positional uncertainty < 0.2 m in width), thus yielding the surface area of these landscape elements with negligible bias (i.e., much smaller than bias associated with the four commonly used databases). The aim was to measure all ditch widths in the polder, but due to inaccessibility a number of ditch widths had to be estimated. In polder Zegveld 91% of all ditches were measured, 8% were estimated in the field, and 1% was estimated using the Top10Vector and aerial photographs. In polder Oukoop 68% of all ditches were measured, 12% were estimated in the field, and 20% were estimated using the Top10Vector and aerial photographs. The boundary between ditch bank and grassland was defined as the line that separates areas with a clear slope gradient from those without a slope or with minimal relief (slope < 1°). The surface area of ditches and ditch banks were then calculated using the widths from the field map and the lengths from the Top10Vector topographic database. This was acceptable because the bias of the Top10Vector in ditch lengths was small compared to the bias in ditch widths. The Top10Vector was used as the basis for the field map, the ditches were adjusted to the measured ditch widths and ditch banks were added. In the database resulting from the field map, a distinction was made between intensively and extensively managed grassland. The extensively managed grassland was managed by a governmental organization, and was unfertilized and grazed by sheep and beef cattle. The grazing pressure is lower than on the intensively managed grassland, used for dairy cattle.

The Top10Vector database is a detailed topographic database of the Netherlands made by the Dutch National Mapping Agency (TDK). The Top10Vector is a vector file with a closed field structure; built up from coded lines enabling the user to select fields with certain characteristics. The Top10Vector is based on aerial photograph interpretation in combination with field investigation. It consists of several point, line, and polygon layers. The database is partly updated every year and the entire Netherlands is updated each 4 yr. The geometric uncertainty of the Top10Vector database is estimated at 2 m (Van Buren et al., 2003).

The CBS soil use database consists of soil use areas and boundaries. For agricultural land cover the only distinction made is between horticulture under glass and other agricultural use. The Top10Vector was used for the basic geometry (water, railroads, and roads). The geometric uncertainty of the topography is also 2 m (CBS, 2002). The main difference is the larger minimum mapping unit of the CBS soil use database (Table 1), which leads to an additional source of geometric uncertainty. In the analysis of the results, the linkage between the two databases was taken into account. The CBS soil use database provides insight into the distribution of different soil use types in the Netherlands and is used by the Statistics Netherlands (CBS) for deriving surface area and density statistics for regional classifications.

The LGN4 is a land use database for the Netherlands and is based on satellite images from 1999 and 2000 (De Wit, 2001). The LGN4 exists of grid data and vector data of crops. The grid data contain the dominant land cover type per 25 by 25 m grid cell. In total 39 land cover types are distinguished. In this research only grid data were used, because only surface areas water and grassland are of interest. The main difference between LGN4 and the CBS soil use database is that LGN4 focuses on agricultural land cover whereas CBS soil use is more focused on urban land cover. The category agriculture is split into ten classes and the category nature has seventeen classes where a distinction is made between intensively and extensively managed grassland. Validation of the LGN4 was estimated by checking 4000 points using aerial photos and the Top10Vector. The overall thematic accuracy of the LGN4 was estimated to be 92.2% (GeoDesk, 2006). However, large differences exist between classes. Classes with large abundances are generally more accurate than less abundant classes.

The CLC2000 database is produced by the European Environment Agency (EEA). The database was made as part of the project COoRdinate INformation on the Environment (CORINE). CLC2000 is a raster image, which has a resolution of 100 m. The CLC2000 is based on satellite images, which were interpreted by national teams. In the Netherlands vector databases of land cover (Hazeu, 2003) were developed for 1986 and 2000 where changes in land cover between these years were also mapped. The minimum mapping unit for these vector databases is 25 ha and for changes in land cover between 1986 and 2000 the minimum mapping unit is 5 ha. These national databases were joined together and converted into the raster database CLC2000 using the majority rule (Büttner et al., 2002). This database distinguishes 44 land cover classes. The thematic accuracy of the CLC2000 was estimated to be 87.0 ± 0.5% (EEA, 2006).

Nitrous Oxide Emission Estimation
For the Zegveld and Oukoop polders N2O emissions were estimated using four methods: IPCC Tier 1, Tier 2a, Tier 2b, and Tier 3. The IPCC Tier 1 method estimates emissions by multiplying global activity data by default emission factors (Table 2 ). The emission factor is the fraction of N emitted as N2O.


View this table:
[in this window]
[in a new window]

  		Table 2. Emission factors for N2O emission from agriculture for different tier levels.

 
Emission factors and activity data from the Good Practice Guidance (IPCC, 2000) and the IPCC Guidelines (IPCC, 1997) were used. When activity data were not indicated in the Good Practice Guidance, estimates from CBS (2007) were used. In the Tier 1 method land cover data are used for the estimation of the emission due to the cultivation of histosols. The estimated surface area of grassland on peat soil from each land cover database and the default emission factor were used. In the polder Oukoop, only negligible N2O emissions from ditches and ditch banks were measured (based on weekly closed chamber measurements by Schrier (personal communication, 2005, 2006, and 2007) see also Table 3 ). The emissions from ditches and ditch banks in polder Zegveld were also assumed negligible, because the soil, land use, and hydrological conditions in this polder are very similar to those of polder Oukoop.


View this table:
[in this window]
[in a new window]

  		Table 3. Nitrous oxide emission from grassland, ditch, and ditch bank. Emissions were non-continuously measured using flux chambers.

 
Two alternative specifications of the IPCC Tier 2 method are considered in this study hereafter referred to as Tier 2a and Tier 2b. The Tier 2a method uses activity data and emission factors as reported in the most recent Dutch inventory report (Klein Goldewijk et al., 2005) while the Tier 2b method uses polder-specific activity data (i.e., number of animals, amount of fertilizer used) gathered from door-by-door interviews with farmers. Five of the twenty farmers in Zegveld were interviewed; together they own 31% of the area in the polder. In Oukoop all eleven farmers were interviewed. All farmers could give animal numbers, separated into mature dairy cattle, yearlings, calves, sheep, lambs, goats, and pigs. Based on these interviews, an estimation of the total amount of cattle, manure amendment, and fertilizer application was made. For the Tier 2a method activity data from the municipality or agricultural region (CBS, 2007) were used. This information was scaled down to the scale of the research polders as follows. Agricultural activity data in the Netherlands, such as the number of cows and the amount of chemical fertilizers used in an area, are correlated to the amount of grassland in that area. Therefore, activity data for the polder were estimated by multiplying the activity data from municipality/agricultural region by the ratio of grassland in the polder to grassland in the municipality/agricultural region.

For the Tier 3 level inventory the process model INITIATOR (De Vries et al., 2003b) was used. The model was developed to represent the crucial processes in the N chain by simple process descriptions, calculated in yearly time steps. Input data were taken from the CBS (2007) concerning animal numbers, manure management systems, and fertilizer use. Inputs from the land cover databases were also used. Soil characteristics from the Dutch Soil map (Stiboka, 1969) and hydrological characteristics (Wolf et al., 2003) were added. INITIATOR uses a process model in which N2O emission is a function of denitrification and nitrification in the soil (De Vries et al., 2003b). Unlike the IPCC methods, the emission factors and denitrification and nitrification fractions vary as a function of soil type and groundwater level in INITIATOR (Table 2).

Analysis of variance (ANOVA) was used to analyze whether differences between land cover databases for the polders are significant and whether differences between inventory methods are significant.

Regional Upscaling
In addition to comparing the calculated emissions for the two research polders an assessment of the regional implications of the use of different databases was made for the entire Dutch western fen meadow landscape. The field map of the research polders was scaled up to estimate the surface areas of different landscape elements and derive emission inventories using the various approaches for the larger region. For upscaling the fen meadow landscape was subdivided into three different reclamation landscapes: ‘cope’ reclamations, reclamations with wide ditches, and irregular reclamations. This distinction was made because these three reclamation landscapes differ in the prevalence of landscape elements due to differences in the shape of grassland parcels and open water due to differences in reclamation history. All three reclamation types are common in the western fen meadow landscape. The Top10Vector database was used to assign the type of reclamation landscape to each of the 315 polders in the western fen meadow landscape. The surface areas of landscape elements found in the two research polders were used to estimate the distribution of these landscape elements in the fen meadows. Polder Zegveld was considered to be representative for ‘cope’ reclamation patterns with a regular pattern of predominantly rectangular parcels divided by small ditches. The length/width ratio of the parcels is approximately 10:1. The selection procedure for this type of reclamation is therefore based on the length/width ratio of the parcels. Polders where more than 70% of the parcels have a length/width ratio equal or greater to 10:1 were considered to be ‘cope’ reclamations. Twenty to thirty percent of the parcels in ‘cope’ reclamations have smaller width/length ratio, because these are situated at the edge of the polder or are dissected by a road. The second reclamation landscape can be described as polders with significant areas of open water. Usually these polders have wide ditches in between the parcels. Polder Oukoop was used as a reference polder for this reclamation landscape. The procedure to distinguish these polders at the regional level was based on the occurrence of open water and ditches wider than 3 m in the polders. Note that the Top10Vector database represents ditches smaller than 3 m as line elements. Polders with surface areas of water equal to or larger than 10% of the grassland surface areas, according to the Top10Vector, were therefore classified as reclamations with wide ditches. This percentage was derived from the standard width of parcels in this area and the average ditch width. The remaining polders were classified as irregular reclamations. A representative for this reclamation landscape is polder Menningweer where Molenaar (unpublished data, 2006) made a detailed field map. After classifying the polders based on the three reclamation landscapes and assignment of the accompanying surface areas of landscape elements, the total surface areas of grassland parcels, ditches, and ditch banks were estimated. These surface areas were used to estimate the total agricultural N2O emission from the Dutch western fen meadow landscape. For each land cover database the amount of grassland on histosols compared to grassland on mineral soil in each agricultural region (CBS, 2007) was calculated to estimate activity data such as amount of cattle and amount of fertilizer use. When emission factors derived from the Dutch situation were available (Klein Goldewijk et al., 2005), these were used. For some N2O sources, emission factors have not been determined in the Netherlands (i.e., for indirect emissions), and default emission factors were used.

The emission of the western fen meadows was also estimated using INITIATOR, based on data on soils, hydrology, and land use data from the STONE database (Wolf et al., 2003).


   Results and Discussion
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results and Discussion
Conclusions
 REFERENCES
 
Land Cover Representations
Representations of landscape structure in the different land cover databases are shown in Fig. 2 and 3 . The enlargements in Fig. 2 and Fig. 3 clarify the difference between the field map and the Top10Vector database. The Top10Vector database represented ditches smaller than 3 m as lines and did not make a distinction between ditch banks and grassland. The field map represented all ditches as polygons and was the only database that distinguished between ditch banks and grassland. The CBS soil use database ignored farmyards, farms, and small ditches due to the large minimum mapping unit compared to the field map and the Top10Vector. Only the village center of Zegveld was represented as a residential area. The LGN4 and the field map databases distinguished between intensively managed grassland and extensive managed nature area. For both polders, the LGN4 database recorded a considerable surface area of ‘urban in agricultural area’ compared to the CBS soil use and CLC2000 database. As a result, the LGN4 database recorded a reduced grassland area compared to the other databases. Raster databases have difficulties representing point and line features that are smaller or equal to the pixel size. The farms with farmyard in the polders are features that were represented by the LGN4 database as square and rectangular shapes, which were often different from their real shape (i.e., the field map). The coarse CLC2000 raster database showed the entire polder covered with grassland, except for one pixel in the Zegveld polder and for one pixel in the Oukoop polder.


Figure 2
View larger version (44K):
[in this window]
[in a new window]

  		Fig. 2. Representations of polder Zegveld using different land cover databases.

 

Figure 3
View larger version (39K):
[in this window]
[in a new window]

  		Fig. 3. Representations of polder Oukoop using different land cover databases.

 
Surface areas of (intensively and extensively managed) grassland parcels, ditches, and ditch banks as calculated using the different databases are given in Table 4 . Except for the field map the land cover data did not have a separate class for ditch banks. These were all classified as grassland. The field map showed the smallest surface area of grassland and the largest surface area occupied by water and ditch banks. The grassland surface area increased with increased minimum mapping unit for vector data and increased with increased resolution for raster data. This can be explained by the dominance of grassland which, when presented at coarser scales, results in a general overestimation of its prevalence (Moody and Woodcock, 1996; Schmit et al., 2006). In vector data, ditches are ignored when they are < 3 m (Top10Vector) or have a surface area < 1 ha (CBS soil use). In raster data, ditches are ignored when another type of land cover is more abundant within a pixel.


View this table:
[in this window]
[in a new window]

  		Table 4. Surface areas land cover types per land cover database.

 
Overestimation of land cover classes with large abundances has also occurred in other landscapes (Turner et al., 1989; Moody and Woodcock, 1996; Ellis et al., 2000; Fassnacht et al., 2006; Schmit et al., 2006). In the fen meadow landscape, grassland has a large abundance and therefore absorbs the other classes, especially for the databases with small accuracies and coarse resolutions. Other landscapes are less sensitive to aggregation errors (e.g., Turner et al., 1989). Moody and Woodcock (1995) analyzed a mountainous forested area in California and found an increasing prevalence of water with increasing resolution due to lakes with a high degree of aggregation situated sparsely across the landscape. The class "conifers" also increased on average by 20% when aggregating from a resolution of 30 to 100 m. They concluded that the large increase in this class was due to the spatial structure of moderately large patches. The results found by Moody and Woodcock (1995) are large compared with the 7 to 8% difference in grassland in our study areas between the LGN4 (25-m resolution) and CLC2000 (100-m resolution) databases. On the other hand, Bach et al. (2006) and Fassnacht et al. (2006) found smaller differences between land use classes when aggregating from 25 to 100 m. Van Oort et al. (2004) compared the LGN database with reference data from randomly chosen areas in the Netherlands. The reference data were based on cadastral information. The grassland surface area was 2.5% larger for the LGN database than for the reference data. We found larger differences between the LGN4 and the field map (20–22%). This is probably due to the fact that Van Oort et al. (2004) only estimated areas of grassland and crops, where we found a large difference due to the presence of ditches instead of grassland. In research where thematic errors are small, positional errors can be large (Bach et al., 2006). Fassnacht et al. (2006) found the class "broadleaf," which forms narrow linear features along rivers, to be particularly susceptible to changes in resolution. This is comparable to our findings of ditches and ditch banks. Ozdogan and Woodcock (2006) also noted that large landscape elements can support large pixels, but when the landscape elements of interest are small, fine resolution is needed to correctly estimate surface areas.

Nitrous Oxide Emission Estimates
Using inventory techniques based on the different IPCC tier levels we calculated N2O emissions with the calculated surface areas (Fig. 4 ). Bias in estimating the area of grassland was propagated in the calculated emissions. For all Tier levels and for both polders the most accurate database represented the smallest area of grassland and accordingly the smallest N2O emission. The N2O emissions from polder Oukoop are about four times smaller than N2O emissions from polder Zegveld, which is consistent with the difference in total surface area between the polders. The Tier 3 method produced the highest N2O emissions for both polders and for all land cover databases (Fig. 4d and 4h). Furthermore, the Tier 3 method showed the largest differences between emission estimates (24% for polder Zegveld and 33% for polder Oukoop) because this method was more dependent on spatial data. Estimated N2O emission per hectare ranged from 20.0 to 47.1 kg N2O ha–1 yr–1, which is comparable to the emissions found by Velthof (1997; Table 3).


Figure 4
View larger version (28K):
[in this window]
[in a new window]

  		Fig. 4. N2O emission from agriculture using different land cover data and different inventory methods.

 
For polder Zegveld the emissions of N2O estimated with the Tier 2b method (Fig. 4c), were higher than the emissions estimated with the Tier 1 (Fig. 4a) and Tier 2a (Fig. 4b) method. From the interviews, it turned out that more cattle were present in the polder than estimated from the municipality data (Tier 1 and Tier 2a). Another reason is that the dairy cattle had, according to the local data, spent more time in the meadow than global and Dutch numbers indicated.

For both polders, the smallest N2O emissions were obtained from the Tier 1 method (Fig. 4a and 4e). The emission factors in the Tier 2 methods were larger and caused higher emission estimations. In polder Oukoop the difference between Tier 2a (Fig. 4f) and Tier 2b (Fig. 4g) was small, indicating that the activity data from the CBS database were close to the activity data estimated from the interviews. Results from the Tier 3 method (Fig. 4d and 4h) were high for both polders compared to the other methods. The INITIATOR estimates for N2O emission are based on the amount of denitrification and nitrification. Because the peat soils in the western fen meadows landscape have excellent conditions for nitrification and denitrification, the emissions estimated by INITIATOR were much higher than the emissions estimated by other inventory methods. As shown in Fig. 4, differences between inventory methods were larger than differences between land cover databases, which was also confirmed by analysis of variance (ANOVA). For both polders, the emission estimates differed significantly between all inventory methods, except for polder Oukoop between methods Tier 2a and Tier 2b. Due to the high emissions estimated by the Tier 3 (INITIATOR) method, the differences between land cover databases were not significant, except for polder Zegveld between the LGN4 and CBS soil database.

Regional Extrapolation
The surface area distribution of grassland parcels, ditches, and ditch banks from the research polders were used to scale up to the entire western fen meadow landscape (Table 5 ). The three polders (Oukoop, Zegveld, and Menningweer) were assumed to be representative for all Dutch fen meadow polders. This is assumed to be correct for the purpose of examining the impact of scale bias in land cover data for estimating regional N2O emissions.


View this table:
[in this window]
[in a new window]

  		Table 5. Surface area distribution of landscape elements in research polders used as reference for upscaling.

 
The reclamation landscape with wide ditches contained about twice as much open water as the other two landscape types. The irregular reclamation landscape had the smallest share in ditch banks, which can be explained by the large abundance of square parcels compared to more elongated parcels in the other reclamation landscapes. Figure 5 shows a map of the western fen meadow including a classification of the polders in reclamation landscapes.


Figure 5
View larger version (33K):
[in this window]
[in a new window]

  		Fig. 5. Reclamation landscapes in the western fen meadow landscape.

 
Intersecting rivers and disappearance of peat due to peat excavation caused fragmentation of the western fen meadow landscape. In the southern part ‘cope’ reclamations were abundant. The irregular reclamation landscape is common in the northern part of the western fen meadow landscape, where there was no common strategy during the reclamation period. The reclamations with wide ditches were most abundant near locations where the peat was excavated for fuel use. The area of grassland based on the field map was considerably smaller than the other estimated areas of grassland (Table 6 ).


View this table:
[in this window]
[in a new window]

  		Table 6. Surface area of grassland in western fen meadow landscape.

 
The field map was used to estimate the extent of ditches in the western fen meadow landscape, where other databases reported larger areas of grassland. According to the Dutch Soil map 36% of the Dutch peat soils are situated in the western fen meadow landscape. In the current national inventory, the total surface area of grassland on organic soils equals 231,000 ha (Klein Goldewijk et al., 2005). Assuming that there are no meaningful nationwide differences between the proportion of land on peat soils occupied by grassland, this suggests that 83,000 ha of grassland is located in the western fen meadow landscape. This estimate is smaller than the estimates from the land cover databases, except for the field map estimate, which is 11% smaller.

In general, vector data are more suitable in representing distinct boundaries and clear landscape elements, whereas raster data are assumed to better represent natural phenomena with gradual boundaries, such as soils, vegetation types, and slopes (Star and Estes, 1990). The landscape structure of the western fen meadow landscape with predominantly sharp boundaries between landscape elements and with long narrow ditches was therefore best represented by vector data. Poor representation of line elements—especially ditches—in this landscape was a large source of bias by both vector and raster data. Note that the bias would be much smaller for landscapes with fewer line elements and larger patches of the same land use.

The N2O emission estimates for the Dutch western fen meadows are shown in Table 7 . The largest source of agricultural N2O emissions was the cultivation of histosols, which demonstrates the importance of this source. The highest emissions from this source were obtained with the CLC2000 database because of the larger estimated surface area grassland (Table 6). The total emissions were larger for Tier 2a than for Tier 1 largely due to larger ammonium losses according to the Tier 1 method. For the western fen meadow landscape the maximum difference between the land cover databases was almost twice as large as between the inventory methods. This difference was largely due to two sources of error. The first error was the varying activity data used for the fen meadow landscape. For the research polders most activity data were relatively constant for all land cover databases. Many Dutch activity data (e.g., number of cows) were reported per agricultural region without information about the distribution (e.g., the amount of cows grazing on mineral soils vs. grazing on organic soils). To estimate these activity data for the western fen meadow landscape estimates about the proportion of organic soils compared to the proportion of mineral soils in the agricultural regions from the land cover databases were used. These activity data, which varied between land cover databases, caused some differences in emission estimates. The second source of error was the bias in representation of landscape elements by the land cover databases.


View this table:
[in this window]
[in a new window]

  		Table 7. Emission of N2O estimated for the western fen meadows using the IPCC Tier 1 and Tier 2a method from different sources.

 
The N2O emission was also calculated using Tier 3 (INITIATOR) and input from the STONE database. The estimated emissions were about twice the emissions estimated with the Tier 1 and 2a method (data not shown). This was partly due to the high denitrification and nitrification estimated by INITIATOR, which was also identified at polder scale, and partly due to the use of STONE, which is a very coarse database (with a resolution of 250 m) compared to the other databases used for the Tier 1 and 2a methods.


   Conclusions
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results and Discussion
 Conclusions
REFERENCES
 
In this research the surface area of grassland was overestimated when using the land cover databases. When moving to a coarser resolution for raster data or to a larger minimum mapping unit for vector data, classes with large abundances ‘absorbed’ classes with small abundances. The choice of a certain land cover database can have drastic effects on N2O inventories, because differences between estimated surface areas sometimes exceed 20% and different surfaces have different emissions. Such differences do not only apply to our study sites; at the regional level the amount of difference is similar.

For the Zegveld and Oukoop polders, the differences in estimated N2O emissions were larger between the inventory techniques than between land cover databases. For the western fen meadow landscape as a whole, the reverse applied because errors in land cover data were mainly systematic errors (bias) and errors from the inventory techniques were mainly random. Bias is consistently in the same direction and does not cancel out when estimates are scaled up to larger regions; therefore, these systematic errors became more distinct for larger areas compared to random errors in emission factors. The effect of using a more detailed land cover database had the opposite effect of using a more detailed inventory method. Highest emissions were estimated using the coarsest land cover database and the most detailed inventory method and vice versa. Although focusing on the reduction of uncertainty by improving emission inventory methods may be efficient at the local scale, this study has shown that for large-scale inventories the careful selection, inventory, and use of land cover data may be as important in reducing inventory uncertainties. While significant effort has gone into improving emission factors and improving inventory techniques, this research demonstrated that with relatively little effort emission inventories can be improved by improving land cover data input.


   ACKNOWLEDGMENTS
 
This work was funded by the Dutch project "Climate Changes Spatial Planning (KvR ME1)". We want to thank Wim de Vries and Hans Kros for making the model INITIATOR available for this research and Jan-Cees Voogd for his help using INITIATOR. We also thank Arina Schrier, Matheijs Pleijter, and Gerard Velthof for making their valuable fieldwork data in the research polders available for this study.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results and Discussion
 Conclusions
 REFERENCES
 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.


   REFERENCES
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results and Discussion
 Conclusions
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Agricola
Right arrow Articles by Nol, L.
Right arrow Articles by Molenaar, K.
Related Collections
Right arrow Nitrogen
Right arrow Global Change
Right arrow Spatial Variability
Right arrow Upscaling
Right arrow Landscape-Atmosphere Interactions

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Soil Science Society of America Journal Journal of Plant Registrations The Plant Genome