Method for estimating soil carbon stock changes in Finnish mineral cropland and grassland soils

ABSTRACT This study presents a method for estimating the soil organic carbon (SOC) stock changes in mineral agricultural soils developed for the Finnish GHG inventory. SOC stock changes in mineral cropland and grassland soils from 1990 to 2013 were calculated by combining agricultural statistics and national conversion factors to estimate the organic inputs to soil, along with the Yasso07 soil carbon model. The effects of selected key assumptions on the simulation results were studied. The method yielded SOC change estimates closer to the observed SOC change than were the results of the previously used Tier 1 method. The SOC stocks of croplands in 1-m soil profiles were slightly decreasing in most regions of the country. At the national level, the decrease was on average 0.05 Mg C ha–1 year–1 (0.01%). Selection of climate data (annual vs. long-term mean) and the initialization procedure had large impacts on the simulated SOC stock and change results, whereas the simulations at regional and subnational levels provided similar results. The method was found to be suitable for the GHG inventory and preferable to the Tier 1 method. The modular structure of the system allows for continuous improvements when more information and data are gathered.


Introduction
Soils hold the largest stock of terrestrial organic carbon (C) in the biosphere [1]. The majority of this soil organic carbon (SOC) stock is concentrated in the northern latitudes [2]. For example, SOC content in the European Union (EU) is clearly larger in northern countries like Sweden, Finland and United Kingdom [3]. This is mainly due to the vast area of peatlands in these countries, but the C stock of mineral soils is also higher in northern Europe. SOC stock development is mainly affected by climatic conditions and litter input driven by ecosystem production and human activities on each land-use type.
Parties to the UNFCCC [4] are required to report losses of SOC from managed soils as part of their greenhouse gas inventories. Guidelines of the IPCC for greenhouse gas inventories provide methods of different complexity (Tier levels) for the reporting. Tier 1 corresponds to the simplest default methods, Tier 2 employs country-specific parameters and the most complex Tier 3 methods apply measurements and/or modelling. The reporting system should be based on verifiable land-use and cultivation data and provide area-based yearly values for changes in SOC, with consideration given to the special characteristics of the country, such as climatic gradients [101]. The reporting should cover all land areas under human influence including land-use change (LUC). Uncertainty of the data has to be assessable and the calculated results should be transparent, complete, consistent, comparable and accurate [101].
In the Finnish greenhouse gas inventory, emission source categories cropland remaining cropland (CRC) and grassland remaining grassland (GRG) are identified as key categories [5]. SOC stock changes from Finnish mineral soils of CRC and GRG have so far been reported using the Tier 1 method, which is based on the assumption of default SOC stocks for each soil type and the IPCC default C stock change factors. On the other hand, SOC stock changes in mineral forest soils have recently been estimated for GHG reporting with a model-based Tier 3 method. In this method, National Forest Inventory (NFI) data [5] are used to calculate carbon inputs to soil and the soil model Yasso07 [6] to estimate SOC changes. A long-term goal in Finland has been to develop a uniform SOC modelling approach that is applicable to land-use classes on mineral soils, because land areas can change from one class to another and most realistic estimates are obtained when the calculation method does not change between classes. Recently, it has been reported that Yasso07 gives reasonable SOC change estimates for Nordic and continental agricultural soils [7À10]. The Yasso07 model, which was originally developed to require only a little easily available input information, is simpler than some other soil models, such as ECOSSE [11], C-Tool [12] or CENTURY [13], and it omits the soil management and texture effects on soil C, for example, that are included in some other models. The management of different crops is covered as far as it changes litter input amounts to soils. In Yasso07, the decomposition of litter and SOC depends on the chemical quality of litter inputs and on climatic conditions only. The model is based on wide empirical litter decomposition data from different ecosystem types [14], and the Bayesian calibration method applied in the model development process allows for the estimation of parameter uncertainties [6], which also supports its use in greenhouse gas inventories.
The aim of this study was to develop a Tier 3 method for estimating the SOC stock change factors for mineral agricultural soils applicable in the Finnish greenhouse gas inventory. The method was used for calculating SOC stock changes in mineral cropland and grassland soils from 1990 to 2013 by combining agricultural statistics and the Yasso07 soil C model. The study contains a description of a national calculation system for the C inputs from crop residues and manure, an assessment of annual SOC and SOC change results highlighting the effects of annual versus mean climate data used in the calculations and a comparison of different spatial resolution levels as well as model initialization methods for the simulations. The method was built bearing in mind the principles of transparency and simplicity, and it allows for continuous improvement as new knowledge and data are gained.

Agriculture in Finland
The majority of Finnish agricultural land area lies between the latitudes 60 and 65 N (Figure 1). The mean annual temperature of the southern part of Finland is about C5 C and in the northernmost agricultural area about C3 C. Mean annual precipitation varies typically between 500 and 700 mm. The most important crops and their recent average annual yields are silage (16 800 kg ha À1 ), barley (3600 kg ha À1 ), oats (3300 kg ha À1 ) and spring wheat (3700 kg ha À1 ) [102]. Historically, most arable land has originally been either forest or mire. The greatest degree of conversion from forest to cropland took place at the turn of the 20th century [15]. Soils in Finland are generally acidic, requiring liming to maintain their productivity.
Agricultural land areas of all cultivated crops, green fallow and grass leys in crop rotation were categorized as cropland following the IPCC guidelines [101]. This includes most forage grasses that are commonly cultivated in rotation with annual crops in Finland. Abandoned fields, extensively cultivated grasslands, ditches wider than 3 m and reed canary grass with infrequent renewal were categorized as grassland. Because the agricultural support system favors green fallows and open fallows are uncommon, in this work we consider fallows vegetated ecosystems where the biomass is not harvested, and we call them green fallows hereafter. The majority (90%) of cropland and grassland is on mineral soils. Organic soils were excluded from this study as the soil C model Yasso07 is only applicable to mineral soils.

Calculation method
The calculation method is based on agricultural statistics of crop yields and land cultivation areas ( Figure 2). Hectare-based SOC stocks and stock changes are estimated with the aid of soil C modelling. The litter production from crops to soil is estimated using the approach proposed by Bolinder et al. [16], which uses national biomass conversion factors that estimate litter input based on crop yields. SOC changes per hectare À that is, SOC change factors À are used to calculate national SOC changes of different land use categories by multiplying them with land areas of corresponding categories. The approach is essentially the same as that used in the inventories of forest SOC balances [5,17].
In this study, the calculation method was applied to estimate the SOC changes in the categories CRC and GRG of mineral agricultural soils in Finland. Soil model Yasso07 [6,18] was used to simulate the SOC stocks and their changes using the calculated C input and climate data ( Figure 2). The calculations were done at the regional Centres for Economic Development, Transport and the Environment (ELY center) level ( Figure 1) for which the statistical data on crop cultivation areas, crop production and animal numbers were available [102]. The land-area data for land use categories were taken from the Finnish National Forest Inventories (NFI) that covers all land-use types. Details related to calculations for both land-use categories are described below.
Cropland remaining cropland SOC stock change factors for CRC were calculated based on simulated changes in SOC stocks from 1990 to 2013. The annual C input for the Yasso07 model was calculated for each ELY center as described below and fed into the model together with the long-term (1961À2013) regional average climate data. The annual soil C input from crop residues varies according to the regional cultivation areas for each crop, and the yield levels. Regional animal numbers are the basis for manure C input, which also varies annually.

Grassland remaining grassland
As the land use in this category containing mainly abandoned fields was extensive, no changes in the management-driven productivity during the inventory years were assumed. This yields a constant C input which, together with the average climate data, gives a zero SOC stock change estimate. However, the results of the initial state of GRG are required for the modelling of the soil C stock changes in the land conversion categories and are therefore reported here.

The Yasso07 model
The soil model applied in the study was Yasso07 [6,18], which describes the decomposition of organic matter based on information on climate and C input quality. The model and its predecessor Yasso [19] have been widely applied to assess the C balances of forest soils [e.g., 17,20] and Yasso07 is being used in the greenhouse gas inventory of Finland to assess changes in soil C stocks of forests [5]. However, litterbag measurements used in model calibration [14] also cover other land uses, and the tests of the model have shown that Yasso07 is able to predict the agricultural SOC changes in response to different organic amendments [8] and also in the case of afforestation and deforestation [7].
Yasso07 consists of four litter compartments and one humus compartment. The litter compartments are compounds soluble in a non-polar solvent, ethanol or dichloromethane (denoted using E); compounds soluble in water (W); compounds hydrolysable in acid (A); and neither soluble nor hydrolysable at all (Ns). Litter is divided into these compartments according to its chemical quality. Each compartment has a specific decomposition rate that is independent of the origin of the litter. Temperature and precipitation affect the decomposition rates. Mass flows between compartments and the formation of more recalcitrant humus is determined by decomposition rates and mass flow parameters ( Table 3 in [14]). The decomposition of woody litter in the model also depends on the diameter of the litter [21]. In this study, the original parameterization of the model ( Table 3 in [14]) was applied, as it was applied in the agricultural model tests [8,9]. The simulated estimates using this parameter set represented soil layers down to a depth of 1 m.

Model input
Estimation of the plant carbon input amounts. The plant C input calculation scheme followed the approach proposed by Bolinder et al. [16] for estimating the C input from plants to soil based on easily available and comprehensive agricultural statistics. For the method, a literature review using Finnish and Nordic sources was carried out to find the most appropriate parameters for all main crops ( Table 1). All crops were allocated to the most appropriate main crop categories; for example, soil C input of vegetables was calculated similarly as for peas, and the soil C input of perennial fruit trees and berry shrubs was assumed to be as for green fallow. As a result, the system covered all agricultural land areas.
C input from the residues of cultivated plants was calculated assuming a typical cultivating scheme for each plant group in Finland. For CRC that meant spring-sown cereal varieties with autumn harvest. Crop residues were assumed to be left in the soil, as typically in Finland they are chopped and left on the soil surface or returned to the soil after being used as animal bedding. The use of straws as energy in Finland has been minimal À less than 3000 ha annually [103] À and was therefore excluded from the analysis.
C input was divided into three compartments: C input from aboveground biomass (CI AB ), root biomass (CI RB ) and rhizodeposition (CI rhizo ). C input from aboveground biomass was assumed to consist of harvest residues (straw, leaves and stubble) and was calculated as follows: where C yield,i is the C content of harvested product of crop i and HI is the harvest index which is the ratio of harvested product to total aboveground biomass. The C content of the harvested product was calculated by multiplying the annual yield (kg ha À1 ) taken from national statistics [102] by dry matter (DM i ) and C contents (CC i ) of the product, which was assumed to be 0.45 according to Jensen et al. [22]. Losses of yields due to technical reasons (e.g., grains lost in harvest or during transportation or storage) were also taken into account by multiplying yields with a yield loss factor (LO i ; Table 1). National crop failure statistics were not applied in the study, as that data did not include information on the phase of failures À that is, whether they were due to failed sowings or devastated yields at a later phase in the season, for example. That information would have been needed to reliably estimate the plant litter input due to crop failure. C input from the root biomass of annual crops was assumed to be equal to their annual root biomass C and was calculated using: where SR i is the ratio of shoot and root biomass of crop i. Root biomass-derived C in root crops and potatoes were calculated similarly; that is, root biomass was assumed to be proportional to aboveground biomass. For perennial crops À mainly grass leys with perennial forages that belonged to cropland categories À root biomass C input was estimated simply by using the measured average dry matter of root biomass (RootDM): where CC is the C content of the root dry matter and L is the average length of continuous cultivation of the perennial crop before renewal of the ley or changing to some other crop in rotation. This simplified approach was taken as the shoot and root dry matter of perennial crops in our national and Nordic sources (Table 1) did not show any correlation (data not shown). The average length of rotational grass cultivation in leys was estimated to be about 3.5 years, based on reported average rotation lengths during the period 1995À2009 in the Land Parcel Identification System [104]. Rhizodeposition for both annual and perennial crops was estimated using: where TR is root turnover rate (1/year) and C RB,i is root biomass.
Estimates of C input in long-term grasslands and abandoned fields were derived from a study of biomass potential of abandoned fields and buffer zones [23]. Since green fallow with no biomass removal is the dominant way of fallowing, estimates for aboveground C input in fallowed land were based on biomass estimates for managed uncultivated fields, which were 5375 and 4845 kg dry matter ha À1 for south and north Finland, respectively. Belowground C input for grasslands and green fallows was assumed to be the same as for perennial crops.
Estimation of manure carbon input. Manure-derived C input (CI manure ) was calculated as follows: (5) where N i is the number of head of livestock species i and VS i is the average annual excretion of volatile solids in manure per head of species i. C content of manure (CC manure ) was assumed to be 50% of VS [105]. Livestock species were divided into 13 groups: dairy cows, suckler cows, bulls, heifers, calves, swine, sheep, goats, horses, laying hens, broilers, turkeys and fur animals (i.e., minks and foxes). For cattle, the amount of VS was calculated using the IPCC equation (Equation 4.16 in [106]). For most other animals (swine, sheep, goats, horses and poultry), IPCC default values were used. The numbers of livestock for ELY center regions were taken from national statistics [102]. Straw used for animal bedding was not reported separately with manure, as all crop residues were assumed to be left in the soil. All manure was assumed to end up in CRC as GRG only included extensively managed land areas.
Other organic amendments, such as peat or sewage sludge, were excluded from the study as their use in Finland is minor [5] and no reliable statistics were available to assess their amounts.
Carbon input quality. Yasso07 requires information on the C input quality in terms of proportions of AWEN fractions. This information for manure was taken from Karhu et al. [8]. For plant residues, the fractions were calculated based on a Nordic dataset [22], and conversion was performed using a common method to characterize the organic chemical composition of decomposing plant litter, namely the van Soest method [24]. The AWEN fractions used are provided in Table 2 [22,24].
Climate data. The climate data applied in the study were the 10 £ 10 km gridded monthly climate data provided by the Finnish Meteorological Institute [25] and were available from 1960 onward. Data were used to calculate the ELY center-specific mean annual temperatures, precipitation and temperature amplitudes between the warmest and coldest months using the grid points located inside each of the ELY centers. As agricultural land area is not equally distributed within the ELY center regions, the climate was calculated as a weighted average À that is, the climatic data of each grid point were weighted with the agricultural land area of the corresponding grid. The cultivated area in the 10 £ 10 km grid used in the weighting is shown in Figure 1. The climate for the subnational levels of Northern and Southern Finland was determined in a similar way as for ELY center division but using northÀsouth regional division.
Model initialization. Initialization of the model was carried out by assuming the soil to be in a steady state, with the average C input from plant litter and manure of the first 10-year simulation period (1990À1999) for each land-use category and mean climate data (1961À2013). Alternative approaches for model initialization were also tested to determine their effects on predicted SOC stocks and SOC stock changes (see the section on "Options for initialization of the soil model" below).

Testing key assumptions
Annual versus mean climate variables used in simulations The effect of the annual versus different average climate data on the simulated SOC stocks and stock changes was studied by comparing simulations performed with different climate input data. The base case also used in all other simulations assumed there was no climate impact À that is, both initialization and simulations were done with the same mean climate data (period 1961À2013). The comparison was done to simulations using annual climate input with the same initialization, and to simulations using different longterm mean climate data during the simulation period (mean climate 1991À2013). In the last case, the climate input for initialization was the average of a 30-year period before the simulations (1961À1990).

Spatial simulation unit
The SOC stocks and stock changes were also compared using two different regional divisions for SOC simulations: the ELY center level at which the agricultural statistics are provided, and a coarser northÀsouth subnational division (Figure 1). In the latter case, the C input and climate were aggregated to the subnational level before the simulations by weighting C input and climatic data with agricultural area in different ELY centers. The aggregation of climate data was again performed by weighting the climatic data of each grid point with the agricultural land area of the corresponding grid, and litter input was aggregated by weighting the input of ELY centers with their agricultural areas. In the former case, the same area-based weighting was used for aggregating the stocks and stock changes simulated at the ELY center level to subnational results.
Options for initialization of the soil model In this study, three alternative approaches were tested for the initialization of the model pools of the emission source category CRC. First, the soil was assumed to be in a steady state with the average agricultural litter input. Then, considering the relatively young age of cultivated areas in Finland, those simulations were compared with an approach in which the effect of historical LUC from forest to cropland was taken into account. This was done by starting the simulations from the estimated SOC stock of forests for 1900 and pre-running the model for a period of 90 years with agricultural litter input. The forest stock for 1900 was taken from Akuj€ arvi et al. [9], and it was 160 Mg C ha À1 down to a depth of 1 m. The third initialization tested was based on the observed stocks reported by Heikkinen et al. [26]. Their observations were for the topmost 15 cm soil layer, and those stocks were extrapolated for a 1-m layer using soil carbon profiles gathered from Finland (references and density functions are provided in the Supplementary material, Figure S1).

C inputs to soils
Cropland remaining cropland Figure 3 shows the mean C inputs from plants to soil and crop net primary production (NPP) of CRC category crops for the 1990À2013 time series of one ELY center, Varsinais-Suomi (region #2 in Figure 1). The highest NPP in that region was for sugarbeet (5900 kg C ha À1 year À1 ) and silage (4800 kg C ha À1 year À1 ), and the lowest for peas (2200 kg C ha À1 year À1 ). Regional and annual differences in the crop-specific NPP and C input figures were driven by yield differences. Generally, the interannual variation in yields was larger than the variation between regional mean yields (data not shown). The highest annual C inputs to soil per hectare in the Varsinais-Suomi region were for green fallows (3700 kg C ha À1 year À1 ), for which all aboveground biomass was assumed to be left in the soil, and mixed crops and wheat (2900 kg C ha À1 year À1 ). They were the lowest for peas (1500 kg C ha À1 year À1 ), and turnip rape and dry hay (both 1600 kg C ha À1 year À1 ). Mean plant C input into the CRC category soils of the ELY centers varied from 1700 kg C ha À1 year À1 in the Lappi (region #15 in Figure 1), the northernmost ELY center, to 2400 kg C ha À1 year À1 in Varsinais-Suomi, which is where most of the cultivated areas in Finland are located (region #2 in Figure 1). Plant C input varied annually (coefficient of variation varied from 3 to 10%) due to interannual variation in yields and cultivation areas. There were mostly no clear temporal trends in plant C input levels ( Figure 4). In some regions (e.g., Satakunta #3 and Pirkanmaa #5), the inputs were somewhat larger in the early 1990s due to a larger share of green fallows during those years. C input from manure applications was at a clearly lower level in comparison to plant C input. There has been a slightly decreasing trend in manure C input due to  . Carbon input to soil from plants (black line) and manure (gray line) for cropland remaining cropland as the national total (Tg C year À1 ; large cell) and by regional ELY centers (Centres for Economic Development, Transport and the Environment) (Mg C ha À1 year À1 ; small cells). The numbering of the ELY centers is as shown in Figure 1.
decreasing animal numbers. National annual plant C input ranged from 3.2 to 4.2 Tg C year À1 , and manure C input from 0.6 to 0.7 Tg C year À1 .
Grassland remaining grassland For GRG, the C inputs were assumed to be constant from year to year. The total C inputs were 3700 and 3400 kg C ha À1 year À1 for Southern and Northern Finland, respectively.

SOC stocks and SOC changes
Cropland remaining cropland National mean SOC stock of CRC over the study period simulated to the depth of 1 m was 48 Mg C ha À1 . Across the ELY centers, the simulated mean SOC stock of croplands varied from 44 Mg C ha À1 in Uusimaa to 57 Mg C ha À1 in Pohjanmaa (regional SOC stock trends are shown in Figure S2). The stocks were slightly decreasing in all regions except in Varsinais-Suomi (#2) and Pohjanmaa (#12). At the national level, the decrease was on average 0.05 Mg C ha À1 year À1 (0.01%). The greatest decrease was in Ahvenanmaa (0.06 Mg C ha À1 year À1 , 0.1%) and the greatest increase in Pohjanmaa (0.03 Mg C ha À1 year À1 , 0.05%). Interannual variation in SOC changes was high ( Figure 5), the widest range being in Uusimaa, where the highest SOC change (sink) was 0.4 Mg C ha À1 year À1 and the lowest (source) ¡0.6 Mg C ha À1 year À1 . The year 1998 in particular showed a C source for many regions due to low yields in that year that subsequently led to low C inputs in soils.
Grassland remaining grassland For GRG, constant C input into soil was assumed, which is why SOC changes depended only on the climate. The national mean SOC stock of grasslands calculated with the steady-state assumption was 68 Mg C ha À1 . Among the ELY centers, the simulated mean SOC stock of grasslands varied from 63 Mg C ha À1 in Uusimaa to 80 Mg C ha À1 in Lappi (data not shown).

Effects of key assumptions on results
Climate SOC stocks and their changes simulated with averaged climate variables were more stable than the simulations done with annual climate input at all ELY centers, as illustrated for the Varsinais-Suomi ELY center in Figure 6. For this one example region, the mean SOC stock and stock changes of CRC simulations made with long-term mean climate variables were 45.2 Mg C ha À1 and 0.02 Mg C ha À1 year À1 , respectively. The mean SOC stock and SOC stock change in CRC simulations carried out with annual climate data were 45.0 Mg C ha À1 and ¡0.00 Mg C ha À1 year À1 , respectively (Figure 6a, b). The standard deviation of changes simulated with mean climate data was 0.04, and with annual climate data it was 0.07 Mg C ha À1 year À1 . The effect of climate data can be seen even more clearly for GRG, where the C inputs to soils are stable over the time series (Figure 6c, d).
Even small changes in the climate mean data used for model initialization have an effect on the initial SOC stocks of the model and thereby affect both the Figure 5. Simulated soil organic carbon changes for cropland remaining cropland (Mg C ha À1 year À1 ). National mean SOC is in the large cell and regional results for ELY centers (Centres for Economic Development, Transport and the Environment) are in small cells (Mg C ha À1 year À1 ). These simulations were done using the mean climate of the period 1961À2013, and annual litter and manure C inputs. Initialization was done using the mean climate data and litter input of 1990À1999 to run the Yasso07 model to steady state. SOC: Soil organic carbon. simulated stocks' stock changes. Figure 6a and c shows that changing the period of mean climate data used in initialization from 1961À2013 to 1961À1990 increased the stocks of both CRC and GRG in Varsinais-Suomi by 3% and made soil a larger source (or smaller sink), particularly in the early years of the study period (Figure 6b, d).

Testing regional versus sub-national simulations
Simulations completed at a sub-national level (Southern and Northern Finland) yielded very similar SOC stock and stock change results to the simulations performed at the ELY center level and aggregated by weighting them according to the share of agricultural area within their borders (results shown in Figure S3). In Northern Finland, the initial stocks reached by the two simulation levels differed by 0.25 Mg C ha À1 (0.5%). The mean difference in SOC changes in Northern Finland was 0.002 Mg C ha À1 year À1 with the maximum being 0.005 Mg C ha À1 year À1 . For Southern Finland the differences were clearly smaller.

Initialization of the SOC simulations
The initialization method selected affected both the level of simulated SOC stocks (Figure 7a) and the stock changes (Figure 7b). Forest C stock was so high and the decomposition dynamic projected by the model with average climate of Varsinais-Suomi so slow that the stock after the 90-year pre-simulation period was still at the level of 82 Mg C ha À1 . The steady-state stock with mean agricultural litter input and climate for the same region was 45 Mg C ha À1 . Stock changes were on average 0.16 Mg C ha À1 year À1 lower (this meant that there were higher emissions or a smaller sink) when simulated with forest stock than when simulated with the steady-state assumption. Initiating the model in year 1990 with stocks estimated based on observations led to fast decrease of the stocks and, thereby, high emission estimates.

Reliability of C input data and parameters
Agricultural statistics provide a generally reliable source of yield and cultivation area data. According to the quality report of the agricultural statistics, the coefficient of variation of the yield estimates of the main cereals is approximately 1À2%, and for crops with a smaller share it is somewhat higher [102]. For years with harvest losses, the applied approach may underestimate the C inputs into soils in the case of low-quality harvests being left on the field. However, this effect could not be included reliably on the basis of the available statistics. In this study, the factors used to convert the crop yield data to the above-and belowground biomass C were mostly based on Finnish experiments, but some published data from other countries and the IPCC default values were also used when national data were not available ( Table 1). The conversion factors were comparable to those reported by Bolinder et al. [16] for Canada. National data for all crop groups used in the calculations were not found, and in those cases the factors for the most appropriate other crop groups were used. The representativeness of the available data in those cases and related uncertainties are difficult to assess. In addition, the same conversion factors were used for all crop varieties in all years, even though harvest indices of cereals vary across varieties and growing seasons, for example [27]. SOC change estimates for GRG that are currently done assuming constant C input over years could be specified if improved information on C inputs from extensive and uncultivated fields becomes available. Ultimately, the largest uncertainty in plant C input is probably related to factors determining the root biomass and root turnover [16], and this is the subject that has the least empirical data available. As it is known that the chemical quality and decomposition of roots differ from those of aboveground residues [28,29], the modelling results could be further refined by acquiring more data on the roots and treating the root biomass input separately in the model with their specific quality data.
The plant-derived C input values of this study were of the same magnitude as those determined with allocation functions created by Andr en and K€ atterer [30] for Sweden, based on a wide review by Kuzyakov and Domanski [31]. The results were also comparable to C input estimates calculated for Norway by Borgen et al. [32], or for south-eastern Germany by Wiesmeier et al. [33].
Uncertainties related to manure C input are mainly due to the estimates of the amounts of manure and volatile solids. Animal numbers are considered to be quite certain, but the method does not take into account differences in animal characteristics or their feeding. However, compared to crop residues, the significance of manure in terms of total C inputs to soils is small (Figure 4).
Uncertainty estimations for this type of calculation systems have been assessed by using Monte Carlo simulations that combine uncertainties of input data and model parameters [34]. For example, Karhu et al. [8] evaluated the performance of the Yasso07 model in predicting SOC changes resulting from different agricultural management, taking into account the uncertainty in the model parameters and model inputs. Ortiz et al. [35], on the other hand, estimated uncertainties related to SOC stocks and SOC changes in Swedish forests. In their analysis, the Yasso07 uncertainty range (95% confidence limit) for the mean Swedish SOC stock of forest soils was about 13% (9 Mg C ha À1 ). This study did not include such an analysis, so as to avoid an incomplete view of the related uncertainties [36]. A thorough analysis of related uncertainties with comprehensive expert assessments and uncertainty analysis is seen as a natural follow-up study. Instead, here we have tried to highlight important assumptions and decisions needed in the calculation procedure that had a considerable effect on the results. Reporting the sensitivities of results on such assumptions is often neglected in numerical uncertainty analyses.

Sensitivity of results to the climate data resolution and the method of initialization
Our results highlighted the difference due to the use of constant (mean) versus annual climate data and the trend effects resulting from the use of different climate periods for the initialization and the simulations ( Figure 6). The aim of the GHG inventories is to report on the anthropogenic effect of GHG emissions and SOC stocks [4]. If we consider the warming of the climate to be an anthropogenic effect, annual climate data with an increasing temperature trend should be used. However, if the aim is to restrict anthropogenic effects to agricultural management, the use of constant climatic data calculated as a mean of a time series is justified. Analogically, the use of the same climate data for the initialization and simulation period shows the effect of changes in C inputs, whereas a different mean climate period used in these reflects the effect of changed climate.
ELY center regions were chosen as the basic unit of our calculations due to the fact that agricultural yield statistics were available directly for those areas. The smaller areal unit is also preferable because of the more uniform climate within one unit. Based on the comparison ( Figure S3), the results achieved with simulations carried out at subnational level, as was done in the greenhouse gas inventory calculations of forest soils in Finland, for example, were very similar to those obtained using ELY centers as the basic calculation units. This indicates that the differences in climate within Southern and Northern Finland are not very large, and the simplified method would also be justified.
One example of an important assumption is the initialization of SOC compartments of the soil model that affects both simulated SOC stocks and SOC stock changes (Figure 7). The initialization of model pools with observed data is often not straightforward, since model compartments do not necessarily represent measurable SOC fractions [37]. In this case, the observed SOC stock values from the Finnish soil survey [26] extrapolated with national soil profile data (See Figure  S1) were also tested in the initialization, but the following model trajectory clearly showed unrealistically high C losses for the first simulation years (Figure 7). For this reason, initialization by assuming a steady state with historical litter estimates with optional spin-up periods was chosen as the preferred approach.
Due to the slow dynamics of humus decomposition in the Yasso07 model, it takes hundreds of years before the soil reaches steady state after a change in driving factors, input levels or climate. As large areas of forested land in Finland were converted to cropland at the start of the 20th century [15], the period of intensive crop cultivation in Finland is short in comparison to the time required to reach the simulated steady state with the model [9]. Here it was shown that initialization with only agricultural litter input results in lower SOC stocks and SOC change estimates than initialization using steady state calculated with forest litter input together with a pre-simulation period with the agricultural litter input, as shown for example by Akuj€ arvi et al. [9]. However, uncertainties in the resulting SOC stocks and data requirements related to the latter approach are high. Implementing this initialization method for Finnish cropland and grassland would be difficult due to the lack of information of the historical C input to soil. On the other hand, applying observed stocks in model initialization also showed a large source as the modelled projection quickly subsided toward an equilibrium determined by the input levels. Assuming that the main emphasis of GHG reporting is on changes due to human-induced impacts since 1990, we propose in that context to use initialization with the steady state using agricultural litter input. In addition, the purpose of the GHG reporting of CRC is to report the effects of management changes, not LUC. This justifies the initialization method that disregards the background effect of the forest preceding the period of cultivation.

SOC and SOC change of mineral agricultural soils in Finland
The SOC stock size in Yasso07 simulations is mostly determined by the initialization procedure together with the parameters related to the decomposition of the most slowly decomposing compartments. The empirical basis of these parameters is much narrower than those determining the dynamics of the faster model compartments [14], which increases the uncertainties related to SOC stock estimates. With the simplified initialization procedure using the steady-state assumption with agricultural litter input, the mean SOC of CRC to a depth of 1 m was 48 Mg C ha À1 . This is a low SOC stock estimate compared to the observed range of 105À160 Mg C ha À1 for the topmost 1 m of the Finnish agricultural soils reported based on a Finnish soil survey [26] and scaled for layers beneath the plow layer using national soil profiles ( Figure S3). Relatively poor compatibility between simulated and observed stocks ( Figure S4) may reflect the omission of the soil texture effect in Yasso07, since soil profiles showed a relation between soil type and stocks for deeper soil layers. In the study by Karhu et al. [7], Yasso07 could simulate C stock changes effectively, irrespective of soil texture, and it was deduced that clay content is not an essential parameter in boreal conditions with low temperature. This study suggests the opposite, but still leaves the question open until more information on the effect of clay content is acquired.
The general decreasing trend of SOC in the simulations is explained with decreasing C inputs in the time series 1990À2013 (Figure 4). The SOC change estimate for CRC, ¡0.05 Mg C ha À1 year À1 (0.01%) was somewhat smaller than the observed declining trend (¡0. 22 Mg C ha À1 , 0.2À0.3%) in Heikkinen et al. [26], which also reflects the missing effect of historical forest clearance when using agricultural litter input in initialization [9]. Thus, the simulated loss rate was considered a conservative estimate of the true loss, but a significant improvement to the Tier 1 method.
The Tier 1 method is based on the division of a few cropping types and soil management and input classes, as well as a coarse classification of climate regions. Our Tier 3 method takes into account the most important factors affecting soil C sequestration, the cropping type, the C input level and climate, but it omits the management practices. A significant advantage is that this method allows for the effects of dynamic changes in cropping systems and cultivation conditions to be taken into account. Probably the largest difference between the end results of the Tier 1 and Tier 3 methods was caused by omitting the soil management method (no-till and reduced tillage). The Tier 1 method used to indicate a relatively large C sink for mineral croplands, 0.08 Mg C ha À1 year À1 [5], which did not correspond to the results from field measurements indicating a loss of C in mineral soils in the period 1974À2009 [26]. The sink of the previous estimates was due to the use of high default C stock change factors for reduced tillage and no-till practices [101] that were considered too high for Finnish conditions. The effects of no-till practice and reduced tillage on SOC stocks are negligible in humid and cool climates [38À40] and thus it was found unnecessary to include their effect in the Tier 3 method, although these methods are widely applied (30% of the cropped area). The reasons for the low potential of SOC accumulation in minimum tillage practices may be the already high concentration of C in the soil [38] or the relatively high precipitation favoring decomposition in soil [41].
The method is comparable to the Tier 3 methods applied in the GHG inventory of Sweden [42] or Norway [32], for example. It is in line with the methods used in other land-use classes and land-use change classes in the Finnish GHG inventory. The results were clearly better than those obtained by the Tier 1 method, but correspondence with the observed C stock changes could be improved. The method can be further developed by using more accurate data on the input of C, and probably by including the soil texture effect. The modular structure of the calculation system allows for the use of multi-model ensembles to assess the model uncertainty in future uncertainty assessments.

Conclusions
Soil C modelling, together with national agricultural statistics as the basis for SOC change factors of CRC and GRG, has advantages over the previously used Tier 1 method. The method yielded SOC change estimates that are closer to the measured SOC change values than the results of the Tier 1 method, and it allows for analysis of factors driving the SOC dynamics. The method was developed with the principles of transparency and simplicity in mind. It synthesizes readily available national statistics and data of various kinds and allows for continuous improvement as new knowledge and data are gained. Model initialization and the temporal resolution of climate data affected the SOC change factors most among the assumptions tested here. It is clear that the results achieved are sensitive to such assumptions and the best assumptions depend on the purpose of the work. This study outlined the criteria for selecting the method that best serves the purposes of the national GHG inventory of Finland.