Greenhouse gas emissions from soils—A review

Soils act as sources and sinks for greenhouse gases (GHG) such as carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O). Since both storage and emission capacities may be large, precise quantifications are needed to obtain reliable global budgets that are necessary for land-use management (agriculture, forestry), global change and for climate research. This paper discusses exclusively the soil emissionrelated processes and their influencing parameters. It reviews soil emission studies involving the most important land-cover types and climate zones and introduces important measuring systems for soil


Main greenhouse gas source and sink aspects of soils
Political agendas of individual countries and international initiatives proclaim greenhouse gas (GHG) neutrality, e.g., by the year 2050 (G7, 2015; Law and Harmon, 2011;UBA, 2013;Willson and Brown, 2008). Whether such declarations can be and must be seen as realistic cannot be an issue in this review. Yet, it is of vital interest to properly assess soils as GHG-sources and to develop a better understanding for the source and sink strengths of this important and critical environmental compartment. Carbon dioxide (CO 2 ), methane (CH 4 ), and nitrous oxide (N 2 O) are important climaterelevant trace gases. While being highly unequally distributed ( of 3-4 orders of magnitude for C), about 1500 Pg of total C and 136 (92-140) Pg of total N are stored in the uppermost meter of the global soil layer, representing the largest terrestrial carbon and nitrogen pools (Batjes, 1996;Kutsch et al., 2009;Nieder and Benbi, 2008;Schaufler et al., 2010;Schlesinger and Andrews, 2000). However, soil structure changes can influence its source and sink function (Jungkunst and Fiedler, 2007) and the total storage capacity is limited. As an example, a review of the European terrestrial GHG balance (Schulze et al., 2009) showed that the fluxes (flow rate per unit area) of CO 2 , CH 4 andN 2 O between 2000 and were near neutral; agricultural CH 4 and N 2 O emissions being largely offset by the CO 2 sink of grasslands and forests. Yet, the authors show that the trend towards more intensive agriculture and logging "is likely to make Europe's land surface a significant source of greenhouse gases". Bahn et al. (2010) compiled and evaluated global data across vegetation types and biomes, concluding that soils emit about 98 Pg C a −1 (resulting in a roughly 15-year turnover time for the reservoir). This is considerably more than emissions from fossil fuel consumption. Older sources suggest lower fluxes with 68-77 Pg C a −1 Raich and Schlesinger, 1992). Bahn et al. (2010) nevertheless emphasized the still high uncertainty with various assumptions and in upscaling attempts.
GHG emissions from soils need to be better quantified for global budgets, since 35% CO 2 , 47% CH 4 , 53% N 2 O, and 21 % nitric oxide (NO) of the respective total annual emissions relate to soil degassing (IPCC, 2007). The annual global NO emissions from soils are within the range of NO emissions from fossil fuel combustion . The increase of GHG emissions from soils mainly stems from CH 4 and N 2 O since the onset of industrialization in the mid 18 th Century and has been caused by agricultural practice (Forster et al., 2007). This makes GHG emissions from soils a key topic in global change issues, in climate research, and for agricultural and forestry management.
First measurements of CO 2 emissions from soils were performed in laboratories in the 19 th Century (Boussingault and Levy, 1853), motivated by the tracer function of soil respiration for soil fertility (Russell and Appleyard, 1915). Today climate change-related topics dominate the study of CO 2 , N 2 O and CH 4 behaviour (Cox et al., 2000). Nitric oxide (NO) has been studied because of its role in acidic precipitation, ozone formation and destruction (Kampfl et al., 2007). Field measurements with chambers were introduced at the beginning of the 20 th Century (Lundegardh, 1927). Determinations of other trace gas emissions started later, as gas analyzers became available (N 2 O in the 1950s: Arnold, 1954; NO in the 1970s: Galbally et al., 1987;CH 4 in the 1980s: Holzapfel-Pschorn et al., 1985. 1.1. Main gas species: properties and processes CO 2 fluxes can be separated into three types: 1) Soil respiration includes root, anaerobic and aerobic microbial respiration. Root respiration contributes on average up to ∼50% of the total soil respiration, yet may vary between 10 and 95% subject to season and vegetation type (Hanson et al., 2000), 2) Ecosystem respiration additionally includes aboveground plant respiration, 3) Net ecosystem exchange (NEE) is the difference between photosynthesis and ecosystem respiration. A positive NEE indicates a CO 2 source, whereas a negative NEE reveals a CO 2 sink. CH 4 in soils is produced by methanogenesis under anaerobic conditions and is consumed by methanotrophic microorganisms that use O 2 and CH 4 for their metabolism under aerobic conditions (Dutaur and Verchot, 2007). N 2 O and NO releases are driven by nitrification (oxidation of NH 4 + to NO 3 − via NO 2 − ) and denitrification (reduction of  to N 2 O and N 2 ). Nitrous oxide (N 2 O) is produced mainly by denitrification under anaerobic conditions, where the Water-Filled Pore Space (WFPS) is >50% (Ussiri and Lal, 2013). The released NO amount is negligible, compared to the NO production of nitrification . N 2 O is produced by nitrification under aerobic conditions (Ussiri and Lal, 2013). Still, quite a few open questions remain, particularly with respect to processes and their controls  that cannot be comprehensively addressed in this work.

Review context and objectives
Analyzing the influence of the different parameters on soil emissions remains challenging due to the multitude of interacting processes. Laboratory studies help in solving this problem )-see Section 3.3. Montzka et al. (2011) reviewed the non-CO 2 GHGs from a global perspective, showing the fluxes, the relevance of atmospheric residence time and flux rate alterations, and their potential in climate change mitigation.
This paper discusses soil emission-related processes and their influencing parameters. It reviews soil emission studies from the most important land-cover types and climate zones and introduces important methods for measuring soil emissions. Thus this work can be seen as complementary to important preceding reviews, e.g., Butterbach-Bahl et al. (2013), Dalal and Allen (2008), Montzka et al. (2011), Raich and Tufekcioglu (2000), Schaufler et al. (2010), and Snyder et al. (2009).
We focus on CO 2 , CH 4 and N 2 O because available high-quality data for these gases cover a wide spectrum of land-use types, crops and climates. Standard deviations are given in the text, when these were mentioned in the original publications. Emission rates have been recalculated throughout to obtain uniform units (mol × m −2 s −1 [CO 2 ] and h −1 [CH 4 and N 2 O]). To enable a direct comparison between the individual GHGs, the individual gas flux rates have been re-calculated to CO 2 equivalents (CO 2 e) and normalized to 100 years, considering their respective atmospheric residence times (IPCC, 2013).

Drivers of soil emissions
Microbial activity, root respiration, chemical decay processes, as well as heterotrophic respiration of soil fauna and fungi produce GHGs in soils (Chapuis-Lardy et al., 2007). Related emission flux rates largely depend on soil water content (humidity), soil temperature, nutrient availability and pH-value (Ludwig et al., 2001), plus land-cover-related parameters. Thus, meteorological and climatological parameters as well as land-use management information are paramount (Fig. 1). Drivers can be separated into proximal drivers that influence soil emissions in the direct environment (e.g. local climate, soil type) and distal drivers that effect soil emissions on larger scales (e.g. temperature, humidity); Robertson (1989).

Humidity
Soil humidity is the single most important soil parameter for soil gas emissions, since it controls microbial activity and all related processes (Fig. 1). Nitrifying bacteria require oxygen residing in soil pores. Soils with less water-filled pore space (WFPS) show higher emissions by nitrification, with a maximum at 20% WFPS (Ludwig et al., 2001). Nitric oxide (NO) emissions decrease in soils having below 10% WFPS due to inhibited nutrient supply . Nitrification yields a higher potential for NO production than for N 2 O production (Fowler et al., 2009). In contrast, CH 4 -and N 2 Oproducing bacteria require anaerobic conditions. N 2 O production is optimal around 60% WFPS and lowest when WFPS is below 30% (Gao et al., 2014). Even an increase of WFPS above 80% still leads to an exponential increase of N 2 O emissions (Keller and Reiners, 1994). CH 4 production requires strictly anaerobic conditions and correlates positively with soil humidity (Gao et al., 2014;Smith et al., 2003). Soils are CH 4 sinks under aerobic conditions (Fiedler et al., 2005). Wetlands and rice paddies are strong CH 4 sources (Dutaur and Verchot, 2007). Long periods of drought can significantly reduce soil emissions. Soils may then turn into a net sink for N 2 O (Goldberg and Gebauer, 2009). Grain-size distribution influences soil moisture. Soils with a high proportion of large pores retain less water and therefore foster the emission of gases produced under aerobic conditions (van der Weerden et al., 2010). NO emissions are highest in soils with coarse soil textures (McTaggart et al., 2002). Soils with dominant fine pores support the formation of CH 4 and N 2 O produced under anaerobic conditions (Dutaur and Verchot, 2007;Gu et al., 2013). Higher CO 2 emissions were encountered with fine textured soils, especially compared to sandy soils during warm dry periods (Dilustro et al., 2005). Stable soil aggregates (concretions, crusts) lead to lower soil emissions since C and N are less available for soil microbes (Kögel-Knabner et al., 1998).
Precipitation after extended dry periods causes the pulsing or Birch effect (Birch, 1958). Emissions increase within some minutes or hours after the onset of precipitation and return to background levels within a few days (Sponseller, 2007). This is driven by the renewed mineralization and the availability of easily decomposable material (Borken and Matzner, 2009) for the metabolism of reactivated microbes (Ludwig et al., 2001). The Birch effect decreases with higher frequencies of wet-dry cycles (Borken and Matzner, 2009).

Temperature
Soil temperature is important to explain the variations of trace gas emissions from soils ( Fig. 1). Soil moisture and soil temperature can explain 74% and 86% of the variations of NO and N 2 O emissions, respectively (Schindlbacher et al., 2004). An increase of soil temperature leads to higher emissions and to higher soil respiration rates as a positive feedback response of increased microbial metabolism. Methane and N 2 O emissions are additionally forced by increasing soil respiration rates with increasing soil temperatures, leading to decreasing O 2 concentrations in the soil . The positive temperature effect may be overlain by soil water stress, since water is needed as a transport medium for nutrients required by microbes (Fowler et al., 2009). Nitric oxide and CO 2 emissions increase exponentially with temperature (Fang and Moncrieff, 2001;Ludwig et al., 2001;Tang et al., 2003). Soil respiration from bacteria was observed down to soil temperatures of −7 • C (Brooks et al., 1997).
The temperature dependency of gas emissions from soils can be described with the temperature sensitivity factor Q 10 . It expresses the rate of change in a chemical or biological system with a temperature change of 10 • C (Berglund et al., 2010) and increases with soil depth (Tang et al., 2003). Q 10 is 2.4 with a range of 1.3-3.3 for soil respiration; based on a data review by Raich and Schlesinger (1992). The average values are confirmed by current studies of Hu et al. (2015) with a range of 1. 7-2.5 and Jiang et al. (2015) with an average of 2.2. The response of CH 4 emission to temperature increases with a Q 10 -value of ∼4 (Dalal and Allen, 2008). N 2 O emissions increase with temperature up to about 37 • C; thereafter denitrification and N 2 O emissions decrease. Reported Q 10 -values for N 2 O range from 1.7 to 9.3 (compiled by Abdalla et al., 2009a).
Under field conditions, moisture and temperature effects overlap, which may make it difficult to observe clear correlations (Fang and Moncrieff, 2001). Temperature is important for the regulation of freeze-thaw events, forcing gas emissions from soils (Holst et al., 2008), and may be responsible for up to 50% of the total annual N 2 O emissions (Groffman et al., 2009). In contrast, winter CO 2 emissions are considered less important for the annual emission budget since root respiration is low in temperate or more polar environments (Groffman et al., 2006). Therefore, soil water content has to be close to saturation to reduce the O 2 content (Groffman et al., 2009). During freeze-thaw cycles, additional nutrients are released for microbial metabolism through the disaggregation of soil particles (Christensen and Christensen, 1991). After thawing, dead organic material (e.g., dead plant roots) forces increased microbial soil respiration and N 2 O emissions (Mørkved et al., 2006). We argue that winter emissions are relevant for the temperate climate zone. Independently, maximum soil emissions result in spring.

Exposure and air pressure
Site exposure (elevation, morphological position, plant cover) influences soil temperature and moisture. Nitrous oxide emissions are higher in depressions than on slopes and ridges due to higher soil moisture (Izaurralde et al., 2004). Lower air pressure supports higher soil emissions due to reduced counter pressure on the soil (Reicosky et al., 2008).

Vegetation fires
Fires in ecosystems can affect the greenhouse gas balance of soils, depending on temperature and duration of the fire, with burned areas showing lower CO 2 and N 2 O fluxes than non-burned reference sites for around one month after burning (Kim, 2013; Table 1). This is caused by reduced root respiration in the absence of plant cover and the related pH change. Lower N 2 O emissions are caused by charcoal (Kim, 2013). After burning, soil temperatures increase due to missing canopy, while soil moisture does not change since lowered plant transpiration compensates for missing or reduced plant canopy (Castaldi and Fierro, 2005). No or minor changes were observed for CH 4 uptake in soils (Kim, 2013). Thermal ammonification causes ammonium increases after a forest fire. Nitrate decreases with time due to suppressed nitrification (MacKenzie et al., 2004).

Soil pH-values
Microbial activity is influenced by soil pH. Therefore, management practices such as liming influence soil emissions; additional carbonate can be released as CO 2 (Snyder et al., 2009). Acidic soil conditions lead to lower soil emissions. The optimal pH-value for methanogenesis (CH 4 production) lies between pH 4 and 7 (Dalal and Allen, 2008). CO 2 emissions were observed to be highest at neutral pH-values (Cuhel et al., 2010). N 2 O emissions decrease only under acidic soil conditions. Nitrification increases with higher pH-values, since the equilibrium between NH 3 and NO 3 − shifts to ammonia (Nugroho et al., 2007). However, no significant correlations between NO and N 2 O emissions and pH-value were found (Pilegaard et al., 2006). NO emissions are caused by denitrification under acidic soil conditions, whereas alkaline conditions foster NO emissions produced by nitrification (Remde and Conrad, 1991).

Nutrients
Nutrient availability is paramount to microbial and plant respiratory processes. Hence, natural N and C content in soil, as well as atmospheric deposition, manure or fertilizer applications play an important role. N 2 O emissions negatively correlate with the C/N-ratio (Pilegaard et al., 2006), with N 2 O emissions being lowest at C/N-ratios ≥30 (limited disintegration of organic material) and highest at a C/N-value of 11 (optimum disintegration and humus build-up); Gundersen et al. (2012a,b). Combined with drought and low pH-values, N 2 O emissions can be significantly suppressed at C/N-ratios <20 (Gundersen et al., 2012a,b). Emissions of CO 2 and CH 4 positively correlate with the C/N-ratio (Shi et al., 2014;Weslien et al., 2009). Yet, depending on the availability of other electron donors such as Fe 3 + , Mn 4 + , SO 4 2− and NO 3 − , the CH 4 production in soils may be reduced (Achtnich et al., 1995;Dalal and Allen, 2008;Fumoto et al., 2008;Kögel-Knabner et al., 2010;Sahrawat, 2004), which is of particular relevance in many (sub)tropical soils and particularly in rice paddies. Increasing soil N content generally leads to higher soil respiration and to higher net ecosystem exchange (NEE), if carbon is not limiting (Niu et al., 2010;Peng et al., 2011). With limited C availability, N fertilizer application has limited influence on soil respiration (Micks et al., 2004). The application of N fertilizers leads to a higher sensitivity of soil respiration against soil moisture and to a lower sensitivity to soil temperature . In short-term studies, application of NO 3 − or NH 4 + leads to a decrease or to no change in the soil respiration rate of forest soils (Micks et al., 2004). Soil respiration decreases during long-term experiments of N addition (Bowden et al., 2004). The application of liquid manure (urea) led to higher N 2 O emissions under aerobic soil conditions, whereas NH 4 + fertilizers caused higher N 2 O emissions under saturated conditions (Tenuta and Beauchamp, 2003).
To minimize N 2 O emissions from agricultural lands, fertilizer application rates need to be adapted to plant needs since not all forms of nitrogen can be taken up by plants. Non plantavailable N amounts lead to increasing N 2 O emissions (McSwiney and Robertson, 2005). Cover crops increase CO 2 and N 2 O fluxes after their incorporation and positively influence soil respiration during their lifetime (Sanz-Cobena et al., 2014b). Using controlledrelease fertilizers or denitrification inhibitors prevents increased N 2 O emissions (Shoji et al., 2001); nevertheless this effect can be disturbed by heavy precipitation events . Soil water content is important for selecting the fertilizer type to inhibit increased N 2 O emissions (Sanz-Cobena et al., 2014a). Fertilizer applications are also influenced by the tillage system. Using urea, N 2 O emissions were higher under no-till and conservation tillage, while no differences could be observed using urea-ammonium nitrate fertilizers (Venterea et al., 2005). However, there are discordant findings on the influence of the tillage system: N 2 O emissions decreased with no-till practice (Omonode et al., 2011) and were explained by lower soil temperatures (Six et al., 2002), while others found a positive effect of no-till on N 2 O emissions and explained this with higher microbial activity (Baggs et al., 2003). Higher soil moisture during no-till practice leads to higher N 2 O emissions. Higher N 2 O emissions under no-till cannot be balanced by higher C sequestration and CH 4 -uptake rates (Li et al., 2005;Six et al., 2002).
Yet, no information about the starting point of the no-till practice for this study was included. This is essential, since conventional and no-till approaches can only be compared once an equilibrium has been established after extended periods of time. Six et al. (2004) reviewed long-term studies on the influence of no-till practices on N 2 O emissons. They found that N 2 O emissions were only elevated during the first ten years after shifting from conventional tillage to conserving no-till farming practice. Soil compaction caused by the use of farming machines leads to lower soil emissions (Mordhorst et al., 2014). A reduction of N deposition led to decreasing soil NO emissions at forest sites, whereas N 2 O emissions were not affected by the reduction (Eickenscheidt and Brumme, 2012).

Vegetation
Vegetation age and tree type (deciduous, coniferous, and species) influence soil respiration. The highest soil respiration rates were found for young spruce forest stands, when comparing 10-, 15-, 31-and 47-year-old stands (Saiz et al., 2006). Soil respiration decreased with stand age, caused by a lower fine root biomass. The decrease levelled out with stand age since lower root respiration rates in old forest were partly compensated for by higher microbial respiration due to higher organic inputs (Saiz et al., 2006). Similar results were found for 45-and 250-year-old pine stands (Law et al., 1999), 20-and 40-year-old fir stands (Klopatek, 2002), and for 5-and 15-year-old poplar plantations (Gong et al., 2012). A high biodiversity with a balanced ratio of Leguminosae and C3 and C4 plants (carbon fixation) at grassland sites resulted in an increased C-sequestration potential (Fornara and Tilman, 2008). On agricultural sites, N 2 O emissions from legume-N were significantly lower than fertilizer-N derived N 2 O emissions (Schwenke et al., 2015).
Vegetation affects CH 4 emissions and positively correlates with net ecosystem production (Dalal and Allen, 2008). Elevated CO 2 concentrations in soils can also be caused by higher root mass due to elevated atmospheric CO 2 concentrations (Dorodnikov et al., 2009). Furthermore, soil moisture increases, since the opening time of stomata can be reduced. Hence, conditions for N 2 O and CH 4 emissions improve (Kim, 2013) and drive denitrification (Ding et al., 2003;Smart et al., 1997). Soil temperatures decrease due to the higher leaf area and related shade (Kim, 2013).

Land-use change
Land-use change is very important for GHG emissions from soils, particularly when forests, grass and peat lands are being converted to agricultural land (Fig. 1). Within the first 30 years after turning forest into agricultural land, 30-35% of the soil carbon stored in the top 7 cm is lost, while below plough depth no changes are recognizable (Degryze et al., 2004). The global C-sequestration potential of agricultural soils amounts to 0.73-0.87 Pg C yr −1 (Lal and Follett, 2001). These aspects are further elucidated in Section 2.

Effects of land-use and land-cover types on soil GHG emissions
The global land surface of 149,430,000 km 2 is covered by 31.5% grassland and pastureland (including shrubs, herbaceous and sparse vegetation; ca. 47,070,500 km 2 ), followed by 27.7% treecovered areas (including forestland; ca. 40,794,000 km 2 ), 15.2% barren land (including [semi]arid areas; ca. 22,713,000 km 2 ), 12.6% cropland (with annual and perennial crops; ca. 18,828,000 km 2 ), 9.7% snow and glaciers, 2.7% water bodies, wetlands and mangroves (ca. 4035 km 2 ), and 0.6% artificial surfaces (including urban environment), according to FAO (2015) and Latham et al. (2014). The most prominent land-cover types are examined in this paper-and include the latest discussion on wetlands and their supposedly particularly important level of GHG release. Neither snow-and glacier-covered areas, nor sealed artificial surfaces deliver respiration rates of any global relevance. Thus, for some calculations these areas are excluded and an effective land surface of 135.39 Mio km 2 is being used (Bahn et al., 2010). Fig. 2 presents a world map of the locations of the studies reviewed in this work and indicates the land-cover types. That map suggests a certain bias with respect to representative geographical coverage of the compiled data. The vast majority of studies have delivered data from Europe, followed by P.R. China and the United States. The southern hemisphere is comparatively underrepresented.
Soil organic carbon (SOC) accumulation largely depends on vegetation cover. Any change in land use may significantly alter related source or sink characteristics for atmospheric CO 2 and other GHGs (Poeplau and Don, 2013;Sainju et al., 2008). Plant species differ in root depth and spatial distribution (Breecker et al., 2012). Therefore, soil-flux data from studies involving various land use, land cover, and different climate regions need to be evaluated separately in order to properly characterize soil degassing (Bahn et al., 2010). In addition, agricultural practices such as tillage and fertilization have to be considered. Food systems alone -everything from growing plants to the disposal of biomass -contribute to 19-29% of global anthropogenic GHG emissions. Of this, 80-86% relate to agricultural production (including indirect emissions associated with land-cover change), albeit with significant regional variation (Vermeulen et al., 2012). Short rotation forestry (SRF) has been integrated into the cropland section, since sites are harvested at 2-3-year intervals and replanting is done about every 20 years.
Wetlands (albeit with limited land cover: 2.7%) show the highest average absolute emission rates; considerably higher than all other land cover types. Forestland, grasslands, croplands and barren lands follow in this order with decreasing emission rates. Yet, individual data of all land-cover types show a high variance (Fig. 3). This reflects climate zone and land management-related conditions and clearly demonstrates the key drivers for soil degassing. Using a conservative average of 300 mg CO 2 e m −2 h −1 for all land-cover types combined as compiled from our literature review, global annual soil emissions of ≥350 Pg CO 2 e result, corresponding to roughly 21% of the estimated global soil C and N pools (Fig. 3).

Grassland, including pastureland and savannahs
Permanent grasslands cover about 31.5% of the total global land area and about 70% of the total agricultural expanse (FAO, , 2015. This high grassland cover percentage and the above average GHG emissions underline the potential of grassland and pastureland in mitigating global warming. Table 1 clearly corroborates the inadequate data situation and the strong bias towards temperate climate regions.

CO 2 -soil respiration
CO 2 -emission rates from perennial grasslands and from Tundra were estimated at 1.54-1.61 and 0.48-0.57 mol CO 2 m −2 s −1 , respectively . Under a semi-arid, temperate climate (grassland in Inner Mongolia), the highest CO 2 -emission rates occurred during summer (3.09 ± 0.23 mol CO 2 m −2 s −1 ) and the lowest rates during the snow-covered winter period (1.26 ± 0.15 mol CO 2 m −2 s −1 ) with Q 10 values ranging between 1.63 and 1.74, depending on the fertilization system . WFPS plays an important role in CO 2 emission rates; 20-40% WFPS in grassland soils results in maximum emissions . Higher soil respiration rates were measured at managed grassland sites in Finland (peak values 11.6-25 mol CO 2 m −2 s −1 ; seasonal average 1.82-7.95) with the higher values recorded in peat soils and the lower ones in sandy and clayey soils . Fertilizing grasslands enhances soil respiration. A 24-57% increase was measured in the annual total soil respiration in the first year after fertilizing under semi-arid temperate climate conditions (Inner Mongolia; Peng et al., 2011).

N 2 O emissions
The influence of livestock grazing on pastureland showed that the type of animals grazing on pasture does influence soil N 2 O emissions. Sheep-grazed pasture sites emitted less N 2 O than cattle-grazed ones, while the lowest emissions were reported from non-grazed pastures (1. 63-2.44, 4.07-4.89 and 0.41-0.82 Saggar et al., 2008). However, Wolf et al. (2010) found during a year-round study in Inner Mongolia on steppe grassland that the pulse of spring-thaw emission was lowest at sites with intense grazing (0.03 ± 0.03 mol N 2 O m −2 h −1 ) and highest on non-grazed sites (0.45 ± 0.03 mol N 2 O m −2 h −1 ). This applies also for the annual N 2 O-emission budget, since spring-thaw emissions contribute most to the annual budget. A sheep-grazing site (permanent pasture) showed significantly lower N 2 O emission rates (0.68 mol N 2 O m −2 h −1 ) than conventional tillage (CT) and no-tillage (NT) sites under winter oat (3.75 mol N 2 O m −2 h −1 ) and fodder maize cultivations (4.89 mol N 2 O m −2 h −1 ) and did not significantly differ between CT and NT sites . For a good compilation of N 2 O and CH 4 emissions from savannahs and seasonally dry ecosystems, see Castaldi et al. (2006) who generally found low emissions for N 2 O and an uptake for CH 4 .

CO 2 , N 2 O, and CH 4 dynamics
Establishing grasslands on drained organic soils enhanced GHG emissions (Kasimir-Klemedtsson et al., 1997). Emission rates from this land-use type in Finland were 7.98 mol CO 2 m −2 s −1 and 1.52 mol N 2 O m −2 h −1 . Using reclaimed coal mine areas to establish managed grassland (fertilized and harvested hay fields) in south-eastern Ohio (USA) emitted 3.18-5.30 mol CO 2 m −2 s −1 , 1.35-1.60 mol CH 4 m −2 h −1 and 0.63 mol N 2 O m −2 h −1 . Our group determined similar values for soil respiration on a winter rye field (measurements took place after seeding between plant rows with plant height of 5 cm) at a former lignite site in Lusatia, Germany, with 4.41 ± 1.15 mol CO 2 m −2 s −1 . The non-fertilized meadow at the reclaimed mine in southeastern Ohio (USA) showed emission and uptake rates from 2.6 to 3.18 mol CO 2 m −2 s −1 , −3.54 to −1.63 mol CH 4 m −2 h −1 and 0.85 mol N 2 O m −2 h −1 .
From the compiled data, it can be calculated that grassland soils emit on average 350 mg CO 2 e m −2 h −1 -the third highest flux rate after wetlands (which represent a small area only) and forestlands (Fig. 3). Given the wide range of results that partly cover several orders of magnitude and the distinct lack of data for Mediterranean and (sub)tropical environments, particularly in the southern hemisphere, it appears premature to compile reliable global assessment data from the existing studies − considerable effort is needed to bridge these gaps.

Tree-covered land, including forestland
Tree-covered land including forestland covers about 28% of the terrestrial surface. Yet, tree cover and related leaf area indices vary widely, from extremely sparse vegetation (e.g., Caatinga in north-eastern Brazil and similar savannah-type vegetation-which we discuss due to their dominant characteristics in the preceeding section on grassland) to multi-storey tropical rainforests with extremely dense foliage. A similarly high variability occurs with the respective soil carbon and nitrogen pools. Dixon et al. (1994) compiled global forest ecosystem pool data. Their ranges extended from 471 Pg at high latitudes via 100 Pg at mid latitudes to 216 Pg at low latitudes, with a total of 787 Pg. Available data are by far not yet sufficient to obtain a highly representative and precise picture of related GHG emissions, yet decent estimates are available ( Table 2). Particular focus is given to longer-term measurements, since many studies deliver relatively short (and potentially less representative) data series.
2.2.1. Global flux estimates from forests 2.2.1.1. CO 2 -soil respiration. Bond-Lamberty and Thompson (2010) summarized soil-respiration (R S ) measurements from 818 studies with 3379 records for the years 1961-2007. Mean (±1 s.d.) R S was 1.01 ± 0.60 for boreal (ca. 15 Mio. km 2 ), 1.97 ± 1.11 for temperate (ca. 7 Mio. km 2 ), and 3.40 ± 1.67 mol CO 2 m −2 s −1 for tropical ecosystems (ca. 19 Mio. km 2 )-resulting in a global average of 2.28 ± 50% mol CO 2 m −2 s −1 for the global forestland. A more detailed review of the CO 2 balance of forests is available from Luyssaert et al. (2007), based on publicly available data and including measurement data and model results ( Table 2). Schulze et al. (2010) reviewed flux measurements of trace gases of CO 2 and other gases within the CarboEurope network. Forest mean autotrophic respiration was around 1.56 ± 0.23 mol CO 2 m −2 s −1 and heterotrophic respiration was determined at 0.97 ± 0.28 CO 2 mol CO 2 m −2 s −1 . Oertel et al. (2015a) present 3.5 mol CO 2 m −1 s −1 (autotrophic + heterotrophic) for forests in Saxony, Central Europe, in summer. Monson et al. (2006) compiled winter forest soil respiration fluxes between 0.05 and 0.25 mol CO 2 m −2 s −1 . (2008) reviewed N 2 O emission data from forest soils. They found that the mean N 2 O-emission rate from tropical forest soils (1.23 ± 2.09 mol N 2 O m −2 h −1 ) was at least 2-3 times higher than those from temperate forest soils (0.41 ± 0.15 mol N 2 O m −2 h −1 ). These results were based on 22 studies in tropical forests and 15 studies in temperate forests. The dataset was very limited for boreal forest soils; only four studies were included. The average annual flux of N 2 O emissions from boreal forest soils was small with 0.10 ± 0.08 mol N 2 O m −2 h −1 .

N 2 O-soil emissions. Dalal and Allen
2.2.1.3. CH 4 -soil uptake. Forests are generally sinks for CH 4 . This is the key result of CH 4 flux measurements for different ecosystems, as reviewed by Dalal and Allen (2008). Mean tropical forest CH 4 consumption was given with −2.75 ± 0.42 mol CH 4 m −2 h −1 (based on 13 studies). Temperate forests consumed on average −3.42 ± 0.44 mol CH 4 m −2 h −1 (18 studies), with deciduous forest soils showing higher CH 4 consumption rates than coniferous forest soils and undisturbed forest soils showing higher CH 4 consumption rates than disturbed forests or regrowth and fertilized forests (Smith et al., 2000). Studies in boreal forests are still scarce (4 studies). Their annual average CH 4 flux is given by −1.38 ± 2.80 mol CH 4 m −2 h −1 , with a range from −9.38 mol CH 4 m −2 h −1 to 3.53 mol CH 4 m −2 h −1 . The latter value represents one boreal aspen forest in Canada that produces CH 4 (Simpson et al., 1997).

Results from long-term soil flux measurements
The interannual variability of GHG emissions is controlled by climate variability and ecosystem-internal parameters (Phillips et al., 2010;Urbanski et al., 2007;Fig. 1). Therefore, soil flux datasets comprising only sporadic measurements or not covering complete seasons and/or years may not be representative for a site (Luo Table 2 Mean soil respiration fluxes in mol CO2 m −2 s −1 with standard deviations (recalculated after Luyssaert et al., 2007 (Phillips et al., 2010), Höglwald spruce site, Germany  and Solling, Germany (Borken et al., 2002). Here we focus on the results of these longterm measurements to obtain information about the relationships between environmental parameters and soil trace gas fluxes.  (Phillips et al., 2010). It may thus be concluded that climatic conditions during the dormant or early growing season might determine annual soil respiration rates in temperate forest ecosystems. Increased soil respiration rates may also result from increased plant carbon assimilation and associated litter production in the previous year . Their data also showed a close correlation of gross primary production and soil respiration. Years with extended summer droughts revealed reduced annual soil respiration rates at all three sites Luo et al., 2012;Savage and Davidson, 2001  . These high fluxes are related to the elevated nitrogen deposition at the Höglwald Forest, lying in a range of 25-30 kg N ha −1 yr −1 for more than three decades . Similar results were found for NO fluxes at the Speulderbos Douglas fir forest site in the Netherlands, also receiving high loads of atmospheric N input (Pilegaard et al., 2006), whereas sites with low atmospheric nitrogen input (<10 kg N ha −1 yr −1 ) showed much lower mean N 2 O fluxes (0.08 mol N 2 O m −2 h −1 ; Bowden et al., 1991;Castro et al., 1993). Results from various forest sites across Europe were summarized within the EU-funded NOFRETETE project (Pilegaard et al., 2006). Their results show that the correlation between atmospheric N-input and nitrogen emissions is highly correlated for NO and not significantly correlated for N 2 O. However, a strongly negative correlation between N 2 O emissions and the C/N ratio was found.
Results from the Höglwald site show that occasional freeze-thaw pulse emissions dominate the annual N 2 O fluxes in certain years   Smith et al., 2000). However, rates of atmospheric CH 4 oxidation were highly variable in a range of −0.08 to −13.75 mol CH 4 m −2 h −1 (e.g., Ambus and Christensen, 1995;Borken and Brumme, 1997;Butterbach-Bahl et al., 1998).
Forests receiving high loads of atmospheric nitrogen and acid deposition show low CH 4 oxidation rates Butterbach-Bahl et al., 2011;Chan et al., 2005;Liu and Graever, 2009;Luo et al., 2012). The long-term CH 4 -flux measurements at the Höglwald site showed only weak negative correlations with soil moisture and weak positive correlations with soil temperature . Drawn from the compiled data (again with similar imperfections as with all other land-cover types), it can be calculated that forest soils emit on average 438.9 mg CO 2 e m −2 h −1 . As forest soils exhibit the second highest flux rate after wetlands with their small area coverage, this represents a very important global flux (Fig. 3).

Bare soil/barren lands
Here, all types of barren land from all climate zones, including tundra, are included in the discussion. The global contribution of CO 2 emissions from tundra and arid ecosystems amounts to about 10% . The sensitivity of these ecosystems to soil respiration is lower than that of other ecosystems, apparent from its low Q 10 value of 1.39. Soil respiration rates in arid areas under continental climate ranged between −0.24 and 1.82 mol CO 2 m −2 s −1 (Zhang et al., 2009), raising explicit open questions about what determined the negative rates and suggesting additional research. Emission rates were lower at a desert shrubland in Inner Mongolia under semi arid continental climate with 0.26 ± 0.04 and 0.58 ± 0.07 mol CO 2 m −2 s −1 (Jin et al., 2009). Scrubland ecosystem sites under arid climate in north-western China showed soil respiration rates between 0.57 and 0.86 mol CO 2 m −2 s −1 at places with natural vegetation and 1.08 mol CO 2 m −2 s −1 at sites with non-native vegetation (Lai et al., 2012).
A desert-shrub ecosystem located at the transition between arid and semi-arid climates showed emission rates between 0.23 and 2.56 mol CO 2 m −2 s −1 (Wang et al., 2014). Higher soil respiration rates and elevated Q 10 values were reported from shrubland and bare land in the Loess Plateau of China under semi-arid conditions with emission rates of 2.03 and 2.66 mol CO 2 m −2 s −1 , and Q 10 values of 1.43 and 2.25, respectively (Shi et al., 2014). Annual GHG emissions were determined from different biological soil crusts in the Tengger desert ecosystem, China. Emission rates ranged between 0.24 and 0.66 mol CO 2 m −2 s −1 , and yielded −1.24 mol CH 4 m −2 h −1 and −0.08 mol N 2 O m −2 h −1 (Hu et al., 2014). Under semi-arid temperate desert ecosystems in the eastern Inner Mongolia Plateau in northern China, CH 4 in soils over two years showed a mean uptake rate of 3 mol CH 4 m −2 h −1 during the growing season (Hou et al., 2012) (Table 3). Correia et al. (2014) measured a soil respiration of 2 mol CO 2 m −2 s −1 from shrubland sites under Mediterranean climate (Table 3). Similar to the situation explored in Section 2.1 on grassland soil emissions, data for Mediterranean and (sub)tropical environments are underexplored.
Soil respiration rates at sites of different land cover (grass, woody shrub and tree landscapes) under dry conditions in Table 3 Soil emission rates in mol CO2 m −2 s −1 and mol X m −2 h −1 for CH4 and N2O from bare soil, (semi)deserts and related land under temperate, mediterranean and (sub)tropical climates. Rates are given without standard deviations. Ranges have been rounded for clarity (see original literature and text for more details).  Maljanen et al. (2004) Finland. 10 Shi et al. (2014) China. 11 Thomas et al. (2011) Botswana. 12 Wang et al. (2014) China. 13 Zhang et al. (2009) China.
Botswana ranged between 0.035-0.14 mol CO 2 m −2 s −1 . The values increased to 0.2-0.96 mol CO 2 m −2 s −1 when the soils received 50 mm of precipitation (Thomas et al., 2011). Higher soil respiration rates were measured in a tree ecosystem under semi-arid conditions in India, with 1.60 ± 0.36 in winter and 9.89 ± 0.78 mol CO 2 m −2 s −1 in summer (Gnaana Saraswathi et al., 2008). Not surprisingly, bare soil demonstrates the lowest average fluxes from all land-cover types (Fig. 3). Calculations from the compiled data on bare soil in (semi)deserts and related lands results with an average emission of 126.7 mg CO 2 e m −2 h −1 . Such a flux would represent ca. 6.5 % of the global emissions as calculated from the data in this work.

Cropland
Agriculture influences global warming due to related direct and indirect GHG emissions from C and N dynamics Wang et al., 2013b). Arable lands and lands under permanent crops cover about 17 Mio. km 2 or 12.6% of the global land surface . Other assessments deliver 15 Mio. km 2 or 12% with an error margin between 12.2 and 17.1 Mio km 2 (Ramankutty et al., 2008). Deforestation and other land-use changes to increase the surface area for crop production further contribute to global warming. In addition, croplands generally stand for intensive agricultural management (e.g., intensive tillage, application of fertilizers and other chemicals), enhancing GHG emissions. Cereals and oil crops (e.g., seed cotton and rapeseed/canola) are the World's most prominent crops (FAO, 2015) ( Table 4).
Plant residues are being used as soil cover to decrease erosion, maintain soil humidity and to improve soil quality (mulching). This may influence soil emission rates. Cotton residuals with urea fertilization decreased N 2 O emission by 25% from fertilized fields without (2.05 ± 0.14 mol N 2 O m −2 h −1 ) and with cotton residue (1.5 ± 0.11 mol N 2 O m −2 h −1 ), according to Muhammad et al. (2011). The effect on soil emissions depends on the chemical properties of the residue cover. Yet, all types of plant residues (sugarcane trash, maize and sorghum straw, cotton residues and lucerne) increased the cumulative CO 2 emission by about a factor of 3 . Bean fertilization (Phaseolus vulgaris L.) under sub-humid temperate climate conditions did not significantly change soil respiration. However, soil respiration rates increased by 137.5% in soils supplemented with vermicompost and by 237.5% in soils with vermicompost and wastewater sludge (WWS). Different composts or fertilizer additions resulted in higher N 2 O-soil emissions that spanned one to two orders of magnitude: urea (factor 19), low dose of WWS (factor 19), vermicompost (factor 47) and high dose of WWS (factor 97) according to Fernández-Luqueño et al. (2009).

Temperate climate
Average soil emission data from the growing season vary by one order of magnitude in temperate regions with a spread of 0.45-8.2 mol m −2 s −1 CO 2 , 0.16-0.31 mol m −2 h −1 CH 4 and 0.03-10.7 mol m −2 h −1 N 2 O; the latter high value refers to a fertilization experiment (Table 5).
Methane uptake in soils decreases with increasing precipitation and increases with addition of N fertilizer; soils with wheat and oilseed rape cultivation under different rotations take up atmospheric CH 4 in a range of 0.25 ± 1.29 to 0.31 ± 0.70 mol CH 4 m −2 h −1 . Re-forestation of organic soil croplands did not change CH 4 and N 2 O emissions, but showed reduced CO 2 emissions (Maljanen et al., 2007).
The rate of soil N 2 O emissions was five times higher from annual bioenergy crop fields (wheat and oilseed rape with 0.30 ± 0.79 to 1.15 ± 2.51 mol N 2 O m −2 h −1 ) than perennial crops such as the willow short rotation crop (SRC) and grass, Miscanthus, with 0.01 ± 0.21 to 0.13 ± 0.97 mol N 2 O m −2 h −1 (Drewer et al.,  were reported after 1-2 months of fertilizing bioenergy sites .

Mediterranean climate
Dry Mediterranean conditions prevail over about 2% of the total global land area. The mean soil respiration rate at rain-fed olive grove sites under such conditions was 1.12 ± 0.12 mol CO 2 m −2 s −1 (Almagro et al., 2009). Yet, cultivating bioenergy crops may negatively affect CO 2 savings if other GHG emission rates are high, e.g., fertilizing maize crops may cause a >70% increase in soil N 2 O emissions (Sanz-Cobena et al., 2014b). Abandoned agricultural areas still emit GHGs. A mean respiration rate of 1.71 ± 0.09 mol CO 2 m −2 s −1 was determined over a former cereal field that lay fallow for about 25 years (Almagro et al., 2009). The authors found a positive correlation between soil respiration and soil temperature only when soil water content was above 15%.

(Sub)tropical climate
Sugar cane is an important crop in Brazil, contributing to more than 37% of the global harvested area (FAO, 2015. The related soil respiration rate was 7.01 shortly after tillage, and measured 13.76 mol CO 2 m −2 s −1 in no-tillage (NT) and conventional tillage (CT) sugar cane fields. One month after tillage, emission rates changed to 4.86 mol CO 2 m −2 s −1 from NT and 17.87 mol CO 2 m −2 s −1 from CT (La . Measured N 2 O fluxes from fertilized tea fields in North Eastern India (humid subtropical conditions) showed emissions ranging between 0.27 and 2.26 mol N 2 O m −2 h −1 .
From the compiled data, it can be calculated that croplands, composing 12.6% of the global land area, emit on average 297.7 mg CO 2 e m −2 h −1 , which is the second lowest flux rate of all land-cover sites (Fig. 3).

Wetlands, including rice paddy soils
Wetlands deliver important ecosystem services. They cover 2.6% of the Earth's surface and are equivalent to 25% of the total anthropogenic and natural sources of methane (FAO, 2015;Latham et al., 2014). Different parameters such as water depth, temperature and type of plant (floating, submersed, emersed, etc.) affect CH 4 emissions Whalen, 2005). Of the global CH 4 emissions from wetlands, 75% are estimated to stem from tropical latitudes (EPA, 2010;Whalen, 2005). The average emission rates from natural wetlands were determined as highest in tropical areas (508 mol CH 4 m −2 h −1 ), lower in temperate areas (391 mol CH 4 m −2 h −1 ) and lowest in Nordic areas (104 mol CH 4 m −2 h −1 ; Cao et al., 1998). Average CH 4 emissions of 326 mol CH 4 m −2 h −1 emerged from temperate wetlands and 216 mol CH 4 m −2 h −1 from subarctic/boreal wetlands in both Scandinavia and North America . Methane emissions also decreased from temperate via boreal to subarctic zones in bogs and rich fens, whereas nutrient-poor fens showed highest emissions in the subarctic zone . Grunwald et al. (2012) found the highest rate of CH 4 emissions from rice paddies in the mediterranean zone (278 mol CH 4 m −2 h −1 ) and the lowest from temperate wetlands (85 mol CH 4 m −2 h −1 ) and from wetlands in cold climates (80 mol CH 4 m −2 h −1 ). Aselmann and Crutzen (1989) provided a global estimate for wetland CH 4 -emission rates in bogs (39), shallow lakes (112), fens (208), swamps (219), floodplains (260), marshes (659) and rice fields (807 mol CH 4 m −2 h −1 ) ( Table 6). Methane emissions from soils in the subartic zone can vary largely and depend on presence of permafrost (124.74 mol CH 4 m −2 h −1 ) or its absence (252.08 mol CH 4 m −2 h −1 ) .
Tropical wetlands in northern Australia yielded average emission rates of 2.00 mol CO 2 m −2 s −1 and 2108 mol CH 4 m −2 h −1 . CH 4 fluxes from natural wetlands, independent of climate zone, were 260.4 mol CH 4 m −2 h −1 (Whalen, 2005). Wetlands cultivated with different plants were sinks for CH 4 (−1.84 ± 0.18 mol CH 4 m −2 h −1 ) and sources for CO 2 (2.55 ± 0.13 mol CO 2 m −2 s −1 ) and N 2 O (0.85 ± 0.33 mol N 2 O m −2 h −1 ) . The highest CH 4 emissions were registered with 95 % WFPS. About 97% of the global paddy rice is produced in wetlands areas (FAO, 1998). Peak N 2 O-fluxes from rice fields in Italy were observed following fertilization at eight locations, revealing emission averages of 42.4 ± 46.1 mol N 2 O m −2 h −1 which decreased after flooding to around zero . Average CH 4 emissions from rice paddies were 130.2-1563 mol CH 4 m −2 h −1 . Bhattacharyya et al. (2013) studied the effect of increasing atmospheric CO 2 (550 ± 30 mol mol −1 ) on CH 4 and N 2 O emissions at lowland rice fields under tropical climate in eastern India: 26% (CH 4 ) and 24.6% (N 2 O N) higher emissions were registered. Different tillage systems were compared in rice fields under humid mid-subtropical monsoon climate in China . They reported that the no-tillage system reduced the total seasonal CO 2 emissions by 19-33%. In general, emission rates from paddies with early maturing rice crops ranged between 0.63-9.24 mol CO 2 m −2 s −1 and −157.5-7813 mol CH 4 m −2 h −1 . GHG-emission rates from rice paddy fields in Eastern China varied between 1.78-1.86 mol CO 2 m −2 s −1 , 7.1-7.8 mol CH 4 m −2 h −1 and −0.84-0.74 mol N 2 O m −2 h −1 ; emission rates were affected by the fertilizing scheme .
Methane emission rates of 95 ± 63 mol CH 4 m −2 h −1 were determined in floodplain areas of the Inner Mongolian steppe . Higher values were measured from different types of wetlands located in a floodplain in central Ohio, USA, with mean emission rates between 92 and 300 mol CH 4 m −2 h −1 . Summer N 2 O-emission rates from waterlogged and from relatively dry tundra marsh sites in maritime Antarctica were −0.07 and 0.95 mol N 2 O m −2 h −1 , respectively (Zhu et al., 2014). Constructed wetlands (CW) for wastewater treatment also emit GHGs. Søvik et al. (2006)   .
A fen site in Finland (peat soil, average air temperature −1.1 • C) delivered CH 4 -emission rates of up to 68 mol CH 4 m −2 h −1 during the cool season and 347 mol CH 4 m −2 h −1 during the warm season. Much lower annual CH 4 -emission rates of 1.73 ± 2.10 mol CH 4 m −2 h −1 were registered from moorland in the United Kingdom (8.1 • C average annual air temperature) in comparison with the Finnish site (122 ± 28.5 mol CH 4 m −2 h −1 ). The wetland site in Finland was a net sink for CO 2 ; its average NEE over two years was −0.06 mol CO 2 m-2 s −1 .
Peatlands are a type of wetland ecosystem. Their organic-rich soils are considered to be one of the important sources of the GHGs CO 2 and N 2 O, mainly when used for agriculture after drainage. High emission rates occur even 20-30 years after agricultural activities have ceased. In contrast to other land-cover types, these peatlands may become a CH 4 sink. Insignificant low emission rates of CH 4 were noticed in short periods from drained peatland sites. Annually, they acted as a sink for atmospheric CH 4 . Soil properties affect soil potential to take up atmospheric CH 4 . Soils with higher organic matter content have higher CH 4 uptake (1.49 mol CH 4 m −2 h −1 ) than those with lower organic content (1.0 mol CH 4 m −2 h −1 ). In addition, soils without vegetation cover have the lowest uptake rate: 0.52 mol CH 4 m −2 h −1 . A mangrove ecosystem under dry tropical climate in northern Colombia exhibited emission rates between 0.8-26.8 mol N 2 O m −2 h −1 and 0-1973 mol CH 4 m −2 h −1 . The authors also found a negative correlation between salinity and N 2 O emission.
On a global scale, wetlands are still being drained for different purposes. About 50% of global wetland areas have been converted to other land-use forms (Verhoeven and Setter, 2010). Using dried-up peat soils for agriculture enhances CO 2 and N 2 O emissions because of fertilization and tillage (Maljanen et al., 2007). Soil under barley crops at drained organic sites in Finland emitted 7.52 mol CO 2 m −2 s −1 , while nearby bare soils released 3.66 mol CO 2 m −2 s −1 . The rate of N 2 O emission was 4.26 mol N 2 O m −2 h −1 from soil Table 6 Soil emission rates (mol CO2 m −2 s −1 , mol CH4 and N2O m −2 h −1 ) from natural wetlands, including river stretches and reservoirs under temperate, mediterranean and (sub)tropical climates. Rates are given without standard deviations. Ranges have been rounded for clarity (see original literature and text for more details).  27-1.41 Constr. wetland 30 3.16-3.79 Rewetted fen

Continental
Continental 5-50 Mio km 2 Sub-continental 0.5-5 Mio km 2 Regional Large regional 50,000-500,000 km 2 Medium regional 5000-50,000 km 2 Small regional 500-5000 km 2 Since there seems not to be a unified definition for spatial dimensions in Geosciences, Table 7 suggests a related definition, based on practical experience. It is expanded from Reimann et al. (2009). under barley cultivation and the highest emission emerged from the bare soils with 147.8 mol N 2 O m −2 h −1 during the growing season . Soil respiration rates from barley fields ranged between 2.5-8.2 mol CO 2 m −2 s −1 on mineral and peaty soils . From the compiled data, the relatively highest average flux rate of wetlands can be calculated at 629.2 mg CO 2 e m −2 h −1 . Given their low total global land area of 2.7% and compared to flux rates for other land-cover types, these emissions become almost negligible for the planetary surface (Figs. 2 and 3).

Methodologies to quantify GHG emissions from soils
We introduce the most important methodologies for determining trace gas emissions from soils and discuss their individual strengths and weaknesses. In general, trace gas emissions from soils are being directly measured in both field and laboratory (chamber techniques and micrometeorological methods), obtained through space and airborne measurements, and calculated with empirical and process-oriented models (Fig. 4). Given the relevance of reliable soil respiration data for land-use management, chamber systems receive some more attention in this work, since their output allows for more differentiated localized information.

Chamber systems
Flux chamber-based analysis is widely used in soil emission studies of CO 2 , CH 4 , N 2 O and NO (Heinemeyer and McNamara, 2011;Kitzler et al., 2006;Oertel et al., 2012;Pumpanen et al., 2004;Simek et al., 2014). A box or cylinder (diameter XX cm 2 to X m 2 footprint) is placed onto the soil surface so that the section of its base is open to the ground (Fig. 4a). Emitted gases accumulate in its chamber headspace. The change of mixing ratio can be analyzed with various gas sensors, e.g., gas chromatography (CO 2 , N 2 O, CH 4 ), IR-spectrometry including NDIR and FID with and without pumps (CO 2 , CO, CH 4 ), chemiluminescence (NO x ), Cavity-Ring-Down spectrometry (CO 2 , CO, CH 4 , N 2 O, H 2 S) or photoacoustics (CO 2 ,CO,CH 4 ,NO and N 2 O). Problems of too high headspace with inhomogeneous gas concentrations inside the chamber can be prevented by decreasing the chamber height (Rochette, 2011) and by lower detection limits . Results deliver data for the spot scale, being measured at the small spot dimension (Table 7). However, to cover site-dimension fluxes representatively, chamber systems need to be easily and rapidly moved to measure multiple predetermined spots (Oertel et al., 2015a,b).
Chamber systems can also be used to analyze isotopic ratios of C and O species online in combination with a quantum cascade laser-based spectrometer (Kammer et al., 2011). Such methods can be applied to quantify CH 4 oxidation, since CH 4 -oxidizing bacteria use 12 C methane (Börjesson et al., 2007). Delta 14 C of the emitted CO 2 serves to determine the age of the originating carbon source (Gorczyca, 2013). C-isotopes can also be used to distinguish between plant root and microbial respiration (Pausch et al., 2013).
Chamber systems can be divided into closed and open chambers, with closed chambers being subdivided into closed static and closed dynamic ones (Kutzbach et al., 2007;Rochette et al., 1997). Closed dynamic chambers (see Section 3.1.1) may also be referred to as non-steady state flow-through chambers. There is still no standardized chamber system, which may inhibit direct comparison of data sets from different research groups (Pumpanen et al., 2004).
All chamber systems should be equipped with auxiliary sensors to register soil emission-influencing parameters. Sensors for air temperature, pressure and relative humidity should be installed inside and outside the chamber to log ambient conditions and to register the differences from within the chamber. A photosynthetic active radiation (PAR) sensor needs to be additionally installed outside of the chamber, which is indispensable for net ecosystem exchange (NEE) measurements (Burrows et al., 2005). All chambers have to be installed on a collar (of steel or inexpensive PVC) in order to prevent gas leakage from the chamber to the atmosphere. To minimize the influence of the collar on the soil structure and plant roots, the collar should, if possible, be embedded to a depth of a few centimeters . Since collars influence the soil profile, they also influence the flux measurements and need to be installed at least 24 h prior to the first measurement (Bahn et al., 2012). Some chambers may work better without a collar, yet this is not recommended for use on forest soils (Pumpanen et al., 2004). To measure NEE, transparent chambers have to be used (Wang et al., 2013a), while opaque ones serve the determination of ecosystem respiration and of other gases (Sanz-Cobena et al., 2014b). The opaque material also insulates against temperature increases inside the chamber that would lead to pressure changes and influence soil emissions (Xu et al., 2006). A rapid change (few minutes) between transparent and opaque mode is possible with some chamber systems (Oertel et al., 2015b). In such a configuration, NEE and ecosystem respiration can be measured with one system.
If gas production in different soil depth is of interest, gas concentration profiles can be assessed (Chirinda et al., 2014). Samples taken with syringes from specific soil depths may be analyzed in the laboratory by gas chromatography (Petersen et al., 2011). Gas sensors may also be directly installed at specific soil depth for automatic and continuous measurements (Tang et al., 2003).

Closed chambers
Closed static chambers are most common for the analysis of CH 4 and N 2 O fluxes (Pihlatie et al., 2013). These chambers also offer an absorption method for CO 2 analysis, where CO 2 is trapped in an alkaline solution. This permits fluxes to be measured over longer times and replicates at several measurement points without the need for additional sensors (Yim et al., 2002). Yet, this method systematically underestimates CO 2 fluxes (Nay et al., 1994) and is rarely used.
In closed dynamic chamber systems, gases accumulating in the chamber are analyzed either externally and pumped back into the chamber (Rochette et al., 1997;Heinemeyer and McNamara, 2011) or are being analyzed inside the chamber with a compact NDIR-sensor that continuously monitors the atmospheric CO 2 concentrations (Oertel et al., 2015a). N 2 O and CH 4 emissions can be analyzed as well with closed dynamic chambers (Cowan et al., 2015;Yu et al., 2013), but are rarely used compared to static chambers.
Closed static and dynamic chambers are available as both manual and automatic versions. They differ in how ambient conditions inside the chamber are restored. Chambers need to be purged from the soil gases after each measurement. With manual systems the operator needs to lift the chamber off the ground. Automatic chambers may have a moveable lid, comparable to the hinged lid of a traditional beer mug, which enables gas exchange (Edwards and Riggs, 2003;Pape et al., 2009). Alternatively, for some models, the entire chamber is lifted (LI-COR, 2012). A vent needs to be included to prevent pressure gradients during chamber deployment particularly with manual systems (Bain et al., 2005). An additional fan in all chamber types intermixes the inside air to maintain a constant and homogeneous level of increasing emitted gases (Christiansen et al., 2011). Manual chambers are able to cover spatial variability since they are portable and easily repositioned or relocated. No mechanical parts are needed, however, the main drawback is that they have to be permanently operated by hand. To achieve a similar performance with automatic chambers, many more systems are needed. Automatic systems can be used for continuous monitoring and do not have to be assisted, however, they involve higher material costs.
Accumulation times for the gas measurements need to be adapted to the emission rates of the different gases. CO 2 fluxes require the shortest accumulation times (2-4 min; Caprez et al., 2012;Correia et al., 2012;Vesterdal et al., 2012). This requires fast IR-spectrometers that analyze CO 2 fluxes in less than 10 s. During a measurement, the CO 2 -mixing ratio may change from several tens to hundreds of ppm v . Methane measurements take about 60-90 min with sampling intervals of about 20 min, using a gas chromatograph with a manual chamber (Fiedler et al., 2005;Liu et al., 2009a;Sanz-Cobena et al., 2014b). The accumulation time for N 2 O measurements lies between 30 and 90 min (Hayakawa et al., 2009;Lamers et al., 2007;Yao et al., 2009). On average, 5-30 min are needed to accumulate NO (Liu et al., 2009b;Oertel et al., 2012;Yan et al., 2013). Cavity ring-down spectroscopy (CRDS) exists for monitoring systems, where CO 2 , CH 4 , and N 2 O are analyzed from one sample, similar to gas chromatography. Yet, CRDS is faster, measuring every 8 s compared to 3-4 min by gas chromatography (GC), achieves higher precision and does not need additional equipment such as gas generators or gas bottles, thus providing better portability (Fleck et al., 2013;Rella et al., 2013). However, high acquisition costs are involved.

Open chambers
A third type of chamber system is the open dynamic chamber. Its two openings draw in ambient air and generate a continuous gas flow (Kutsch et al., 2009). Gas concentrations are analyzed at the air inlet and outlet of the chamber. The gas flux is calculated by the difference of the concentrations at both ends. Consequently, there are no accumulation times needed, since the flux is analyzed continuously. Continuous measurement systems do not need mechanical parts. Problems due to high headspace mixing ratios do not occur (Balogh et al., 2007). Open chambers are suitable for dry and hot conditions in summer with low gas exchange rates. Closed chamber systems require longer accumulations times under such conditions, leading to temperature and pressure gradients (Balogh et al., 2007). Obviously, open dynamic chambers are technically more sophisticated and more expensive as compared to closed systems. For this reason, economically priced closed dynamic chambers are still the most common systems (Pumpanen et al., 2004).

Data evaluation
Soil flux for all gases can be calculated by linear and nonlinear (exponential) regression, using the slope of the concentration change inside the chamber headspace (Christiansen et al., 2011). The linear model is easier to handle and works best for short chamber closure times, ideal for CO 2 (Forbrich et al., 2010). Therefore, it is widely used and least biased for curves with a convex-upward shape . These authors found that linear regression is more sensitive to relative flux changes and is useful for studies with changing experimental parameters. The lowest detection limits for flux calculations are gathered with linear regression (Parkin et al., 2012). Yet, soil fluxes are significantly underestimated in contrast to the exponential model (Kutzbach et al., 2007), especially for curves with a convex-downward shape . Burrows et al. (2005) explain this with increasing plant stress during measurements. The exponential model is suitable for longer closure times with few data points, e.g., when using a gas chromatograph (Forbrich et al., 2010). We recommend testing the best fit of linear and non-linear models for the measurements of net ecosystem exchange (NEE) during the day, where convexdownward curvatures are common. Evaluation of our own data, calculated with the linear model, showed that NEE during daytime can be more than 50% lower than the flux rate calculated with a suitable non-linear model-even for short measurement periods of five minutes. Nighttime measurements or ecosystem respiration measurements showed only small deviations of less than 10%. Sometimes values calculated with a non-linear model, which should be chosen when calculating fluxes of CH 4 and N 2 O, delivered lower values than values calculated with a linear model.

Micrometeorological methods
The eddy covariance method is a direct micrometeorological approach. It uses vertical turbulences to analyze the turbulent heat and gas exchange between soil surface and atmosphere (Launiainen et al., 2005). A 3-D ultrasonic anemometer and a gas analyzer attached to a tower or mast of at least a 2-m height are needed for this method (Myklebust et al., 2008) (Fig. 4b). The most commonly analyzed gases are CO 2 , CH 4 and N 2 O (Nicolini et al., 2013;Wang et al., 2013a), yet substances like carbonyl sulphides (Asaf et al., 2013) or volatile organic compounds can be determined too (Ruuskanen et al., 2011). Measurements may run continuously and incorporate areas of up to several square kilometers (Myklebust et al., 2008), the medium local scale (Table 7). Eddy covariance integrates plants and trees, and thus completely covers soil, biosphere and atmosphere to determine NEE. The method does not work properly if very low near-ground turbulent mixing occurs. This leads to an underestimation of fluxes (Papale et al., 2006). This also applies if the system is installed within a forest (Kutsch et al., 2009). It is recommendable to perform measurements on levelled ground, above or within low-density vegetation (Baldocchi, 2003;Baldocchi et al., 2001). Data post-processing is complex and measurement gap filling is important to calculate fluxes (Du et al., 2013).
The open-path Fourier Transform Infrared spectroscopy (FTIR) is another near-ground micrometeorological method, using an instrument that can be mounted on a tower or a pole. A radiation source emits the entire IR-spectrum simultaneously (Griffith et al., 2012). The gases modify the response signals that are received by a telescope along the pathway. Such pathways may have standard lengths of 100-500 m (Griffiths et al., 2009);Kelliher et al. (2002) measured CO 2 and N 2 O on a 97-m-long path. Methane, H 2 O and other gases as well as C-isotopes can also be analyzed by FTIR (Griffith et al., 2012).

Laboratory experiments
Laboratory approaches help when the influence of single parameters (e.g., soil temperature or nutrient availability) on soil emissions shall be assessed. Single parameters can be changed, while others are kept constant . Soils from different climate zones can be investigated under controlled temperature and humidity conditions . As an example, Gritsch et al. (2015) analyzed soil monoliths from nine stations across Europe, representing different land-use types. The authors could clearly show dependencies between CO 2 emissions with soil temperature and moisture. Climate chambers that allow full control of temperature and humidity as well as light conditions are used for such experiments. Sieved and homogenized or undisturbed soil (cores) material may be used (small spot scale; Table 7). While undisturbed material does not negatively influence soil structure and microbial life (Petersen et al., 2013;Schaufler et al., 2010;van der Weerden et al., 2010;Yao et al., 2010), the heterogeneity among soil cores demands a larger sample size. Gritsch et al. (2015), for instance, took 33 undisturbed samples per site. Problems lie in destroying roots during sampling and in maintaining constant physical soil core conditions during transport. Influencing parameters can be observed better with homogenized soil material, which is a widespread approach (Aranibar et al., 2004;Feig et al., 2008;Laville et al., 2009;Oertel et al., 2011;Patino-Zuniga et al., 2009). Yet, soil structure is destroyed in the lab and before sieving soil material needs to be air-dried, which inadvertently influences microbial activity.
Lysimeters are an additional option to study soil emissions under controlled conditions in the field. Emissions can be analyzed jointly with the analysis of nutrient leaching (Velty et al., 2007;Zhou et al., 2013). Small field chamber systems can be used both in the laboratory and on lysimeters in the field, while some research groups use chambers, especially designed for laboratory use (Jäger et al., 2011;Schaufler et al., 2010;Yao et al., 2010).

Spaceborne measurements
Remote sensing from satellites may deliver information on GHG soil emissions in two different ways. One approach is to estimate tropospherical, near-surface CO 2 and CH 4 concentrations based on the measurement of the intensity of the reflected sunlight in small wavelength bands in the visible and short-wavelength IR portion of the spectrum (Fig. 4c). Earlier earth observation missions of the European Space Agency (ESA) like ERS-1 and ENVISAT carried low resolution scanning imaging absorption spectrometers for atmospheric cartography (ERS-GOME; ENVISAT-SCIAMACHY; Frankenberg et al., 2005) with a precision of 1-2% (Schneising et al., 2008). Simultaneously, the Japan Aerospace Exploration Agency (JAXA) operated the GOSAT system with a thermal and near-infrared sensor for carbon observations (TANSO), which is operational until present (Kuze et al., 2009). All previous instruments are limited in their applicability for flux estimates (Yoshida et al., 2011). The OCO-2 (Orbiting Carbon Observatory), a NASA satellite with a precision of 1-2 ppm v for CO 2 , can cover the variability of CO 2 and CH 4 sinks and sources with high spatial and temporal resolution (Boesch et al., 2011). It was successfully launched on July 02, 2014, and is designed for a two-year lifetime (Greicius et al., 2014;NASA, 2014). The continuation of the global time series of CO 2 and CH 4 concentrations after the lifetime of GOSAT and OCO-2 is being planned with the Carbon Monitoring Satellite (CarbonSat). CarbonSat is a candidate mission for ESA's EarthExplorer 8 (EE8) satellite to be launched at the end of this decade. Unlike the previous missions, it is designed to map natural and anthropogenic GHG sources and sinks at high spatial resolution (2 × 2 km 2 ) and spatial coverage in order to localize strong emission sources (Buchwitz et al., 2013).
Mapping the spatial distribution and change of land cover types that represent sources or sinks for CO 2 and CH 4 is the alternative to direct estimations of GHG concentrations from remote sensing systems. Coarse to medium resolution remote sensing data deliver a globally consistent and objective source of information for a spatially explicit mapping of the distribution of potential C stocks in terms of land cover type maps (regional to global scale with local resolution; Table 7). However, there is still considerable uncertainty in the spatial agreement of the area and distribution of the relevant land cover types (e.g., grassland, forests, barren land, cropland, wetland) and hence of the globally stored C (Herold et al., 2008;Pflugmacher et al., 2011). These uncertainties are attributed to a number of limitations that are determined by either the technical specification of the sensor (wavelength, spectral and spatial resolution) or the derived data products (e.g., land cover maps). Differences between land cover maps have important implications on modeling global emissions. Thus, the choice of a map might introduce a significant bias in any regional to global carbon balance model.

Airborne measurements
Airborne methods use direct sampling approaches to collect gases from transects, e.g., over different types of land use or from near-surface environments to higher tropospheric altitudes. Amelio et al. (2009), for instance, collected air samples on an ascending and descending flight path of an airplane (between 305 and 3600 m a.s.l.). These samples were stored in flasks and analyzed in the laboratory by gas chromatography for CO 2 ,CH 4 and N 2 O. Yet, airborne measurements deliver data over a short time period only and with common spatial distances of about 10-100 km (Desjardins et al., , 2010Kustas et al., 2006;Pattey et al., 2007; medium local scale; Table 7).

Modeling soil emissions
Since measurements deliver only punctual data, modeling GHG emissions from soils is important to regionalize such measurements and to calculate global budgets. Next to empirical models (e.g., Freibauer and Kaltschmitt, 2003), process-based models that include general physical and chemical processes are used to process the obtained field data (Pattey et al., 2007). Such approaches may cover local to global scales (Table 7; Fig. 4d).
A widely used process-oriented simulation model is the DeNitrification-DeComposition Modell (DNDC), consisting of four sub-models (Li et al., 1992). It can model daily decomposition, nitrification, ammonia volatilization, CO 2 production (soil microbial and root respiration), N-uptake of plants and plant growth (Li et al., 1994). This model is often used to calculate GHG emissions from soils, especially from agricultural lands (Abdalla et al., 2011;Gu et al., 2014;Li et al., 2012). Meteorological and soil parameters (e.g., soil texture, pH-value, bulk density, organic C) as well as vegetation type and management practices (tillage system, fertilizer application, grain yield) serve as important input data (Abdalla et al., 2009b). GHG fluxes from forest ecosystems can be simulated with the Forest-DNDC model (Abdalla et al., 2013). Falloon and Smith (2012) listed further models for the simulation of C and N trace emissions from soils. Butterbach-Bahl et al. (2004) compared model output with onsite measurements for agricultural and forest ecosystems. While they conclude that the modelling approach delivers higher quality estimates, results may deviate by one order of magnitude. However, the average error for soil emissions is generally much lower Hastings et al., 2010) and modeling data may often be used with some confidence.

Method comparison
To compare the different methods with respect to applicability (spatial variability, observable area, continuous monitoring, analyzing processes), the aspects of accuracy and precision (bias, e.g., influence on soil structure) and of costs and workload involved need to be taken into account. There is no single best technique out there. Ideally, investigations would allow for multiple method approaches such as a combination of chamber systems with eddy covariance and remote sensing (Liang et al., 2004;Myklebust et al., 2008).
3.7.1. Applicability 3.7.1.1. Spatial variability, observable area. Chamber measurements are an ideal tool if emissions from distinct sites are to be analyzed. The spatial footprint of an individual chamber measurement generally covers less than 1 m 2 (spot size; Table 7). Different agricultural management practices on soil or plants can be analyzed easily with chamber systems as well. To observe larger areas of up to 10 4 m 2 (Wang et al., 2013a), the number of chambers has to be increased or chambers have to be moved consecutively to cover a larger area (site size, Table 7). When regionalizing chamber flux measurement results, heterogeneity of soils and emission values on small spatial scales have to be kept in mind (e.g., Stoyan et al., 2000). Therefore, the use of representative sampling sites is fundamental. An area of some hectares can be observed with the eddy covariance method (Baldocchi, 2003). The open-path FTIR covers an area of 100-27,000 m 2 (large spot to medium site size; Table 7). The spatial extent is unlimited with modelling and remote sensing approaches. Since modelling or remote sensing data can differ from on-site measurements, validation using ground-based measurements (e.g., chamber measurements) is important, as well as area wide soil property and meteorological data.
3.7.1.2. Continuous monitoring. Chamber systems are suitable for continuous monitoring, except when winter conditions yield higher snow levels. Eddy covariance systems have well known problems at night and during periods of low turbulence (Papale et al., 2006). The open-path FTIR method can be applied during nighttime and in periods without turbulence (Kelliher et al., 2002). Remote sensing and airborne data depend on the orbit of the satellite or the flight frequency and track of the airplane, yet the measuring network is rather sparse over oceans and the tropics. Consequently, more remote sensing data are needed to fill this gap, although it is still challenging to distinguish between sinks and sources (Hungershoefer et al., 2010).
3.7.1.3. Resolution of different processes. To distinguish, e.g., between soil respiration, net ecosystem exchange or photosynthesis, chamber systems have to be used, since this is impossible with eddy covariance (Pumpanen et al., 2012). Nevertheless, entire ecosystem fluxes, e.g., for CO 2 , cannot be measured since bigger plants and trees cannot be included inside chamber systems. Here, eddy covariance, remote sensing or modelling is required. 3.7.2. Accuracy and precision 3.7.2.1. Introduction of bias. Chamber systems are invasive, thus influencing soil emissions (use of collars and the chamber itself). Changes of temperature, pressure and humidity inside the chamber may also lead to biased soil emission values (Rochette and Eriksen-Hamel, 2008;Suleau et al., 2009). Short closure duration of the chamber can minimize these artefacts. Own data showed that the temperature inside the non-transparent chamber increased up to 3 K at non-forest sites during summer. Using the transparent mode, temperature differences can be up to 10 K. In forest ecosystems and under colder conditions, temperature differences are below 2 K. Air humidity measurements differed on the agricultural and forest sites by about 10-20% during summer and <10% on cold, humid days. Using a vent can minimize pressure differences, while it increases the sensitivity to wind forcing (Bain et al., 2005).
Field measurements and modeling for the same plots deliver different results. Pumpanen et al. (2004) tested different chambers under constant conditions with a calibration system. Closed chambers tended to underestimate CO 2 -fluxes by 10%. Closed static chambers, sampled with syringes, underestimated fluxes by up to 35% and decreased errors with shorter accumulation times. Small overestimations of the fluxes between 2 and 4% were found for open dynamic chamber systems (Pumpanen et al., 2004). Myklebust et al. (2008) compared the NEE measured with a dynamic chamber system and with eddy covariance and found no relevant deviations during the non-growing season and wind speeds above 0.2 m s −1 . Wang et al. (2013a) found 4% higher fluxes for NEE on an agricultural site measured with a static chamber system compared to values from eddy covariance. Especially on agricultural sites, comparisons of measurements between eddy covariance and chamber systems are still needed (Wang et al., 2013a).
Compared to N 2 O field measurements with static chambers, Abdalla et al. (2009b) found an underestimation with a deviation of up to 360% for the denitrification-decomposition (DNDC) model. Pihlatie et al. (2013) found an underestimation of about 33% for CH 4 fluxes with static closed chambers. Abdalla et al. (2013) compared eddy covariance data with modeled NEE from the forest-DNDC model and found a good agreement for seasonal and annual values. Field measurements better detect peak values of NEE (Abdalla et al., 2013). Butterbach-Bahl et al. (2009) tested both DNDC models for simulating NO soil emissions from agricultural and forest soils and found flux underestimations of up to 60%.

Costs and workload
Chamber systems are easy to handle and can be operated by one person. These devices are easy to transport and due to this, numerous locations can be analyzed within a short period. By contrast, portability of eddy covariance systems (mounted on poles) is limited.
Chamber systems have the lowest material costs compared to the other methods (Wang et al., 2013a). Material costs for manual and automatic chamber systems that perform online analysis of CO 2 range from EUR 6000-8000 (USD 6700-9000). Commercially available solutions cost about EUR 20,000 D (USD 22,500), e.g., from LiCOR. A cavity-ring-down-spectrometer costs about EUR 40,000 (USD 45,000) and eddy-flux systems ∼EUR 30,000 (USD 33,700). Open-path FTIR-systems range around EUR 60,000 (USD 67,500).
The workload is high, especially with manual chambers, since they have to be supervised continually. Data processing is highest for remote sensing and eddy covariance and data gap filling for nonturbulent periods is important (Wang et al., 2013b). Compared to all other solutions, chamber systems are easy to install by one person without any further equipment.

Synthesis of major findings and conclusions
While the GHG effect has been described a long while ago and Svante Arrhenius published first calculations in 1896 already (Arrhenius, 1896), it was not until the 1970s that the contribution of soils in the global GHG budget was recognized and moved into focus (Hutchinson and Mosier, 1979;Schmidt et al., 1988). It began with individual site (spot) investigations (Table 7). Soon the entire suite of methodological approaches delivered an increasing amount of data and interpretations. It became progressively obvious that the source and sink characteristics of soils with all influencing parameters (as discussed above) could not be neglected in comprehensive global and regional GHG budget considerations (Houghton et al., 1990;IPCC, 2013). This in turn triggered international agreement that individual nations should report to the United Nations about their GHG inventories, including net soil emissions (e.g., Rösemann et al., 2011). Such assessments are mostly based on the IPCC guidelines for national GHG inventories (2006) and their constant advancement. The resulting data deliver the base for global assessments and related mitigation and abatement activities. Yet, we must acknowledge the fact that remote sensing data still do not provide the necessary accuracy and precision and that groundbased data are strongly biased with respect to temperate climate regions and the northern hemisphere at large (Fig. 2). It would be highly desirable to have unified standards and protocols to ensure quality control and data comparability. This includes the necessity for a definition of required meta-data, such as precise information on location, soil and air temperatures, time frame and its relation to annual average meteorological and possibly land-use conditions, instrumentation and its calibration, etc.
Any up-or downscaling of measured data yields a multitude of potential errors and biases (also independent of GHG emissions from soils) that need to be addressed-and overcome (see Reimann et al., 2009Reimann et al., , 2010, for an analogous example). Site data acquisition with chamber systems or micrometeorological methods can deliver reliable and highly precise data, with the latter capable to include small local dimensions. However, attempts to upscale their results to a regional level will likely end in rather improbable results, unless a dense station network can be maintained for an extended period of time, representing land cover and land-use types of that region. So far, no such clusters exist for larger spatial scales (e.g., continental) and long-term monitoring studies are very scarce. Nevertheless, high quality site-dependent data are indispensable to allow for proper model validation and to safeguard model output quality.
Model calculations, as the general base for regional and sub-continental inventories (e.g., in the European Union), are increasingly well able to deliver robust results. Nevertheless, it would be premature to fully trust model outputs as representing reality. The output rather serves obtaining an overview and triggering critical questions, which will eventually further improve input database and algorithms. With better satellite data (higher resolution, enhanced representation of small-scale variation), we can expect input data to improve. Rösemann et al. (2011) give an in-depth evaluation of errors and biases in their inventory for Germany.
For global assessments, a combination of site and localityrelated data, regional data acquisition/reporting and satellite data is used-again through the filter of appropriate models. All real and possible errors discussed above influence the obtained results. It is no major surprise to see differences of at least a factor of two, and often more, when comparing results from different groups or sources. Even differences in the accurate assessment of land-use and land-cover types (e.g., Ramankutty et al., 2008versus FAO, 2014 still lead to noteworthy uncertainties (Fig. 3). Further ambiguity arises, given the rather fast conversion of forestland and other biomes into croplands, especially in (sub)tropical areas with a rate of 48,000 km 2 per year (1999Phalan et al., 2013). Therefore, any current assessment can only deliver a momentary outlook onto this dynamic system. Given the as yet non-satisfactory results from modeling and remote sensing, this shortcoming of bias towards the temperate climate zone also needs to be overcome.
There is not one methodology that covers all spatial dimensions. Also, at this point there is no established standard methodology-neither for determining soil emission rates nor for related GHG-budget calculations. Both scientists and policy makers may not always be free from interests that counter the ideal to obtain and deliver unbiased and highly reliable data (Gillenwater et al., 2007;Winiwarter, 2007). This should be kept in mind, when dealing with this topic.
Various parameters enhance soil respiration. Changes in soil organic content and land-use management have a significant influence on respiration rates. Average CO 2 -emission rates from wetlands exceed those from forestlands, grasslands, croplands and barren lands (Fig. 3). The highest rates were registered at sites located over peat and drained organic soils, followed by sites with managed grasslands.
Methane, the second most potent GHG, has a 25 times higher global-warming potential (per molecule) than CO 2 over a time horizon of 100 years calculated atmospheric residence time (IPCC, 2007). Wetlands are the major source of methane; available data cover a wide spectrum of wetland types under different climates ( Table 6). The limited land cover of wetlands compensates for the high relative emissions, however. Thus, on a gobal scale, wetland emissions appear neither to alter global balances nor to act as a "hidden driver". Emission rates vary widely (0.1-6950 mol CH 4 m −2 h −1 ), because of the diversity in wetland types and climates. At the same time, it needs to be emphasized that data are still rather scarce and certainly not representative on a larger scale. The highest CH 4 emissions were registered during warm seasons and in tropical environments. When wetlands dried up or were used for other purposes, emissions decreased or the soil even became a sink for atmospheric CH 4 ; the related uptake rate depended on its organic content. The average grassland CH 4 -emission rate was below 3 mol m −2 h −1 and for croplands ranged from negative values to a very low rate of <1 mol m −2 h −1 .
Nitrous oxide yields a 298 times higher global-warming potential per molecule than CO 2 (100 years time horizon; IPCC, 2007). The average N 2 O-emission rate from grasslands was low with 10 mol N 2 O m −2 h −1 . This is higher, however, than the rate for croplands (<5 mol m −2 h −1 ) except from barren dry organic soils. Grazing and no-tillage practices on grasslands enhanced N 2 Oemission rates. If we neglect wetlands for the global balance due to their very limited land cover, grasslands show the highest total CO 2 -equivalent emission when compared with other land-cover types and are the most widespread type of land use (about 25% of land area worldwide). Thus, grasslands have the most promising opportunities for GHG mitigation. However, the N 2 O-emission rate from wetlands ranged from negative values (sinks for N 2 O) to more than 23 mol N 2 O m −2 h −1 during warm seasons.
Through our literature review, limitations in the database became evident, particularly in respect to measured soil respiration in the field using chamber systems. Also, a deficiency in crop diversity information was obvious, with rather limited information about fruit trees (orchards) and vegetable production. The need to explore GHG emissions from fields under other crops appears obvious (Table 4). Most studies that focused on wetlands registered only CH 4 , as compared to investigations on croplands and grasslands that yielded a more comprehensive GHG coverage.
Other land-cover types such as urban soils are currently radically underrepresented. Based on first and certainly insufficient data from our group, we assume that soil respiration from partly sealed areas in the temperate climate zone is comparable with wintertime emissions from agricultural fields (cropland). Pavements with plant or moss in the interstices can considerably exceed wintertime emissions, however if the non-sealed proportion of the pavement is extrapolated to the whole measuring spot. Emissions from urban meadows or flowerbeds are in the range of natural grasslands and agricultural fields in the respective season.
While our understanding is obviously increasing -and many new questions that now demand answers were no issue only a couple of years or decades ago -there still remain various sources of errors and uncertainties. These start with the calculated estimates for climate zones, biomes and land-cover areas. Depending on the initial questions, the tools used, and on the chosen scale for any specific analysis, errors up to roughly 10-20% may emerge. A larger relative error relates to GHG-emission data. Again, depending on the methodology used (and there are no established standards so far), results may vary considerably from both ends, when upscaling (e.g., from site measurements) as much as downscaling (e.g., from remote sensing observations). Related errors may easily reach and exceed 50%. Another source of errors relates to the rather uneven distribution of real data that still generates a bias between the northern and the southern hemispheres, their biomes and land-use patterns (Fig. 2). Thus, it may not be surprising that data published for comparable global calculations may deviate considerably from reality. If we use analogies from global data assessment, e.g., the World Soil Average (Koljonen, 1992), there is a strong bias resulting from an overrepresentation of studies with data from the northern hemisphere and temperate climate, as a recent study shows (Caritat et al., 2012). More experimental and monitoring data are needed with an area-representative distribution, even if it was just for a better validation of numerical models. Another aspect to be considered relates to homogenization and equalization of data. To this point, no unified norm or agreed-upon recommendation exists that defines quality standards to be applied (1) to site selection, (2) to methodological set-up of appropriate flux and concentration units, and (3) to data display. With many publications, metadata are missing or insufficient-making the set-up and boundary conditions for the measurements indiscernable and thereby obscuring the data.
Despite all criticism and shortcomings, the achievements and progress made on investigating GHG emissions from soils should not be neglected either. There is an increasing understanding of the very large role of soils as sinks and sources mainly for CO 2 , CH 4 and N 2 O, and of the various feedback processes involved. This understanding goes hand in hand with an appreciation of the massive power of intervention in the natural feedback processes through land-use (mis)management. While individual studies exist, their results are far from sufficient to be used in models and for policy counselling. The key drivers of soil GHG emissions are known and yet important geo databases such as truly representative maps for total carbon and nitrogen as well as total organic carbon are lacking or available for very restricted geographical areas only (e.g., Reimann et al., 2014). Such globally incomplete databases impede true and verified global representativeness.
with the figures and the Verein der Freunde und Förderer of TU Bergakademie Freiberg for supporting the possibility to print colour figures.