Forests globally provide ecosystem services including provision of timber, fuel, and water; regulation of local climate and hydrology; carbon sequestration; support of biodiversity; and recreation (Bonan, 2008; Mori et al., 2017). The effects of climate change on forest productivity and biodiversity may be predicted to be negative due to increased evapotranspiration and reduced rainfall in many forested areas; an increase in extreme events like droughts, wildfires, storms, and insect attacks; and local or regional extinctions of plant and animal species (Easterling et al., 2000; Seidl et al., 2011; Anderegg et al., 2013; Urban, 2015). On the other hand, productivity may increase due to the fertilising effect of increased nitrogen deposition and higher atmospheric CO2 levels (Zaehle and Dalmonech, 2011; Luyssaert et al., 2008) and shifts in tree species composition and longer growing seasons at high latitudes caused by higher temperatures (Sitch et al., 2015; Morin et al., 2018).

Forests make up the largest portion of the current land carbon sink and are estimated to have absorbed 20 %–50 % of CO2 emitted by fossil fuel combustion and industry during the first decade of this century (Pan et al., 2011; Le Quéré et al., 2018; Pugh et al., 2019). The suggested basis for this carbon uptake is the recent history of the drivers increasing productivity mentioned above, especially increased CO2, and the recovery of carbon pools in regrowth forests (forests regrowing after natural or anthropogenic stand-destroying disturbances; Pugh et al., 2019). The size of the forest carbon sink has been estimated by using bookkeeping methods (Pan et al., 2011; Houghton et al., 2012) and global vegetation models (Luyssaert et al., 2008; Shevliakova et al., 2009; Pugh et al., 2019), but this sink is associated with relatively large uncertainties, resulting in differing estimates using different approaches and models. Key uncertainties include the magnitude of CO2 fertilisation – which may be limited by soil availability of nutrients such as N and P (Zaehle and Dalmonech, 2011; Jiang et al., 2020) – and the extent of shifting cultivation in the tropics (Heinimann et al., 2017). While the net atmosphere-to-land flux (FL) is relatively well constrained by atmospheric measurements, large uncertainties in the net land-use and land-cover flux (FLULCC) make the size of the residual (land) sink (FRL) itself uncertain (FL=FRL-FLULCC) (Arneth et al., 2017).

Forests cover 33 % of Europe's land area (Forest Europe, 2015) and store approximately 13 Pg C in vegetation and 28 Pg C in soils (Pan et al., 2011). The carbon sink of European forests in 2000–2007 has been estimated at 0.27 Pg C yr-1 or about 12 % of the global carbon sink of established forests (Pan et al., 2011). Europe has been identified as a region where regrowth forests dominate carbon sequestration (Pugh et al., 2019) and has a history of thousands of years of human impact on forest structure and species composition (Perlin, 2005). Forest management practices of the past few hundred years are relatively well documented (McGrath et al., 2015). Depending on the region, different management strategies are applied (Cardellini et al., 2018). The preponderance of young trees and the removal of wood in managed forests influence carbon stocks and fluxes, e.g. by increasing productivity and reducing self-thinning, age-related mortality, and litter production compared to pristine forests (Zaehle et al., 2006). In addition to the effects on radiative forcing by atmospheric CO2, forest management influences local climate by changing albedo, evapotranspiration, and surface roughness (Luyssaert et al., 2014).

Dynamic vegetation models (DVMs) provide a potential framework for predicting the combined effects of climate and forest management scenarios on forest ecosystem structure and carbon balance. Based on such information, the potential of forest landscapes to contribute to climate change mitigation by maintaining or enhancing carbon sinks and to climate adaptation through sustained production of forest products and other ecosystem services in the face of climate change can be assessed. Applications of DVMs to represent climate responses of potential natural vegetation (PNV) have been shown in the past, for example as a basis for nature conservation planning (Hickler et al., 2012). Human management of land, including cropland, pasture, and managed forest, has been introduced in a number of global DVMs (Bondeau et al., 2007; Bellassen et al., 2010; Lindeskog et al., 2013; Arneth et al., 2017). Key elements required to represent managed forests in a DVM framework include the ability to initialise a simulation with historical land use; to represent age and size structure of forests stands and their change over time; to account for tree species composition; and to apply silvicultural treatments that modify stand composition and structure like planting, thinning, and harvesting.

LPJ-GUESS (Smith et al., 2001, 2014) is a second-generation DVM tailored for regional- and global-scale applications. It is one of few globally applicable DVMs that incorporate a detailed representation of forest ecosystem composition and stand dynamics, suitable for the implementation of a forest management scheme. It captures the distribution of European PNV at species level and can make projections of vegetation shifts under future climate scenarios (Hickler et al., 2012). The model has been shown to represent stand-level vegetation growth and succession successfully (Smith et al., 2014). It has been used to estimate forest vulnerability to climate change (Seiler et al., 2015) and carbon mitigation potential of regrowth forests and forests under alternative management scenarios (Pugh et al., 2019; Krause et al., 2020). Earlier versions of LPJ-GUESS have been modified to enable analysis of clear-cut forest management and the effects of wind damage and insect outbreaks (Lagergren et al., 2012; Jönsson et al., 2012). In this study, we describe the implementation of expanded forest management capabilities including even-age and clear-cut as well as uneven-age and continuous-cover management in LPJ-GUESS v4.0. In addition to detailed carbon and water cycle processes, this version of the model incorporates a dynamic nitrogen cycle and nitrogen limitation on plant productivity (Smith et al., 2014). In this way, forest management in LPJ-GUESS is for the first time fully integrated in a model version capable of simulating a landscape containing a mosaic of land-cover types like PNV, cropland, pasture, and peatland and with a sophisticated land-use and land-cover change functionality. Model alternatives for forest stand initialisation (land-use history and species and age distribution) and silvicultural management (detailed and automated harvest strategies) are presented in detail. Simulations using different forest management alternatives are evaluated against observations of standing volume and harvest for even-aged monospecific European beech and Norway spruce stands in central Europe. Using an automated thinning and clear-cutting approach for European forests, we compare modelled carbon stocks and fluxes with observational data and explore the dynamic behaviour of the model under changing climate forcing.

LPJ-GUESS representations of unmanaged forest have previously been shown to compare favourably with observed forest vegetation succession, growth, stand structure, biomass, and regrowth timescales (Smith et al., 2001, 2014; Pugh et al., 2019), and land-use and land-cover change (LULCC) functionality has been included in the model since version 4.0 (Lindeskog et al., 2013). In a recent global study that used the model to analyse the carbon stocks of old-growth and regrowth forests (modelled as primary and secondary forest stands, respectively), without applying wood harvest (Pugh et al., 2019), the total forest carbon sink was found to be about 50 % of values reported by Pan et al. (2011) based on upscaled inventory data. Disregarding wood harvest has been identified as causing underestimation of carbon sinks in vegetation models (Zaehle et al., 2006; Ciais et al., 2008). In an effort to improve the ability to simulate carbon pools and fluxes on managed land, we introduced new forest management options into LPJ-GUESS v4.0 and provide a comprehensive description of forest initialisation and wood harvest alternatives. The initialisation and harvest alternatives in the model are tailored to enable available forest inventory data and harvest information to be used to initialise and guide simulations. Ideally, both age and species structure, as well as land-use history and current wood harvest strategy, should be taken into account, but this is not always possible for simulations with a large spatial extent because of limited data availability. To demonstrate a possible workaround, we used an automated thinning and clear-cutting alternative to represent European regrowth forests, initialised on the basis of inventory-based age and species data but without wood harvest or LULCC data input. In simulations of central European beech and spruce stands, the automated thinning method was shown to result in similar modelled standing volume but often in a higher carbon sink compared to a more detailed harvest scheme based on reported harvest intensities (Fig. 7). The harvested volume was generally substantially higher in the automated thinning simulations, as the optimum harvested volume required to completely avoid self-thinning may not be realised in real managed forest stands. Ideally, automated thinning should be just enough to avoid self-thinning mortality in the model, so the biomass should not be severely reduced, but in old beech stands self-thinning is very low in the model (Fig. C1), and thus in these stands both detailed and automated harvests result in a relatively large reduction in biomass compared to unharvested stands (Fig. 7).

The modelled mean vegetation carbon density in European forests in 2000–2010 is close to observations from several published sources (Pan et al., 2011; Liu et al., 2015; Forest Europe, 2015). Including thinning in the simulation has a rather small impact on vegetation carbon (<5 %), but after clear-cutting starts in regrowth stands after 2010, simulations with and without harvest in regrowth stands diverge strongly (Fig. E7). In addition, the modelled mean European growing stock is close to observations. Modelled carbon sink density (=-NEE) for European forests in a simulation without thinning in the present study is about 46 % of the 2000–2007 value reported by Pan et al. (2011). This is similar to the global carbon sink predicted by a simulation with a similar setup without thinning, which is 49 % of the global value from the Pan et al. (2011) study. The difference in modelled carbon sink in 2001–2010 between old-growth forest (0.04 kg C m-2 yr-1) and regrowth forest without thinning (0.085 kg C m-2 yr-1) is similar to the difference reported for global old-growth and regrowth forests by Pugh et al. (2019). Adding thinning to the European regrowth forest setup increases the carbon sink by 38 % for the total forest area and by 46 % for the regrowth forest area, reaching 64 % and 82 %, respectively, of the Pan et al. (2011) value. Thinning reduces natural mortality due to relaxed competition between trees, and since a large part of harvested biomass is removed from forest stands, litter input to the soil and the resulting heterotrophic respiration are also reduced (Figs. E6–E7), increasing the carbon sink.

Details in the simplified European setup might explain the remainder of the “missing” carbon sink relative to reported values. One potential cause is that old-growth (>140 year) forests in this study are represented by unmanaged PNV (with a low carbon sink; see Table 6), as in Pugh et al. (2019), missing effects of land-use history in Europe but preferred by us to the alternative of introducing arbitrary assumptions of age structure. Furthermore, the GFAD >140 year forest type data contain artefacts manifested in the BE distribution. Including a basic extensive wood harvest method in old-growth forests increased the total carbon sink by only 4 %, resulting in a value of 66 % of the Pan et al. (2011) value. Wildfires also contribute to a lower carbon sink in modelled PNV. A further likely cause of the discrepancy between the modelled and reported carbon balance is that secondary forests are created from PNV stands without taking land-use history into account. Reforestation of cropland, which generally has a much lower soil carbon content than forests in Europe (Guo and Gifford, 2002), has a higher carbon storage potential than regrowth after clearing of existing forests. In addition, soils of existing European forests have been depleted of carbon historically because of higher harvest rates, fuel-wood collection, and litter raking (Ciais et al., 2008; McGrath et al., 2015). Higher initial soil carbon pools will increase the release of CO2 in regrowth forests, especially under rising temperatures. Alternative methods to initialise secondary forests (fate of cleared wood, land-use history) have large implications for simulated carbon pools and fluxes as seen in the example Swedish site in this study, e.g. a mean carbon sink over 150 years spanning from 0.078 to 0.188 kg C m-2 yr-1 (Fig. 2). This has also been shown at the global scale (Pugh et al., 2019). The high value of modelled European soil carbon density in 2000–2010 (34 %–80 % higher than reported values) supports the possibility that the lack of consideration of LULCC history is a main source of the missing carbon sink in this study. The similarity of the modelled mean NAI of European forests in a simulation with thinning to observed values (a 100 % increase compared to a simulation without thinning) also suggests that the missing carbon sink component could be found in heterotrophic respiration and not in vegetation productivity.

The automated thinning and clear-cutting modelling strategy applied in the model in the present study is intended as an example for demonstrating the new forest management capabilities and an improvement on the age structure setup of Pugh et al. (2019) and does not include all available possibilities in the model. In addition to the shortcomings in the setup already noted concerning land-use history, many central European forests are managed by continuous wood harvest and not by clear-cutting and also consist of species mixes (Pretzsch et al., 2021). Estimating the effect of such different wood harvest strategies and monoculture or mixed-species alternatives on carbon stocks and fluxes is now possible and will be done in further studies. The self-thinning and tree-density-based harvest method is less successful in the northernmost and southernmost parts of Europe, where productivity is strongly limited by temperature and precipitation, respectively, and the self-thinning relationship between biomass and tree density in the model is weaker. The low simulated productivity of forests in the Mediterranean points to the need for a review of the parameterisation of tree species to reflect Mediterranean managed forests or the introduction of tree species that are not currently represented in the model (Fig. E8). While the model shows good skills when reproducing reported mean values for Europe's vegetation carbon and productivity, the correlation between modelled results and observations for the individual countries shows a large spread with no simple pattern for the deviations (Figs. E1–E5). However, it is obvious that modelled thinning intensities for countries in the Balkans, except Albania and Greece, are higher than the corresponding reported total harvest intensities. These countries also show a poorer fit to observed NAI values in a simulation with thinning compared to a simulation without thinning. In any case, including thinning in simulations improves the fit to observed national NAI values in most other countries.

Our simulation results using LPJ-GUESS exhibit similarity with results from the ORCHIDEE DVM, which was applied with the same automated thinning method at a central European site (Bellassen et al., 2010). The ORCHIDEE simulation with automated thinning, compared to a simulation without thinning, gave a similar vegetation reduction (7 %) and thinning fraction (0.55), reduced heterotrophic respiration (ca. 20 %), and a carbon sink increase (67 %). The forest NPP reduction over time in ORCHIDEE simulations (ca. 10 %) is also seen in the average value for unharvested regrowth forests in European simulations with LPJ-GUESS (Fig. E7b). The decline of NPP directly after thinnings in ORCHIDEE is not simulated by LPJ-GUESS, but both models display a short-lived increase in heterotrophic respiration after thinnings (not shown). The recovery time after a clear-cut (when the stand turns into a carbon sink) is 6 years in the example southern Swedish site with a standard harvest removal, but it is 18 years if the harvested biomass is left on site (Fig. 2). This is similar to the ORCHIDEE results with a stand recovery time of 10–20 years after a clear-cut. A similar recovery time after clear-cutting, 7–11 years, has been diagnosed based on CO2 flux measurements in Sweden (Lindroth et al., 2009).

Responses of soil carbon and nitrogen cycling to harvest and fertilisation can be complex and qualitatively different in clear-cut and continuous-harvest systems (Parolari and Porporato, 2016). The coupled carbon–nitrogen cycling in LPJ-GUESS (Smith et al., 2014) should enable the investigation of the effect of different management practices on forest productivity and sustainability at both stand and regional scale in future studies. Nitrogen depletion of the soil in previous land-use history reduces forest productivity and causes a shift in species succession in the model (Fig. 2c). At the European scale, removing a smaller fraction of residues (0 % of leaves rather than 30 %) makes a small positive impact on productivity (0.1 %; see Sect. 3.4). However, since many European forests receive large amounts of atmospheric nitrogen deposition, other nutrients such as Ca, Mg, K, and P may be more important for limiting productivity, and acidification of the soil by N and S deposition may further decrease the availability of these nutrients (Sverdrup et al., 2006). Ca is especially close to or below the limit of sustainability in current forest management systems in southern Sweden (Sverdrup et al., 2006). Thus, ongoing development of limitation and cycling of additional nutrient species into LPJ-GUESS may be beneficial for capturing the full effects of different harvest regimes. Also relevant to achieving a better model of nutrient uptake is an improved representation of the soil profile.

While the mean productivity of European forests is captured well by the model (Table 4), and mean productivity of forests in individual European countries is captured reasonably well (Figs. 10, E4), the inability to reproduce observed productivity levels in high-productivity beech and spruce stands in Germany (Fig. 7a) highlights the need for allowing a wider range of productivities. The lack of certain physiological processes in the model, e.g. hardening and dehardening (Bergh et al., 1998), could explain why productivities along the whole temperature gradient in European forests cannot be fully represented in the model. Model tuning that aims for correct mean values of, for example, biomass and carbon fluxes over large geographic areas compensates for an overestimation of productivity in northern Europe by lowering average productivity along the whole temperature gradient. This could partially explain, for example, why the productivity of some southern German sites is underestimated, while average productivity for Germany as a whole is in line with inventory data. Additionally, the selected individual German Norway spruce and European beech sites in this study were generally of above-average site quality and are not fully representative of German forests, which includes forests of other tree species, especially Scots pine (Pinus sylvestris) and oak species (Quercus robur/Quercus petraea), on lower site quality sites. This is likewise in line with the smaller gap between modelled and observed growing stock (ca. 20 %, Fig. E3) seen at country level, compared to individual spruce and beech sites in Germany (Fig. 7a).

The emergent competition between PFTs with similar shade-tolerance values in the model, e.g. beech and spruce, can deviate from actual dynamics, as seen in the poor performance of spruce compared to beech in a succession at the example site in southern Sweden (Fig. 5).

The management systems covered by the new forest management functionality in LPJ-GUESS include the most important features required for the improvement of modelling carbon pools and fluxes and the development of forest stands under future climates, but a few important additions will be desirable to include in the future. These include automated continuous wood harvesting and coppice management. For a good representation of coppicing, the model should also be improved to include plant carbohydrate storage. For better representations of European forests, land-use history, including litter raking, should be included to generate more realistic soil carbon pools by adapting functionality already available in the model.