The approach followed to calculate the peat loss over the last 1,000 years is based on the main steps that are depicted in Fig. 3 and further detailed in the following. First, a historical elevation model was created. To estimate subsidence (elevation change) over that period, a present-day elevation model was subtracted from the historical one (providing a 1,000-year height difference map). Subsequently, areas that experienced elevation changes that could not be attributed to peatland cultivation were excluded. Then, a distinction was made between elevation changes that were caused by peat excavation, and by subsidence proper (i.e. resulting from drainage). In order to assess CO2 respiration, it is essential to distinguish between subsidence due to the oxidation of organic matter, which contributes to CO2 respiration, and consolidation, which does not. For that purpose, geotechnical data from the literature and unpublished records were compiled.
The large-scale approach required making a number of assumptions and simplifications. The most important simplification in this paper is that all deposits with more than 20 % of organic matter by weight are referred to as peat, including deposits that are normally classified as clayey peat, muck, organic mud, organic matter detritus, or gyttja deposits. This corresponds to the whole range of organic deposits defined in international soil classification standards (e.g. Berendsen and Stouthamer 2000; USDA 2005). An important assumption is that bulk density and organic matter content values derived from current deposits are also valid for peat that has now disappeared. Finally, all emitted soil carbon (C) has been converted into CO2, even though it may have partly been emitted in another form, for instance methane (CH4).
For current elevation (further referred to as DEM2005), the 2005 version of AHN (Actueel Hoogtebestand Nederland) was used, which is a public domain, high-resolution digital elevation model published by the Ministry of Infrastructure and the Environment (Anonymous 2016). It is based on airborne laser-altimetry data, discretised on a grid of 1 × 1 m, the elevation values of which have a vertical precision of 5 cm.
Creating a historic elevation model
As a first step to create an elevation model for AD 1000 (further referred to as DEM1000), the extent of the coastal plain at that time needed to be determined, which was defined as all land in the present coastal plain (grey area in Fig. 1) that is currently below the mean high water of AD 1000 (further referred to as MHW1000). On the basis of detailed palaeogeographic reconstructions, Vos and Knol (2013) put that level at 1 m above mean sea level (amsl) of AD 1000, which, in turn, is 0.3 m below current mean sea level (Van de Plassche 1982). As eustatic sea level rise over the past 1,000 years has been close to zero (Peltier 2002), the 0.3-m rise is entirely attributed to background subsidence, i.e. natural consolidation (auto-compaction), isostacy and tectonic movement (Kooi et al. 1998). Hence, MHW1000 is put at 0.7 m above current msl (i.e., 0.7 m above Dutch ordnance datum). This value was used to clip DEM2005 and define the study area (the coloured area in Fig. 1).
Once the extent of the coastal plain of AD 1000 was established, the historical peat landscape was reconstructed as a next step in creating a palaeo-elevation model for the coastal peatlands. It was assumed that the coastal plain of AD 1000 was still largely in equilibrium, so sedimentation and peat growth had been able to keep up with relative sea level rise until that time, and the entire surface elevation was above msl (cf. Pons 1992). Peatlands are complicated patchworks of different peat types depending on, among other factors, local hydrology, sediment inputs, vegetation succession and subsurface characteristics. For the purpose of this particular study, the only distinction that needed to be made was between bogs and fens, because of the difference in peat volumes held in these two landscape types.
Bogs are domes formed by peat growing above the regional groundwater level, creating relief, unlike fens, where growth is limited by the regional groundwater level, forming the lower elevated landscape. Previous studies used the extent of dykes and parcelling patterns to reconstruct peat bog colonization, and delineate the former peat bog area (Pons 1992); an example of this is given in Fig. 4. Another source of information is presented by nationwide palaeogeographic reconstructions of Vos et al. (2011) and Vos (2015), which are based on detailed sampling and dating of cores taken from the Dutch coastal plain. Both palaeobotanical and palaeogeographic reconstructions were used to arrive at a simplified historical bog and fen landscape, with some open water in between (Fig. 5). Where no information on the palaeo-environment was available, a (low) fen landscape was conservatively assumed to have been present.
There is an abundance of data on the height of peat bogs in current peatlands (e.g., Borger 1992; De Bont 2008), and height estimates from palaeo-landscape reconstructions (Pons 1992; De Bont 2008), which was used to transform the historic peat landscape to the desired elevation model (DEM1000). It was conservatively assumed that all peat bogs had reached an elevation of 2 m amsl (i.e., 1.7 m above Dutch Ordnance Datum to correct for background subsidence), with a range of 1–4 m. The fact that bogs have slopes was not accounted for in this study. The slight overestimation of height at the fringes of bogs due to this simplification is insignificant on the scale and resolution of this study.
A best estimate of fen height of 80 cm amsl (i.e. 0.5 m above Dutch Ordnance Datum), with a range of 50–150 cm, is similarly conservative. Where a fen or bog directly overlay outcropping Pleistocene deposits at the fringes of the coastal plain, the peat was assumed to have sloped parallel to the gradient of that surface, in accordance with palaeobotanical reconstructions of De Bont (2008) and Van Loon et al. (2009). The effects of uncertainty in the bog and fen height estimates were assessed by using the lower and upper elevation values in minimum and maximum volume scenarios, respectively.
Excluding subsidence outside peatlands
Not all differences between DEM1000 and DEM2005 can be attributed to subsidence caused by peatland cultivation. Water bodies in the area may have silted up, or enlarged due to erosion. In addition, artificially raised grounds and mining pits have altered the elevation of the current landscape. Existing topographic maps and a query of DEM2005 for outliers (e.g., sand pits having depths of 50 m) were used to identify and eliminate such areas and processes from the calculations made in this study.
Areas formed by clastic deposition, mainly tidal sediments, may have experienced local elevation changes due to sedimentation and erosion, and regionally due to ripening and consolidation. Although this shows up as elevation difference when subtracting the DEMs, these processes are not the focus of this study and were eliminated from the calculations. In addition, areas where peat oxidation was impeded by a clayey topsoil of at least 1.5 m were discarded. The extent of these areas was inferred using results from GeoTOP and NL3D, high-resolution national 3D models issued by the Geological Survey of the Netherlands. GeoTOP is a systematically produced voxel model, attributed with lithostratigraphic unit, lithologic class and their uncertainties, having a 3D resolution of 100 × 100 × 0.5 m (Stafleu et al. 2011; Van der Meulen et al. 2013; Maljers et al. 2015). GeoTOP does not have national coverage yet; where unavailable, its predecessor NL3D, which has a resolution of 250 × 250 × 1 m, was used (Van der Meulen et al. 2013). The resolution difference with GeoTOP is largely insignificant at the scale and level of simplification of the present study, and extra detail was obtained from the digital soil map of the Netherlands (De Vries et al. 2003) and the palaeogeographic maps of Vos et al. (2011) and Vos (2015).
Virtually all peat considered is the main lithologic constituent of the Nieuwkoop Formation; peat (including other organic deposits) also occurs as a subordinate lithology in the marine clastic Naaldwijk Formation (lithostratigraphic nomenclature cf. De Mulder et al. 2003). GeoTOP and NL3D allowed to query stratigraphically and lithologically in 3D, zooming in on peat within the depth range of interest (cf. Van der Meulen et al. 2005, 2007a). Exclusion from the volumetric calculations took place by clipping the differential elevation map with the extents of all the afore-mentioned features. In the resulting reconstruction, the Dutch coastal peatlands have an area of 7,874 km2 (including the extent of removed peat) and make up 47 % of the total extent of the coastal lowlands around AD 1000.
Two processes causing volumetric loss were distinguished (Fig. 2): peat drainage, which occurred in the entire coastal plain, and peat mining, which was undertaken in designated sites. This differentiation is relevant for CO2 emission estimates: one can safely assume that excavated peat has all been burnt, generating CO2, unlike volumetric loss caused by consolidation.
To estimate the contribution of mining to the total peat volume reduction, a database of nation-wide historical land use (De Bont 2008) was used to delineate former mining areas (Fig. 6). The classification and mapping were supported and supplemented by historical records of windmill constructions over the last centuries (Dutch Wind and Water Mill database), local land reclamation information (e.g. Van de Ven 1993) and parcelling patterns (cf. Fig. 4). Depending on the source(s) of information, a confidence class was assigned to each mining site: high (supported by historical documents); medium (most characteristic landscape elements are present); and low (only based on elevation or parcelling). High confidence and medium confidence sites constitute 61 and 30 % of the total extent of peat mining, respectively. The medium confidence category refers to sites for which it is unclear how much of it was created by mining, and how much resulted from subsequent erosion; however, because this erosion could only have happened after mining, this component was added to the volume lost by excavation. Most of the eroded peat will have been washed ashore, forming a deposit that is prone to oxidisation. A smaller, yet unknown amount was redeposited as detritus on the lake bottom, and may not have oxidised. For low confidence sites (9 %), volume loss could result from either post-mining or natural erosion. These sites were nonetheless included in the budget calculations; the maximum volume overestimation associated with this uncertainty is 2.5 %.
Distinguishing between peat oxidation and consolidation
After isolating mined peat, the focus was put on the areas where peatlands subsided solely as a result of drainage. Volumetric loss of peat caused by drainage is the combined result of shrinkage, oxidation and consolidation (Schothorst 1977). Shrinkage refers to the volume loss caused by the contraction of plant fibres and pores when peat is exposed to air. Shrinkage is largely irreversible, rehydrated peat will not regain its initial volume. Shrinkage happens fairly quickly and occurs only in the zone above the phreatic groundwater level. In the same zone, the introduction of oxygen increases the decomposition of organic matter, i.e. the peat oxidises. Peat oxidation persists as long as there is organic matter available above the phreatic water table, and as subsidence in cultivated peatlands will provoke more drainage, it becomes a self-perpetuating process, continuing as long as peat is available for oxidation.
Consolidation is volume loss caused by compression. The density of peat is so close to that of water that buoyancy significantly reduces autocompaction in a saturated peat column. However, once drained, the weight (load) of the peat above the lowered water table will lead to consolidation of the peat below the groundwater table. The weight of farming equipment or any structure that is put on top of the peat will cause additional loading and consolidation. After applying a load, most of the consolidation occurs within a few years, but visco-plastic deformation (creep) persists for decades.
Shrinkage, oxidation and consolidation all contribute to land subsidence, but only oxidation causes CO2 emission. In order to quantify the CO2 respiration, the contribution of peat oxidation to the overall loss of peat volume in the study area had to be estimated. This is not straightforward, especially in the Dutch case, where a cultivation history of 1,000 years had to be considered. Experimental drainage studies in agricultural-used peatlands (Schothorst 1977) show that as much as 50 % of all surface subsidence occurring during 6 years of deep drainage can be attributed to oxidation, 15 % to shrinkage, and 35 % to consolidation (referred to as compaction in the original paper). In these studies, maximum drainage depths between 50 and 100 cm below the surface were applied. Historical drainage depths will have been considerably shallower, on the order of ~20 cm (Kuhlman et al. 2010). The associated loading effect will have been accordingly lower, so the measured contribution of consolidation by Schothorst (1977) must therefore be considered a maximum for this study. When considered on time scales beyond those of available experiments, all shrunken peat will ultimately oxidise, so the volumetric reduction by shrinkage may be added to that by oxidation, attributing the original density to the aggregate volume. Experimental geotechnical studies indeed suggest that at present at least 70 % of the human-induced subsidence in the Netherlands is due to oxidation (Den Haan and Kruse 2006).
In this study, the amount of peat consolidation was quantified by comparing the dry bulk density of consolidated with fresh, unconsolidated, peat (Bird et al. 2004; Van Asselen 2011). This was done for a database consisting of 985 organic matter and dry bulk density measurements of peat samples, collected over the entire coastal plain. These samples were analysed by the Geological Survey of the Netherlands and by Utrecht University (Erkens 2009; Van Asselen 2010). The organic matter content (percentage by weight) was determined using the loss-on-ignition method as described by Heiri et al. (2001). Bulk density was measured by weighing and drying peat samples with a known volume (Van Asselen 2010). By multiplying the dry bulk density (kg/m3) of the sample with the relative organic matter content (%), the organic matter density is derived (kg/m3; Fig. 7).
Based on measurements by Van Asselen (2011), it is estimated that fresh, unconsolidated Dutch peat has an organic matter density of ca. 70 kg/m3. For a subset of peat samples which have a relative organic matter content > 20 % and range between fresh and consolidated, the method described by Van Asselen (2011) was used to calculate the amount of consolidation (expressed as percentage volume loss) that was needed to obtain the measured organic matter density. The method was not applied to samples that experienced extra consolidation as a result of burial and loading by thick (>150 cm) clastic deposits. Because of the burial depth, this peat is not likely to oxidise, and including these samples would increase the average organic matter density values beyond realistic extents. For the remaining 447 samples, the consolidation was plotted against the organic matter density (Fig. 8). The consolidation of individual samples ranges between 0 and 57 % depending on the organic matter density. The variation in consolidation for single organic matter density values is the result of the clastic content of the samples, as represented in the x-variation in Fig. 7; samples with lower clastic content consolidate more than samples with higher clastic content.
Calculating CO2 respiration
To calculate CO2 respiration (as a result of oxidation or combustion), the amount of organic carbon per volumetric unit is needed. Organic matter density is independent of the relative organic matter density for samples with a relative organic matter content > 20 % (Fig. 6). This implies that clay and silt content, which is highly variable in the Dutch subsurface, has no impact on the amount of organic matter per unit volume, and it is justifiable to attribute a single value per scenario for organic matter content to all organic deposits in the Dutch coastal plain. The average organic matter density of peat in the Dutch subsurface is 103 kg/m3; minimum and maximum scenarios have values of 80 and 150 kg/m3, respectively. When the carbon content of organic matter is assumed to be 50 % (cf. Kuikman et al. 2003), the resulting average carbon density for peat becomes 52 kg/m3, a value that compares well with literature data (e.g. Gorham 1991; Page et al. 2010). This value was used to convert the loss volume attributed to oxidation and combustion to calculate the total mass of soil carbon (soil-C) lost. Ratios of 1.00:3.67 for the conversion of soil-C to CO2, and 1.00:0.47 for the conversion of CO2 to atmospheric carbon in parts per million in volume (cf. Van den Bos 2003) were subsequently applied.
Consolidation values were assigned to the density scenarios using Fig. 8. In the minimum density scenario, consolidated organic matter density is 80 kg/m3, which means that 7 % of the total volume loss in drained areas was the result of consolidation (and 93 % due to oxidation). In the maximum density scenario, organic matter density of the lost peat is 150 kg/m3, in which case consolidation caused 50 % of the total volume reduction. A best estimate scenario uses the average organic matter density (103 kg/m3) and an associated consolidation value of 28 % (and a contribution of oxidation of 72 %). These density scenarios were added to the volumetric scenarios, arriving at a total of seven scenarios: best estimate (best estimates for volume and density), minimum and maximum density (with best estimate for volume), minimum and maximum volume (with best estimate for density), and absolute minimum and maximum (all parameters set at minimum or maximum, respectively).