The European forest sector: past and future carbon budget and fluxes

Abstract. The comprehensive analysis of carbon stocks and fluxes of managed European forests is a prerequisite to quantify their role in biomass production and climate change mitigation. We applied the Carbon Budget Model (CBM) to 26 European countries, parameterized with country information on the historical forest age structure, management practices, harvest regimes and the main natural disturbances. We modeled the C stocks for the five forest pools plus harvested wood products (HWPs) and the fluxes among these pools from 2000 to 2030. The aim is to quantify, using a consistent modeling framework for all 26 countries, the main C fluxes as affected by land-use changes, natural disturbances and forest management and to assess the impact of specific harvest and afforestation scenarios after 2012 on the mitigation potential of the EU forest sector. Substitution effects and the possible impacts of climate are not included in this analysis. Results show that for the historical period from 2000 to 2012 the net primary productivity (NPP) of the forest pools at the EU level is on average equal to 639 Tg C yr−1. The losses are dominated by heterotrophic respiration (409 Tg C yr−1) and removals (110 Tg C yr−1), with direct fire emissions being only 1 Tg C yr−1, leading to a net carbon stock change (i.e., sink) of 110 Tg C yr−1. Fellings also transferred 28 Tg C yr−1 of harvest residues from biomass to dead organic matter pools. The average annual net sector exchange (NSE) of the forest system, i.e., the carbon stock changes in the forest pools including HWP, equals a sink of 122 Tg C yr−1 (i.e., about 19 % of the NPP) for the historical period, and in 2030 it reaches 126, 101 and 151 Tg C yr−1, assuming constant, increasing (+20 %) and decreasing (−20 %) scenarios, respectively, of both harvest and afforestation rates compared to the historical period. Under the constant harvest rate scenario, our findings show an incipient aging process of the forests existing in 1990: although NPP increases (+7 %), heterotrophic respiration increases at a greater rate (+13 %) and this leads to a decrease in the sink in the forest pools (−6 %) in 2030 compared to the historical period. By comparing the evolution of the biomass as a function of the NPP (i.e., the turnover time) for each country, we highlighted at least three groups of countries and turnover times. This means that, contrary to the assumptions proposed by other authors, this relationship cannot be assumed as a constant for all the EU countries, but specific conditions, such as the harvest rate, the current age structure and the forest composition, may contribute to the country-specific evolution of biomass stocks. The detailed picture of the C fluxes condensed in this study, and their evolution under different harvest scenarios, may represent both a benchmark for similar studies and a basis for broader analyses (e.g., including substitution effects of wood) on the mitigation potential of the EU forest sector.


11
The comprehensive analysis of carbon stocks and fluxes of managed European forests is a 12 prerequisite to quantify their role in biomass production and climate change mitigation. We applied 13 the Carbon Budget Model (CBM) to 26 European (EU) countries, parameterized with country 14 information on the historical forest age structure, management practices, harvest regimes and the 15 main natural disturbances. We modeled the C stocks for the five forest pools plus Harvested Wood 16 Products (HWP), and the fluxes among these pools, from 2000 to 2030. The aim is to quantify, 17 using a consistent modelling framework for all 26 countries, the main C fluxes as affected by land-18 use changes, natural disturbances and forest management and to assess the impact of specific 19 harvest and afforestation scenarios after 2012 on the mitigation potential of the EU forest sector. 20 Substitution effects and the possible impacts of climate are not included in this analysis. 21 Results show that for the historical period (2000 -2012) the net primary productivity (NPP) of the 22 forest pools at the EU level is on average equal to 639 Tg C yr -1 , the losses are dominated by 23 heterotrophic respiration (409 Tg C yr -1 ) and removals (110 Tg C yr -1 ), with direct fire emissions 24 being only 1 Tg C yr -1 , leading to a net carbon stock change (i.e. sink) of 110 Tg C yr -1 . Fellings 25 also transferred 28 Tg C yr -1 of harvest residues from biomass to dead organic matter pools. The 26 average annual Net Sector Exchange (NSE) of the forest system, i.e. the carbon stock changes in 27 the forest pools including HWP, equals a sink of 122 Tg C yr -1 (i.e., about 19% of the NPP) for 28 the historical period and in 2030 reaches 126 Tg C yr -1 , 101 Tg C yr -1 and 151 Tg C yr -1 , assuming 29 respectively a constant, increasing (+20%) and decreasing (-20%) scenario of both harvest and 30 afforestation rates compared to the historical period. Under the constant harvest rate scenario, our 31 findings show an incipient aging process of the forests existing in 1990: although NPP is increasing 32

48
Forest management in Europe has a long tradition that has strongly influenced the present species 49 composition (Spiecker, 2003) and it will continue to be the main driver affecting the productivity 50 of European forests for the next decades (Koehl et al., 2010). A comprehensive assessment of the 51 overall carbon stocks and fluxes of managed forests is required to complement the analyses of 52 climate change impacts on forest productivity and composition (e.g. Lindner et al., 2015). Several 53 studies analyzed the European forest carbon budget from different perspectives and over different 54 time periods (Kauppi et al., 1992, Karjalainen et al., 2003, using different approaches, such as 55 process-based ecosystem models (i.e., Valentini et al., 2000) or estimates based on forest 56 inventories (i.e., Liski et al., 2000). Each of these methods has its strengths and weaknesses 57 (Karjalainen et al., 2003). 58 Although several studies tried to harmonize different data sources (i.e., Böttcher  The aim of this study is to provide a comprehensive quantification of the carbon stocks and fluxes 76 of the EU forest sector, including country-level details. We used an inventory-based model 77 (Carbon Budget Model, CBM-CFS3, Kurz et al., 2009) and applied it to 26 EU countries for the 78 historical period 2000-2012 and for future scenarios of different harvest and afforestation rates (up 79 to 2030). 80 In particular, we focus on the effects of forest age-structure, natural disturbances, land-use change 81 and management activities on: (i) the amount of carbon stocked in the five forest C pools (i.e., 82 above-and belowground biomass, dead wood, litter, and soil) and outside the forest (i.e., harvested 83 wood products, HWP), when possible further distinguishing between merchantable biomass, 84 branches, biomass used for energy, etc.; and (ii) the fluxes, i.e., the inputs to and the outputs from 85 each pool, and the exchanges between the forest sector and the atmosphere. The CBM is an inventory-based, yield-curve driven model that simulates the stand-and landscape-93 level C dynamics of above-and below-ground biomass, dead organic matter (DOM: litter and dead 94 wood) and mineral soil (Kurz et al., 2009

Harvest demand and carbon flow 164
The main fluxes modelled in our study are: (1) inputs of C from the atmosphere (i.e., NPP) to the 165 forest ecosystem; (2) outputs due to direct C emissions from the forest to the atmosphere and due 166 to harvest activities; (3) internal fluxes (not affecting the total C balance), mainly from the living 167 biomass to the DOM pool (see also Figure 1S in the Supplementary Materials for more details). 168 Carbon enters the forest as CO2 absorbed from the atmosphere by living biomass (LB); a fraction 169 of this biomass returns to the atmosphere (through natural disturbances such as fires and storms) 170 or moves to the other forest pools (dead wood and litter) through natural mortality and disturbance 171 events. From these pools, C can be directly released to the atmosphere or transferred to the soil 172 pool where some of it can reside for centuries. All these ecosystem carbon fluxes are modeled in 173 CBM with a semi-empirical approach (Kurz et al., 2009). 174 From an ecosystem perspective (Kirschbaum et al., 2001), the sum of all biomass production, 175 during a year, represents the NPP, equal to the difference between the carbon assimilated by plants 176 through photosynthesis (i.e., the Gross Primary Production, GPP) and the carbon released by plants 177 through autotrophic respiration (Ra): 178 Subtracting from this figure all the C losses due to the heterotrophic respiration (Rh, i.e., 180 decomposition), we estimate Net Ecosystem Productivity (NEP): 181 NBP is the difference between NEP and the direct losses due to harvest (H) and natural 183 disturbances (D, e.g., fires): 184 Through the fellings, a fraction of the LB moves to the HWP pool (this is the amount of biomass 186 removed from the forest, i.e. the roundwood removals reported in Figure 1S). Another fraction of 187 biomass is left in the forest as forest residues (i.e., slash, varying according to the specific 9 them from the dead wood pool to the roundwood pool. Adding to the NBP the total changes in the 190 HWP carbon stock (HWPΔC), we estimate the Net Sector Exchange (NSE, Karjalainen et al., 191 2003): 192 In this study, we applied the CBM as a timber assessment model, i.e., we defined a certain harvest 194 level and implemented the model to (i) check if it is possible to harvest that amount and (ii) to 195 simulate the forest development under that harvest level (Schelhaas et al., 2007). The total fellings 196 were inferred, for each country, from the amount of roundwood removals reported by FAOSTAT 197 data (FAOSTAT, 2013), further distinguished between industrial roundwood (IRW, used for the 198 production of wood commodities and mainly provided by stems) and fuelwood (FW i.e., the wood 199 for energy use, mainly provided by residues, branches and coppices). To provide a consistent 200 estimate of the harvest demand for all the countries, these data were compared and, when needed, 201 corrected with other information from the literature (i.e., to account for the bark fraction or other 202 possible recognized biases; Pilli et al., 2015). 203 The EU-26 total past and three alternative future harvest demands considered in this study are 204 shown in Figure 1. For each country, the total harvest was further distinguished between four 205 compartments providing the total amount of wood expected each year: IRW conifers, IRW 206 broadleaves, FW conifers and FW broadleaves. For each compartment we defined: (i) the FTs (i.e., 207 broadleaved species for IRW and FW, and coniferous species for IRW and FW), (ii) the MTs (for 208 example coppices for FW broadleaves) and (iii) the silvicultural practices (for example thinnings 209 for FW conifers). Original values of harvest demand expressed as cubic meter were converted to 210 tons of C using species-specific wood densities values and a constant C fraction equal to 0.50 211 (Penman et al., 2003). A further distribution between FTs and MTs associated with the same 212 compartment was based on the total stock of aboveground biomass available at the beginning of 213 the model run. The C annually stocked as harvested wood products (i.e., IRW) was directly derived 214 by the estimates provided by Pilli et al., 2015, based on the same input data used in this study. 215 During the model run, we also quantified the amount of FW provided by branches and other wood 216 components such as the amount of residues moved from the LB to the dead wood pool (see Figure  217 lack of detailed information on this potential use, this amount was not included in the sum of the 219 total roundwood removals; instead it was assumed as direct emission of C to the atmosphere. the total harvest demand between the four compartments (i.e., IRW and FW, Con. and Broad.), 227 assuming the same proportions as in the historical period, i.e, about 62% of the total harvest was 228 used as IRW coming from coniferous species, 19% was used as IRW coming from broadleaved 229 species, 6% was used as FW coming from coniferous species and 13% was used as FW coming 230 from broadleaved species. 231 [ Figure 1] 232 We assumed that the harvest demand was entirely provided by the FM area, excluding potential 233 harvest from deforestation. For AR we estimated the maximum potential (and theoretical) harvest 234 from afforested areas, assuming a common set of silvicultural practices for all countries, with a 235 single 15% commercial thinning applied to broadleaved forests 15-years or older and a single 20% 236 commercial thinning applied to coniferous forests 20-years or older (Pilli et al., 2014b).

Carbon balance at EU level 248
The average total C stock estimated for EU-26, for the main FM pools is equal to 9,417 Tg C for 249 the living biomass; 1,536 Tg C for dead wood; 1,179 Tg C and 7,717 Tg C for litter and soil (to a 250 depth of 1 m), plus 1,843 Tg C, as average amount of C in the HWP pool during the same period 251 (based on the analysis provided by Pilli et al., 2015). 252 2012, we estimated that, on average, 8 Tg C yr -1 were moved from the living biomass to DOM due 265 to natural disturbances, and, apart from direct CO2 emissions due to wildfires (about 1 Tg C yr -1 ), 266 these processes also increased the indirect emissions due to heterotrophic decomposition of fire-267 killed biomass (Ghimire et al., 2012). Due to the short time frame considered in our study, we 268 could not identify any significant variation of the soil C stock. The slightly negative C stock change 269 reported for this pool (-0.7 Tg C yr -1 ) is mainly due to the effect of deforestation that moves 270 forested lands to other land-use categories (i.e., as reported in Figure 2, it is not a soil C loss to the 271 atmosphere, but it is a C transfer to other land-use categories). Overall, the soil C stock is stable. 272 [ Figure 2] 273 1990. The total heterotrophic respiration (Rh) amounts to 403 Tg C yr -1 , mainly due to the decay 276 of the DOM and soil C pools, plus 6 Tg C yr -1 from the afforested area. 277 The direct C emissions related to fire disturbances are about 1 Tg C yr -1 (see Figure 1S  related to fellings (about 138 Tg C yr -1 ) and can be distinguished between wood removals (110 Tg 281 C yr -1 ) and transfers of biomass residues to DOM pools, (28 Tg C yr -1 ), which will decay over time 282 (see Figure 1S). A consistent fraction (about 20%) of the fellings are used as fuelwood and thus 283 its C content is directly released to the atmosphere (see Figure 1S and Tab We compare our results with figures from the literature (Table 3) Europe: 5.2 ± 0. 7 Mg C ha -1 and 0.6±0.1 Mg C ha -1 , respectively. 308 Taking into account all these fluxes, we estimated a total NBP equal to 98 Tg C yr -1 and 12 Tg C 309 yr -1 for the FM area and the afforested area (146 M ha in total), respectively. Adding to these NPB 310 estimates the C stock increases in the HWP pool, we estimate a Net Sector Exchange (NSE) for 311 the total forest sector of 122 Tg C yr -1 . Luyssaert  rates, and represent about 11% of the total NPP. In contrast, the IRW C sink, equal to 12 Tg C yr -343 1 for the historical period, decreases when assuming a constant (8 Tg C yr -1 ) or a decreasing (2 Tg 344 C yr -1 ) harvest scenario. When we assume an increasing harvest, the HWP C sink in 2030 increases 345 slightly from 12 to 13 Tg C yr -1 . 346 Subtracting from the initial NPP the emissions due to the natural turnover rate (panel E), natural 347 disturbances and deforestation (panel A) and fellings (panel D), we can estimate the final C sink 348 of (i) the FM area (including the effect of deforestation), (ii) the HWP pool (stored outside the 349 forest), (iii) the AR that occurred from 1990 to 2030 and (iv) the total forest sector sink. The C 350 sink of the FM area (excluding HWP) varies from 98 Tg C yr -1 for the historical period, to 92 Tg 351 C yr -1 , 61 Tg C yr -1 and 123 Tg C yr -1 assuming a constant, increasing and decreasing harvest 352 scenario. This means that, even maintaining a constant harvest rate from 2013 to 2030, the final 353 NBP of forests existing in 1990 decreases by 6% in 2030, compared with the historical period. 354 Increasing the harvest demand by 20%, the NBP decreases by 37% in 2030, but in all cases the 355 NBP estimates a C sink. Only when the harvest demand decreases, will the NBP increase by 25%. 356 The declining C sink estimated in the constant harvest scenario, is the results of an increasing NPP 357 (+7%, if compared with the historical period, see Tab. 1S for details), combined, but with an 358 opposite effect, with an increasing natural turnover and consequent emissions from DOM pools to 359 the atmosphere (+13%). This confirms an age-related decline in the productivity of the European 360 forests (Zaehle et al., 2006), and it is consistent with the results from other studies in the literature, Overall, for the historical period, the NBP of the FM area equals 16% of the NPP (i.e., the input 363 to the forests). This means that about 84% of the NPP is lost due to natural and human activities. 364 In 2030, the proportion of NBP in NPP varies considerably: from 9%, for the increasing harvest 365 scenario, to 18%, for the decreasing harvest scenario. Since a fraction of the NPP is still stocked 366 in the HWP products, adding this amount to the FM NBP we can estimate the total C sink, i.e., the 367 Net Sector Exchange. In this case, the NSE increases to 110 Tg C yr -1 (i.e., about 18% of the NPP) 368 for the historical period 2000 -2012. This value is considerably higher than the NSE reported by 369 Karjalainen et al. (2003), equal to 87 Tg C yr -1 , but for a lower area (128 Mha compared to 138 370 Mha) and a slightly different period (1995 -2000). In 2030, the NSE varies from 100 Tg C yr -1 to 371 74 Tg C yr -1 and 126 Tg C yr -1 assuming a constant, increasing and decreasing harvest scenarios, 372 respectively (excluding AR). This means that, excluding the substitution benefits and avoided 373 emissions from the use of harvested wood products ( Therefore, reducing the harvest by 20% will decrease the energy potential of the FW proportionally 384 and, vice versa, increasing the harvest by 20% will increase the energy potential of the FW. 385 Several studies suggest a significant increase in harvest removals at EU level for the next decades, 386 mainly due to increasing wood demand for renewable energy production, i.e., the FW demand 387 same study, because of ageing managed forests, this would result in a 30% decline of the forest C 391 sink in 2030, compared to 2005. In our study, increasing the harvest by 20% resulted in a slightly larger reduction of the C sink, equal to about 38%. Since, in the increased harvest scenario, the 393 HWP C sink equals 13 Tg C yr -1 , reducing the share of IRW, further increases in the FW 394 production, would also further reduce the total C sink. 395 The average annual NBP on AR lands from 1990 to 2012 is equal to 12 Tg C yr -1 , i.e., about 62% 396 of the AR NPP. Assuming different afforestation rates from 2012 to 2030, the final NBP in 2030 397 is equal to 26 Tg C yr -1 , 27 Tg C yr -1 and 25 Tg C yr -1 , with a constant, increasing and decreasing 398 AR rate, respectively (Table 3). Compared with the historical period, the ratio between NPP and 399 NBP considerably decreases (about -46%), because the potential amount of harvest on AR lands 400 increases from 1 Tg C yr -1 for the historical period, to about 6 Tg C yr -1 in 2030 for all three AR 401 scenarios. While the amount of wood available for harvest until 2012 is negligible (because of the 402 young age of the new forests established since 1990), in 2030, the potential amount of harvest 403 from AR increases, but even then it can only provide less than 6% of the total EU harvest. In our 404 study, we assumed that this amount was mainly used as FW, i.e., the C was immediately oxidized. 405 A further potential amount of harvest, eventually used as FW or IRW, can be provided by the 406 biomass removed from deforested areas, equal on average to about 5 Tg C yr -1 for the historical 407 period. Due to the lack of detailed information on this use, this amount, equal to about 20 M m 3 408 yr -1 (i.e., about 4% of the average amount of harvest from 2000 to 2012), was quantified but not 409 accounted in the sum of the total roundwood removals and included in the total emissions due to 410 deforestation (see Figure 2 and Figure 1S). This simplified assumption is consistent with the 2013 411 IPCC KP LULUCF Supplement (Hiraishi et al., 2014), which suggests to assume an instantaneous 412 oxidation of the harvest originating from deforestation. On the opposite, when assuming that this 413 amount is used as FW or IRW, we should reduce the amount of living biomass removed through 414 other management practices (see Figure 1S, arrows (E), (F), (G)). This would slightly increase the 415 living biomass C stock (see Tab 1S: from 7,228 Tg C to 7,233 Tg C, i.e., + 0.07% yr -1 ) and, as a 416 consequence, the NBP of the FM area, but it would not affect the direct emissions due to FW and 417 to the decay process affecting IRW, since the absolute amount of FW and IRW would not change. 418 Adding to the previous estimates the C sink related to AR, the total NSE of the forest system in 419 2030 is equal to 126 Tg C yr -1 , 101 Tg C yr -1 and 151 Tg C yr -1 , assuming a constant (harvest and 420 AR rate), increasing and decreasing scenario (see Table 1S). Compared with the historical period 421 (with a total NSE equal to 122 Tg C yr -1 ) these values are slightly higher (+3%), lower (-17%) and higher (+23%), for the constant, increasing and decreasing harvest and AR scenarios, respectively. 423 Looking at the constant harvest and AR scenarios, these results suggest that the decreasing C sink 424 detected on the FM area is partly compensated by the increasing C sink on the afforested area. 425 These results are based on the assumption that the highest harvest demand is combined with an 426 increasing AR rate, and vice versa. Different combinations of harvest and AR rate however may 427 also be possible (see the Tab. 4) but, excluding the FW energy potential, the maximum C sink is The total losses due to natural processes, such as the decomposition of organic matter, fires and 445 human activities (i.e., harvest, orange slice of each external pie in Figure 4) vary between -2.2 Mg 446 C ha -1 yr -1 in Finland and -8.2 Mg C ha -1 yr -1 in Ireland. The EU average is -3.8 Mg C ha -1 yr -1 . As 447 expected, these losses vary proportionally to the absolute NPP value, and on average the total 448 losses amount to about 83% of the NPP. The highest proportion of losses was estimated for 449 Belgium (>95% of the NPP) and the lowest for the UK (<70% of the NPP). 450 The average NBP (white internal pie in Figure 4) is equal to the difference between the average 451 NPP minus the losses due to respiration (Rh), harvest (H) and disturbances (D) and varies between 452 0.1 Mg C ha -1 yr -1 estimated for Belgium and 2.4 Mg C ha -1 yr -1 for UK. Adding to the NBP the 453 HWP net sink (also highlighted by the external orange pies on Figure 4), we can estimate the NSE 454 (labels in Figure 4). This amount varies between 0.1 Mg C ha -1 yr -1 in Belgium and 2.7 Mg C ha -1 455 yr -1 in the UK. 456 Since forest losses are due to the combined effect of natural processes and harvest and they directly 457 affect the final NEP, a more detailed analysis of these parameters may provide useful information. 458 [ Figure 4] 459 [ Figure 5] 460 In Figure 5 we distinguished the relative amount of C losses due to 9 different processes, including 461 natural (i.e., fires and release of C due to the decomposition of DOM and soil pools) and human 462 factors (i.e., harvest activities) and we estimated the percentage loss of the total NPP due to each 463 process. The largest release of C to the atmosphere from the forest ecosystem is due to the natural 464 decomposition of dead wood and litter pools (i.e., DOM  atmosphere). In all countries, this 465 covers at least 37% of total losses while at the EU level it equals 51% of total NPP. 466 The second factor contributing to the total absolute amount of losses is generally represented by 467 human activities, i.e., the use of the merchantable wood components as industrial roundwood. 468 Unlike the previous factor, the relative contribution of this factor varies considerably among 469 countries. In some cases, this may represent more than 20% of the total NPP (e.g., Belgium), but 470 in other countries this share may be less than 3% (i.e., Greece and Italy). At the EU level, 471 merchantable wood use represents about 12% of total NPP. 472 Releases of C from soil to the atmosphere represent the third factor contributing to the total losses 473 (on average 13% of the total NPP). Of course, due to the lack of data, and similarly to other soil and DOM, is also affected by local climatic conditions. In our modelling framework, we linked 478 the forest area to specific CLUs, associated with values of mean annual temperature and total the decomposition rate for each DOM pool is modelled using a temperature-dependent decay rate 481 (Kurz et al., 2009) which allowed us to consider the effect of regional climatic on decay. Due to 482 the lack of data, we did not differentiate biomass turnover rates by region. 483 For all EU countries, further losses are due to the use of wood for energy. While the IRW is 484 generally provided by the merchantable wood components (or, in some cases, by salvage logging 485 after storms). Based on our assumptions (see also Figure 1S), the FW may be provided through 486 three different sources of materials: merchantable components (e.g., from coppices or early 487 thinnings), other wood components (mainly branches harvested simultaneously with merchantable 488 wood used as IRW) or standing dead trees (i.e., snags, even as salvage logging after fires). The 489 relative share of these three sources varies considerably among countries but it is generally < 5%. 490 In few countries, the total losses due to the use of wood for energy exceeds 8% of the total NPP 491 (e.g., France), but at the EU level equals, on average, 4%. 492 The total losses due to natural disturbances were only accounted for in 22 countries, while 4 493 countries do not report natural disturbance events. At the EU level, for the historical period 2000 494 -2012, these represent about 1% of the total NPP. In some countries, however, this percentage 495 may represent, on average, more than 2%. This is the case of Austria, due to the effect of storms 496 and insect attacks, and Portugal due to fires. Natural disturbances may cause direct losses, due to 497 the biomass and dead organic matter burned by fires (i.e., a direct emission of C to the atmosphere) 498 or indirect losses from the forest ecosystem, due to the salvage of logging residues, after the 499 disturbance events or the decay of biomass that was killed during the natural disturbance and 500 transferred to the DOM pools . 501 We also report the relative amount of losses due to deforestation on the FM area. At the EU level, 502 deforestation represents less than 2% of the total NPP and, for the majority of the countries, less 503 than <1%. In few cases, however, due to the relative large amount of deforestation compared with 504 the total FM area (based on the KP CRF tables, 2014), the deforestation losses may be higher than 505 4% (France and Luxemburg) and, for Netherlands, equal to 19% of the total NPP. This country 506 reports an annual rate of deforestation equal to 2,000 ha yr -1 (KP CRF, 2014), i.e., about 6% of the 507 FM area. 508

Carbon turnover time 509
Overall, our study suggests that, in the majority of European countries, the build-up of biomass 510 stocks results from woody NPP exceeding losses by harvest and natural disturbances, as 511 highlighted by Ciais et al. (2008). While some estimate biomass carbon stocks as a function of Country-specific forest conditions related to management practices, harvest rates, past age 547 structures and forest composition, have varying impacts on the evolution of biomass stock and 548 NPP. Above all, the turnover time estimated for the living biomass seems to be related to the age 549 structure and management practices. Indeed, countries with older forests (such as UK) and longer 550 rotation lengths applied to clearcuts, have the highest τ (>80 yrs). In Italy, where clearcuts are often 551 replaced by other silvicultural practices such as thinnings or partial cuts and where a large part of 552 the forest area (mainly coppices) is aging because of a relative low harvest demand (Pilli et al.,553 2013), τ is also over 80 years. An increasing harvest demand, generally combined with a larger use 554 of final cuts and shorter rotation lengths, gradually reduces the turnover time and the average age 555 of the forests. Moreover, exceptional natural disturbances, such as windstorms or fires, may further 556 modify this parameter. Due to the complex interaction between these variables, further analyses 557 are needed. 558 [ Figure 6] 559

Uncertainties 560
Quantifying the overall uncertainty of these estimates is challenging because of the complexity of 561 our analysis. Indeed, the EU estimate is obtained by summing up 26 country level estimates. For 562 each country, the C stock of each pool is obtained by multiplying the area of each age class (further 563 distinguished between different FTs and administrative units) with the corresponding volume and 564 by applying a species-specific equation to convert the merchantable volume to total aboveground 565 biomass (used as a biomass expansion factor). Therefore, we first consider the uncertainty related 566 to the area, the volume and the equation applied to each FT.
The uncertainty of the area estimates varies among countries. Generally, the information from east 568 European countries have a higher uncertainty because of low updating frequency or heterogeneous 569 data sources (e.g. for forest in Romania, Blujdea, pers. com.), while the most recent NFIs have 570 lower uncertainty (e.g., <1%, at the country level, e.g. for Germany or Italy). Considering that the 571 average reference year of the NFIs applied by our analysis is 2003 (see Tab. 1) we assume that the 572 uncertainty of the area (at the country level) is equal to 2%. 573 The volume reported by the yield tables applied by CBM derives from a linear interpolation of the 574 volume and increment data reported in each NFI. The uncertainty on these data (when reported) 575 may vary considerably, depending on the relative abundance of each FT (i.e., by the number of 576 plots) but, based on an overview of the NFIs applied to our analysis, we may assume that it is equal 577 to 5% (in most cases, however, the uncertainty estimate is missing). 578 Estimating the uncertainty related to the biomass equations applied to each FT is even more 579 challenging. These equations were preliminarily selected comparing some values available at 580 country level (for 8 out of 26 countries, considering the main FTs and biomass compartments) 581 with the values estimated through specific multinomial models developed by Boudewyn et al. 582 (2007). For each FT, administrative region and biomass compartment, we selected the equation 583 that minimizes the average sum of squares of the differences between the values predicted by the 584 equations and reported in the literature (see Pilli et al., 2013). Therefore, the uncertainty on this 585 component is related to both the uncertainty of the original values reported in the literature and of 586 the multinomial model selected by our analysis. The first uncertainty may vary considerably, 587 depending on the original data source selected for each country. For example, based on NFI data 588 reported for Italy, the standard error of the aboveground biomass estimated at the regional level 589 may vary between less than 3% to more than 100% (Gasparini and Tabacchi, 2011). For Germany, 590 and for other countries where no detailed information on the biomass was available and this 591 parameter was estimated through allometric equations applied to the original NFI data, the 592 uncertainty may also be higher. 593 The uncertainty related to the capacity of each model to represent the original values was estimated 594 through the mean percentage difference between the predicted and observed values. This may vary 595 considerably, depending on the forest compartment and the species. For Italy, the mean percentage 596 difference between the total aboveground biomass estimated using the selected stand-level equations and the biomass reported by NFI was ±3.8% (Pilli et al., 2013). For other countries, we 598 obtained similar results. Where no data were provided by the literature (i.e., for 18 out of 26 599 countries), we applied the same equations selected for other countries, for similar FTs. Of course, 600 this may further increase the uncertainty of our estimates. 601 Attributing an overall uncertainty equal to 2% (UA), 5% (UV) and 3.8% (UB) to the input data on 602 the area, the volume and the expansion of the volume to total living biomass, respectively, and 603 without considering further possible uncertainties (i.e., of the original input data reported by NFIs 604 and of singular FTs and regions), and actual correlations between NFI measured variables, the 605 overall uncertainty on the living biomass stock may be estimated as (Penman et al., 2003): 606 = √ 2 + 2 + 2 = 6.6% Eq. (4) 607 The estimates on the C stock change and, indirectly on the fluxes, are affected by additional 608 uncertainties about the amount of harvest and the amount of area affected by natural disturbances. 609 Comparing different data sources such as NFIs or FAOSTAT data, Pilli et al. (2015) highlighted 610 the inconsistencies of harvest statistics and the uncertainties of these data, which may vary 611 considerably among countries. For example, the Italian NFI reports a 13.3% uncertainty on the 612 amount of harvest, while the German NFI reports a 1.2% overall uncertainty. This also affects the 613 uncertainty on the net-emissions associated to the HWP pool, which also depends on the 614 initialization and on the decay rate for each wood commodity (i.e., sawnwoods, wood based panels 615 and paper and paper board), on the relative fraction of HWP coming from domestic forests and on 616 other sources of uncertainty (described in detail by the 2013 IPCC KP LULUCF Supplement, 617 Hiraishi et al., 2014). 618 Quantifying the uncertainty of the input data for natural disturbances is even more challenging. 619 Due to the lack of data, the uncertainty of land-use change (i.e., afforestation and deforestation), 620 dead organic matter and soil C pools is even higher. Based on the information reported in the 621 countries' Greenhouse Gas Inventories, for the forest land category, the uncertainty reported by 622 the individual EU member states ranges between 15-77% for the living biomass, between 22-113% 623 for dead organic matter and between 13-62% for mineral soils (Blujdea et al., 2015). 624 Due to the high number of variables and countries considered by our study, the only way to 625 estimate the overall uncertainty would be through a Monte Carlo approach, as proposed for British Unfortunately, much of this information is often not available or simply does not exist. The yield 628 curves used in CBM are based on field observations, and thus some impacts of environmental 629 changes are represented in the model. However, many of these curves are based on plot 630 measurements over the past decades, and we therefore cannot make any assumptions about how 631 representative the existing yield curves will be for future (2030)  Slovenia and Spain; the average NPP estimated by this model is 17% higher than our estimate but 642 it is also combined with a higher contribution of Rh, equal on average to 72% in EFISCEN against 643 64% in CBM. ORCHIDEE, a process-oriented model, and BIOME-BGC a climate-based 644 ecosystem model, generally reported a higher NPP than CBM: on average +8% and +16%, for 645 BIOME-BGC and ORCHIDEE, respectively. JULES, i.e. a process-based surface exchange 646 scheme similar to ORCHIDEE, generally estimated a lower NPP than CBM (on average -24% at 647 EU level). Many reasons, such as the use of different data sources, different assumptions on the 648 forest area, the effect of the main natural disturbances (generally not considered by EFISCEN) and 649 silvicultural practices (generally neglected by climate-based ecosystem models) may explain these 650 differences. Looking to the standard deviation estimated by these data series, however, the average 651 NPP estimated by these models (5.54 ±1.19 Mg C ha -1 yr -1 ) is not statistically different from the 652 average value estimated by CBM (5.15±1.42 Mg C ha -1 yr -1 ). 653 Further studies will focus on a specific assessment of these uncertainties, but, in the meantime, to 654 overcome these limitations, we successfully validated our results at the country (for Lithuania) and  2030 is similar (+3%), lower (-17%) and higher (+23%), when assuming a constant, increasing 669 and decreasing scenario for both harvest and afforestation rates. In this study we did not quantify 670 the avoided emissions from the use of wood products and fire wood, and changes in NSE may not 671 be indicative of the overall changes in GHG balance resulting from changes in harvest rates. 672 Increased harvest rates will reduce NSE but provide more wood products that can be used to 673 substitute other emissions-intensive materials and fossil fuels. 674 For the forest area existing in 1990 (i.e., the FM area), we show a decline in the C sink, assuming 675 a constant harvest scenario, due to increasing releases from decomposition (Rh +13%) as DOM 676 pools increase with increasing biomass stocks. This confirms the results of earlier studies, 677 suggesting some signs of C sink saturation in European forest biomass (Nabuurs et al. 2013). This 678 result, however, should be combined with further analysis, accounting for the ongoing 679 environmental changes, which could have impacts on NPP and Rh that are not represented in the 680 inventory-based model used in this analysis (Kurz et al. 2013). The non-proportional effect of 681 different harvest scenarios on the 2030 C sink of the FM area suggests that the overall growth of 682 the European forests is slightly decreasing, and by increasing the harvest demand by 20%, we are 683 approaching the maximum harvest potential of the pre-1990 forest area. 684 country level, we highlighted some statistical differences, suggesting that the relationship between 687 biomass stock and NPP cannot be assumed constant for all EU countries. Specific forest 688 conditions, such as the harvest rate, the age structure and forest composition, may affect the 689 country-specific evolution of biomass, dead organic matter and soil stocks.   scenario (i.e., constant harvest and AR rate); (iii) the increasing scenario (i.e., +20% amount of harvest and AR rate compared to the average historical harvest and AR rate); (iv) the decreasing scenario (i.e., -20% amount of harvest and AR rate compared to the average historical harvest and AR rate). For each scenario, the fluxes were further distinguished between (all values in Tg C yr -1 ): (NPP) the Net Primary Production contributed by the FM area (including deforestation), AR, and total (FM+AR); (A) the total losses due to natural disturbances and deforestation (i.e., direct emissions to the atmosphere); (B) the fluxes of C from the living biomass to DOM pools (i.e., internal fluxes for the forest ecosystem), further distinguished between fluxes due to self-thinnings and to fellings (i.e., the harvest residues, equal to the difference between fellings and harvest removals); (C) the total fluxes of C due to fellings and the harvest C removals provided by the  Exchange (NSE) reported by the black labels (in Mg C ha -1 yr -1 ) and proportional to the radius of the external circle. years with a distance greater than 3 interquartile ranges from the median (SAS Institute Inc., 1990)) due to extreme events such as exceptional disturbances. Plots B and D report, for each country, the slope (τ±95% confidence interval) of the linear regression model (y = a + τx) applied to the previous values for each country (reported on the x axis). On plots A and C, we also highlighted the regression model estimated, at EU level, including all the countries, with the corresponding equation and coefficient of regression (R 2 ).