2.1. Radiative Forcing due to Changes in Albedo and Greenhouse Gas Fluxes
 In this study, the net impact on the RF due to a change in albedo and GHG fluxes following the forestation of peatland is expressed as:
where RFΔalbLoc is the local radiative forcing caused by a change in the albedo of the forestation area, Aloc is the forestation area, Aglob is the surface area of the Earth ( = 5.1 × 1014 m2), and RFΔalb is the global RF due to the albedo change. RFΔghg is the radiative forcing due to the changes of GHG fluxes caused by the forestation, which is obtained as a sum of the changes in soil GHG fluxes (RFΔghgSoil) and changes in the CO2 sequestration in the trees (RFΔghgTree). The subscript Ch refers to the changed situation and Ref to the reference situation before the change. The albedo change takes place only in the forestation area and affects the local climate, but as the local radiation balance contributes to the global radiation balance, the albedo change also has global consequences. The GHG emissions, on the other hand, are diluted over the whole globe due to the long atmospheric lifetime of these gases. To make the RFs arising from the different effects comparable, all the calculations were made for one square meter of peatland (Aloc = 1 m2) and the local RF due to the albedo change was scaled by the Earth's surface area (equation (1)).
2.2. Sites and Data for the Calculation of Radiative Forcing
 The calculations were made for four site cases of forestry-drained peatlands in Finland, representing different climates and nutrient levels, with different tree stand successions: for one nutrient-poor and one nutrient-rich site both in southern and in northern Finland (hereafter referred to as south and north). Furthermore, every case had an original undrained mire as a reference case.
 In this study, different data and models were combined: global radiation and albedo data, simulated tree stand characteristics, biomasses and soil GHG fluxes, as well as models for estimating canopy cover, the accumulated carbon in trees from the biomass data, and the atmospheric concentration due to the gas fluxes, and subsequently radiative forcing.
2.2.1. Radiation Data
 Radiation data for the southern and northern Finland cases were taken from the Jokioinen (60°49′N, 23°30′E) and Sodankylä (67°21′N, 26°38′E) observatories, respectively, operated by the Finnish Meteorological Institute (FMI). Average daily global radiation was calculated from the data covering the years 1971–2000. At both sites, global radiation was measured with a Kipp&Zonen CM11 pyranometer.
2.2.2. Albedo Measurements
 Albedo was calculated as a ratio of the reflected and incoming SW radiation, as measured with a Kipp&Zonen CM7B pyranometer above the different ecosystems. Measurements were made at four sites: one open, i.e., a treeless site, and one forest in both the south and north. Measurements from the open sites were used directly to calculate the albedo for the undrained peatlands (both the nutrient-poor and nutrient-rich cases).
 In the absence of albedo data for an undrained peatland in the south, data were taken from a cultivated peatland located at Jokioinen (Table 1). There, the albedo was measured at a height of 5 m above the soil surface. The field was sown with barley (Hordeum vulgare L.) or with barley and undersown grass (mixture of Phleum pratense and Festuca pratensis) typically in May and harvested in in June (grass) and in August–September (grass and barley) [Lohila et al., 2004]. Since the summertime maximum albedo at the Jokioinen cropland exceeded 0.25, which is higher than the values typically measured at bogs [e.g., Berglund and Mace, 1972; Petzold and Rencz, 1975; Solantie, 1988; Kurbatova et al., 2002; Arneth et al., 2006], it was given a value of 0.15 between day of year (DOY) 126 and 304. Therefore, the Jokioinen data provided us with the seasonal dynamics and the albedo values outside this period.
Table 1. Characteristics of the Albedo Measurement Sites
|Site coordinates||60°54′N, 23°31′E||60°39′N, 24°21′E||69°08′N, 27°14′E||67°21′N, 26°38′E|
|Height above sea level (m)||104||129||155||179|
|Time period from which data available||2000–2003||2004–2007||1997–2007||2001–2007|
|Land use and site type||open cultivated peatland||drained nutrient-poor pine forest||open mesotrophic fen||pine forest on mineral soil|
|Soil type||peat soil||peat soil||peat soil||podzol|
|Dominant vegetation||bare soil, barley, grass||Scots pine||sedges, mosses, shrubs||Scots pine|
|Tree density (ha−1)||-||1290||-||2100|
|Tree volume (m3 ha−1)||-||130||-||93|
|LAI (season maximum)||5.5||1.3||0.7||1.2|
|Vegetation height (m)||0.5||15.3||0.4||12.7|
|Average time (DOY) of snowmelta||96||123||138||140|
|Mean albedo in July||0.244 (0.15)b||0.124||0.137||0.112|
 For the undrained peatland case in the north, the albedo was obtained from a mesotrophic fen at Kaamanen (Table 1). The microtopography comprised hummocks and hollows, the height of the hummocks varying from 0.3 to 0.8 m and covering about 40% of the fen [Aurela et al., 2002]. At hollows, the vegetation consisted of sedges and some mosses, while the strings were dominated by different shrubs, e.g., Ledum palustre, Empetrum nigrum, Vaccinium uliginosum, Betula nana, Rubus chamaemorus, V. vitis-idaea and Salix spp. The albedo was measured at a height of 5 m.
 For the forested sites in the south, the data were obtained from the Kalevansuo forestry-drained peatland (originally dwarf shrub pine bog), drained about 40 years earlier (Table 1). Albedo measurements were conducted above the canopy at a height of 21.5 m above the ground. The tree stand composition was uneven and consisted mainly of Scots pine (Pinus sylvestris L.) with some small-sized Norway spruce (Picea abies L.) and Downy birch (B. pubescens). Forest floor vegetation consisted mainly of hummock dwarf shrubs (V. vitis-idaea, V. myrtillus, E. nigrum, V. uliginosum, L. palustre and B. nana), sedges like Eriophorum vaginatum and different mosses.
 The albedo data for forested peatlands in the north was taken from a Scots pine forest (P. sylvestris L.) located on fluvial sandy podzol at Sodankylä [Thum et al., 2007, Table 1]. The albedo was measured on a mast at a height of 48 m above the ground. The forest has naturally regenerated after forest fires. The trees were mostly 55–80 years old with a few much older trees. The sparse ground vegetation was mainly composed of lichens (e.g., Cladonia and Cladina spp.), mosses (e.g., Dicranum spp.) and ericaceous shrubs (E. nigrum L., Calluna vulgaris (L.) Hull).
 Due to the very low radiation levels, especially in the north where there is a period of darkness, the albedo cannot be measured in a reasonable way in midwinter. The radiation sensors may also be snow covered for part of the time, lowering the reliability of the winter measurements. For this reason, upper limits of 0.4 and 0.8 were defined for winter albedos in forests and open peatlands, respectively, based on both our own albedo measurements and on those of Solantie . For the nutrient-poor site cases, it was assumed that such sites already contained some trees, typically pines, in their pristine state. The upper (winter) albedo limit in the nutrient-poor cases before drainage was therefore set lower, at 0.55.
2.2.3. Forest Biomass Data
 The tree biomass data for drained sites were derived from simulations separately for each four site cases similarly to Minkkinen et al. , using the MOTTI stand simulator [Hynynen et al., 2002; Salminen and Hynynen, 2001]. The data included basal area (m2 ha−1), mean height (m) and volume of the tree stand (m3 ha−1). The increase in these variables was simulated at 5 year intervals for a typical full rotation; this was taken as 85 and 100 years in the southern and northern cases, respectively, including thinnings, which for the nutrient-rich cases took place at ages of 40 and 65 years in the south and north, respectively. In the nutrient-poor cases, thinning was only conducted in the south at an age of 65 years. The carbon stocks in the stands were calculated from the simulated stem volume data separately for pine and spruce using models developed for forested peatlands by Minkkinen et al. . At undrained peatlands, the tree biomasses were assumed to stay unchanged.
2.2.4. Estimation of Canopy Cover
 Projected canopy cover, defined as “the proportion of the forest floor covered by the vertical projection of the tree crowns” [Jennings et al., 1999], was estimated, for albedo calculations, using empirical models and the simulated stand characteristics. The data consisted of 46 pine and 88 spruce-dominated sample plots in different parts of Finland, whose site nutrient levels were equivalent to those of our test sites. Canopy cover was measured with a Cajanus tube [Korhonen et al., 2006]. Nonlinear canopy cover models, having stand basal area and tree height as independent variables [Korhonen et al., 2007], were fitted separately for pine and spruce stands:
where CCp and CCs are the canopy cover percentages for pine and spruce, G is the basal area of the stand (m2 ha−1), H is the tree height (m), and N is a dummy variable for latitudes higher than 65°N. The standard errors for the pine and spruce models were 0.079 (r2 = 0.85) and 0.099 (r2 = 0.67), respectively.
 At high latitudes, the real shading effect of a canopy cannot be directly deduced from the canopy cover, which represents the proportion of vertical between-crown gaps. A better estimate of the probability of the collision between a photon and a needle can be obtained by using the angular gap fraction, which depends, for example, on the zenith angle, stand height and spacing [e.g., Davies, 1963]. Because of this, we estimated the relationship between the projected canopy cover and the fractional cover at zenith angles of 45–60°, which the sun typically reaches in Finland between April and August. We used measurement data from Hyytiälä in southern Finland (61°50′51″, 24°17′41″) [Suni et al., 2003] and Sodankylä in northern Finland (see section 2.2.2). The projected canopy cover measurements and the fractional cover data, estimated from hemispherical photographs, were obtained from 130 sample plots representing different species compositions and development stages. Based on these data, a nonlinear model was fitted as follows:
where FC45–60 is the fractional cover observed at zenith angles of 45–60° and CC is the measured projected canopy cover. The standard error of the model was 0.095 with r2 = 0.86.
2.2.5. Estimation of Albedo and Albedo-Induced RF Based on the Canopy Cover Data
 Albedo measurements from the open sites were used directly to calculate the albedo for undrained peatlands. Measurements from the forests were used to calculate the albedo for all the forested cases at the end of the rotation. The albedo for the time period between the drainage and the end of the rotation was estimated for each day of the year at 5 year intervals from the relationship between the measured albedo and the modeled fractional cover FC45–60 by assuming a linear relationship between these two. Using the daily albedo and daily average global radiation data, we calculated the annual average value of absorbed radiation (W m−2) during the whole tree stand rotation for both the undrained and forestry-drained cases at 5 year intervals. The difference between the latter two values equaled RFΔalbLoc.
 Calculating the albedo-induced RF from the data measured on the ground excludes the effects of the absorbing and reflecting properties of the atmosphere, particularly clouds. At the latitudes between 60° and 70°N, the RF observed during clear sky conditions at the top of the troposphere is approximately 20% lower than that observed on the ground. During cloudy conditions, the top-of-troposphere forcing drops on average to 50% of that observed below the clouds, mainly due to the back reflection by clouds (P. Räisänen, personal communication, 2010). Since quantifying the exact magnitude and dynamics of the relationship between the RFs observed at the tropopause and below the clouds is complicated, an average of these, 35%, was used. Consequently, the RF observed on the ground was multiplied by 0.65 to account for the atmospheric effects.
2.2.6. Soil Fluxes
 Annual soil CO2 balances for drained peatlands were estimated using the same soil respiration data [Minkkinen et al., 2007a] and models for litter production and decomposition as were used in the Finnish greenhouse gas inventory [Statistics Finland, 2009]. Litter and soil carbon dynamics were simulated for the four site cases with the Motti stand simulator [Salminen and Hynynen, 2001], while the average carbon store change for the rotation period was used as the soil CO2 emission factor. Methane emissions for drained peatlands were calculated for the four cases using the regression models by Minkkinen et al. [2007b], in which tree stand volume and site type were used as input data. The stand volumes for the different site types were derived from the Finnish national forest inventory, and area-weighted estimates for the four cases were then calculated. N2O fluxes were estimated based on chamber measurements at 68 forestry-drained peatland sites in Finland [Ojanen et al., 2010].
 The GHG balances for the undrained sites were obtained from the literature (Table 2). At both undrained and drained peatlands, the GHG balances were assumed to remain constant during the whole simulation period.
Table 2. Greenhouse Gas Fluxes From/Into the Soil in Each Site Case (g m−2 yr−1)a
|South, nutrient-poor drained||244||0.141||0.029|
|South, nutrient-rich drained||465||−0.259||0.167|
|North, nutrient-poor drained||244||0.487||0.029|
|North, nutrient-rich drained||512||−0.096||0.167|
2.2.7. Calculation of Atmospheric Concentrations and RF
 The RF due to GHG flux changes, RFΔghg, was calculated with a modified version of the REFUGE model [Monni et al., 2003]. In this model, RFΔghg is estimated by time-integrating the response function related to an instantaneous concentration pulse, taking into account the annual variation in the surface exchange fluxes and background concentrations of the long-lived GHGs considered. The concentration change Δχ due to net GHG flux ϕ can be expressed as
where k denotes the emission-concentration conversion factor resulting from the instantaneous and complete atmospheric mixing assumed in the model, and fa is an atmospheric lifetime function that indicates the airborne fraction of the pulse. The removal of CO2 from the atmosphere is modeled by a superposition of three relaxation timescales, while that of CH4 and N2O is approximated by a single exponential. For the present work, the lifetime functions were updated according to Forster et al. . The radiative forcing function for CO2 (RFc), CH4 (RFm) and N2O (RFn) are based on the “simplified expressions” of the Intergovernmental Panel on Climate Change (IPCC) [Ramaswamy et al., 2001; Forster et al., 2007] and include the indirect RF effects of CH4 and the spectral interactions between CH4 and N2O:
Where χc, χm and χn are the mixing ratios of CO2, CH4 and N2O, respectively; the subscript 0 denotes unperturbed concentrations; a = 5.35 W m−2, b = 0.0378 W m−2, c = 1.24 × 10−4 W m−2 and d = 0.12 W m−2 are constants; fo is a correction function for absorption overlap,
where e = 0.47 W m−2.
 In the present calculations, the RFΔghg of CO2 (RFΔghg,c) is determined as a marginal change with respect to a varying reference concentration χc,ref,
and correspondingly for CH4 (RFΔghg,m) and N2O (RFΔghg,n). The reference concentrations are assumed to follow the IPCC SRES A2 scenario until 2050 and thereafter remain constant [IPCC, 2001]. The total RF due to the three GHGs is then
2.2.8. Uncertainty Analysis
 As discussed by Minkkinen et al. , the REFUGE model provides a suitable framework for estimating the radiative forcing for the coming 100 years. Even though the uncertainty of the modeled RF can be as high as 40% it is significantly reduced when the model is applied for a comparison of different land use scenarios, for example, like in this paper [Sinisalo, 1998]. Here we estimated the sensitivity of the RF calculations for some of the parameters, subjectively selected as representing the most critical ones, by changing slightly the initial “best estimate” values one at a time. Of the gas fluxes, the value of CH4 flux in a nutrient-rich, undrained mire was tested by increasing and decreasing the current emission rate of 20 g CH4 m−2 yr−1 by 50%. In addition, the CO2 balance of the forestry-drained nutrient-poor peat soil was changed from a source of 244 g m−2 yr−1 to a sink of −370 g m−2 yr−1. While the former value is based on a large-scale national data set on soil CO2 fluxes and simulations of litter production and decomposition, the latter value is derived from the only direct micrometeorological net ecosystem exchange measurement above such an ecosystem [Laurila et al., 2007]. In addition, the sensitivity of RF to the selected albedo limits was tested. We recalculated the RF for the cases in which (1) the upper limit of winter albedo in the forestry-drained case was lowered from 0.4 to 0.2 (motivated by the study of Viterbo and Betts ), (2) the upper limit of winter albedo in the case of an undrained open mire was lowered from 0.8 to 0.4, and (3) the lower limit of summer albedo of an undrained mire was increased from 0.15 to 0.24, the latter value being the July average of the albedo measured at the Jokioinen grass field (Table 1).
2.3. Calculation of Trends in the Surface Temperatures
 To study whether the land use changes from open mire to peatland forest have increased the spring temperatures in Finland, the surface temperature trends for March, April and May were calculated from the monthly mean, minimum and maximum temperature records based on gridded climate data. The number of weather stations with homogenized temperature time series varied from about 50 in 1909 to 180 in the 1970s. By 2009, the amount of stations had been dropped to ca. 140. Trends for the southern and northern parts of Finland were studied separately; the country was divided by the latitude 65°N. The monthly mean temperature records were extracted from the monthly mean temperature grid values covering the 100 year period from 1909 to 2008 [Tietäväinen et al., 2010]. The monthly minimum and maximum temperature records were extracted from the daily grid values covering a 48 year period from 1961 to 2008 [Venäläinen et al., 2005]. The resolution of both of the gridded data sets was 10 km. The monthly average of the daily temperature range (DTR) was calculated as the difference between the monthly maximum and minimum temperatures.
 For mean temperatures, linear trends were calculated separately for the first 50 years (1909–1958) and the last 50 years (1959–2008), while for the maximum and minimum temperatures 48 years (1961–2008) were used. The significance of the trends was tested with the t test. To study the differences between the trends of the maximum and minimum temperatures, the daily temperature range trends were also calculated.