Impacts of climate Change, Weather Extremes and Alternative Strategies in Managed Forests

ABSTRACT The growth rate of most tree species in boreal forests will increase with changing climate. This increase is counterbalanced by an increased risk of damage due to extreme weather events. It is believed that the risk of storm damage will increase over time, especially if forests continue to be managed as they are today. In this study, a new landscape-level hybrid forest growth model 3PG-Heureka was developed and simulations were performed to predict the damage caused by storm events in Kronoberg county, over a period of 91 years (2010–2100) with different alternative management regimes under various climatic scenarios (historic, RCP4.5 and RCP8.5). The results indicate that damage caused by storm events could drastically reduce the annual volume increment and annual net revenue obtained from forest landscapes if current forest management regimes are used. These problems can be reduced by adopting alternative management strategies involving avoiding thinning, shorter rotation periods and planting alternative tree species. Alternative management strategies could potentially improve annual volume increments and net revenue obtained while reducing storm-felling. Planting Scots pine instead of Norway spruce across the landscape to minimize storm damage is predicted to be less effective than reducing rotation periods.


Introduction
It has been predicted that the increases in temperature and atmospheric CO 2 concentration will promote growth and thus increase biomass production in Swedish forests (Kirilenko and Sedjo 2007;Bergh et al. 2010;Poudel et al. 2011). This in turn is predicted to increase harvested volumes and permit the use of shorter rotations in managed forests (Briceno-Elizondo et al. 2006;Kirilenko and Sedjo 2007;Bergh et al. 2010). Recent studies have suggested that the growth rates currently achieved in Swedish forests are well below their production potential because intensive management methods like using genetically improved seedlings for planting and intensive fertilization are rarely applied (Rosvall 2007;Nilsson et al. 2011). Climate change could further enhance this currently untapped production potential of Sweden's boreal forests (Bergh et al. 2005;Subramanian 2010;Poudel et al. 2012).
In addition to the increase in the productive potential of Sweden's forests, future climate change is expected to increase the frequency of adverse effects including extreme weather events such as frosts and storms (Blennow and Olofsson 2008;Hanewinkel et al. 2011;Jungqvist et al. 2014). Storms are a major problem for growing stock in the forests across Europe, including Sweden (Hanewinkel et al. 2011). Major storm events in central Europe in the last three decades have caused vast damage to growing stock in the forests (Blennow and Olofsson 2008;Hanewinkel et al. 2011;Wallentin and Nilsson 2014). Recent effects of climate change such as increased winter precipitation and reduced frequency of sub-zero winter temperatures, which allows soil to remain unfrozen during winter and thus weaken tree anchorage, can increase storm damage in the future (Hanewinkel et al. 2011).
There have been several changes in management practices that have increased the likelihood of future storms causing significant damage to growing stock. These include increasing the amount of conifer monoculture (Drössler et al. 2014) and standing volume in the forests (Hanewinkel et al. 2011). Changes in current management practices may therefore be needed to effectively utilize the productive capacity of forest landscape, at the same time increasing its resilience against the adverse impact of extreme weather events (Lucash et al. 2017). Previous studies on the impacts of climate change and its risk factors at the stand level have shown that storm damage can be mitigated by using adaptation strategies such as avoiding thinning and reducing rotation lengths, and by promoting the growth of stands with alternative tree species rather than Norway spruce monocultures (Keskitalo et al. 2011;Wallentin and Nilsson 2014;Subramanian et al. 2016a). In addition to protecting against storm damage, cultivating alternative species (as opposed to Norway spruce monocultures) improves biodiversity, increases resistance to root rot (Heterobasidion annosum [Fr.] Bref.) and bark beetle (Ips typographus L.) infestations (Subramanian et al. 2016a), and may create opportunities to use new and more productive indigenous or exotic tree species in commercial forestry (Bolte et al. 2009;Mason et al. 2012;Meason and Mason 2014).
Simulation models are used to predict the potential effects of future climate change on the growth and management of forests. Three main kinds of simulation models are used in forestry: (a) empirical models, (b) process-based models, and (c) hybrid models; each kind has its own advantages and limitations (Korzukhin et al. 1996;Landsberg and Coops 1999;Mäkelä et al. 2000;Peng et al. 2002;Landsberg and Sands 2010;Maroschek et al. 2015).
Empirically based stand level growth functions (Ekö 1985;Elfving and Nyström 2010) and individual tree growth functions (Söderberg 1986;Elfving and Nyström 2010) have been developed using Swedish National Forest Inventory (NFI) data and can be applied to all major tree species found in Sweden (Fahlvik et al. 2014). Empirically based Decision Support Systems (DSS) such as the Heureka system and its predecessor Hugin are widely used in the Swedish forestry sector for long-term forest planning (Hägglund 1981;Elfving and Nyström 2010;Wikström et al. 2011;Eriksson et al. 2015). These growth simulators are based on even-aged monocultures that are traditionally managed, and they can be used to forecast the growth and yield of forests in the near future, when the impact of climate change on forests is minimal. However, the reliability of their predictions for longterm forest planning is questionable. The problem with long-term forecasting using empirically based models is that the accuracy of the predictions decreases as the length of the period over which the predictions are made increases (Kangas 1997). Therefore, given the expected changes in environmental conditions, simulations based on these models cannot be expected to reliably predict growth over the coming decades.
Process-based models (PBM) simulate growth as a function of interacting eco-physiological processes (e.g., photosynthesis, respiration, decomposition, soil-water balance, and nutrient cycling) that are in turn treated as primary effects of environmental factors such as light, temperature, atmospheric CO 2 levels, and the soil water content (Landsberg and Sands 2010). For this reason, PBMs such as 3-PG (Landsberg and Waring 1997;Landsberg and Sands 2010) and BIOMASS (McMurtrie and Wolf 1983) are used in Sweden to predict the states of forests and the way they change at the stand level in response to climatic change (Bergh et al. 1998(Bergh et al. , 2005Subramanian 2010;Getahun 2013). However, existing PBMs have a number of important drawbacks. In particular, their carbon allocation equations, mortality equations, management functions and regeneration equations are associated with high levels of uncertainty (Mäkelä et al. 2000). The mortality functions used in current PBMs are often based on the −3/2 power law of self-thinning, which relates the stand density to the biomass of individual stems (Landsberg and Sands 2010). However, this law was developed for even-aged stands and previous studies have established that the exponent of −3/2 is not universally valid; stands having different ages, species, site qualities, and/or initial densities may require different exponents (Weller 1987;Zeide 1987). Another important drawback is that regeneration functions are not usually included in stand level PBMs (Mäkelä et al. 2000) even though it is important to consider regeneration in simulation models for longterm planning. A third drawback derives from the carbon allocation models used in PBMs. Process-based carbon balance models are used to describe the assimilation of carbon and its allocation to different organizational levels in the stand such as the foliage, stem, and roots (Mäkelä et al. 2000). In a stand level PBM such as 3-PG, carbon allocation functions are implemented as independent sub-models, but the physiological mechanisms underpinning carbon allocation are not sufficiently well understood for these sub-models to be relied on in practical forestry (Landsberg and Sands 2010). Finally, the description of management operations such as cleaning, pre-commercial thinning (PCT), commercial thinning and final felling in PBMs is limited in many respects. After a thinning or PCT in a forest stand, the canopy's development is enhanced because the allocation of carbon to the foliage and branches is increased, leading to an increase in net photosynthesis of the remaining stand (Wallentin and Nilsson 2011). The current assumption in 3-PG is that the only modification that occurs after a thinning or PCT operation in a forest stand is an increases radiation interception and transpiration of the remaining trees (Landsberg and Sands 2010). This means that the 3-PG model does not account for changes in light use efficiency (LUE) or the increased allocation of biomass to foliage and branches after a thinning operation.
As demonstrated above, there is considerable uncertainty in long-term forest planning when using either PBMs or empirically based models. Consequently, there is a growing demand for growth models that can be applied to diverse forest structures in a given operational level and can describe the potential effects of future changes in the global environment, such as changes in the climate and forest management practice (Von Teuffel et al. 2006). In general the weak points of PBMs are strong points of empirical models and vice versa (Peng et al. 2002).
Hybrid modelling thus seeks to combine the strengths of the two approaches such that each one compensates for the other's weaknesses (Kimmins et al. 1999). Combining the growth equations of empirical models and PBMs in this way could reduce the uncertainties associated with long-term forest planning. The major limitations encountered when using a DSS such as Heureka for long-term forest planning could be overcome by implementing growth functions from PBMs that can more reliably model growth under changing climatic conditions. The concept of hybrid forest modelling has been investigated previously (Battaglia et al. 1999;Kimmins et al. 1999;Peng et al. 2002;Pérez-Cruzado et al. 2011;Maroschek et al. 2015), and attempts have been made to adapt the Heureka system's growth simulator to handle the effects of climate change by indirectly using an approximation model (Eriksson et al. 2015). In practice, the implementation of this climate response model involved modifying the relationship between tree size and tree age that is assumed in Heureka, and also making changes to the Heureka system's handling of the site index, vegetation type index and temperature sum (Eriksson et al. 2015).
The generic hybrid model 3PG-Heureka developed by Subramanian (2016) combines growth simulators from the process-based model 3-PG, management and mortality functions of the Heureka-Regwise, one of the models in the Heureka Decision Support System (DSS) (Elfving and Nyström 2010;Wikström et al. 2011;Fahlvik et al. 2014). Forest growth is simulated using a monthly time step and management activities are implemented once every five years.
The aim of this study was to (i) investigate the effects of climate change and extreme weather events on forest production and economy at a landscape level, and (ii) discuss the potential for using alternative management programmes as an adaptive strategy to reduce damage caused by extreme weather events. Alternative programmes considered include a modified thinning programme, shorter rotations, and planting alternative tree species.

Study area
Kronoberg county (56.4°N to 57.2°N, 13.4°E to 15.8°E) is located in southern Sweden (Figure 1(a)). The total annual precipitation in the area is approximately 750 mm and the monthly mean temperature is around 16.5°C and −0.7°C respectively during summer and winter (SMHI 2016). The total productive forest area in Kronoberg county is around 0.65 million ha and the average standing volume in the landscape is around 142 m 3 ha −1 (Skogsstyrelsen 2014). The mean site index (dominant height [m] at a total age of 100 years) for the whole landscape is around 27 m and the dominant ground vegetation type found in the landscape is blueberry (Vaccinium spp.) and grass (Elfving and Kiviste 1997;Fridman et al. 2014). The forest in the region is dominated by conifer stands and the major tree species is Norway spruce. Other tree species present include Scots pine, silver birch (Betula pendula Roth.) and downy birch (Betula pubescens Ehrh.), which are both henceforth referred to simply as 'birch', European beech (Fagus sylvatica L.) and oak (Quercus spp.).

The 3PG-Heureka model
The hybrid model 3PG-Heureka includes elements of the simplified 3-PG process-based carbon balance model (Landsberg et al. 2005;Landsberg and Sands 2010) and the empirically based forest management model Heureka-Regwise (Wikström et al. 2011;Fahlvik et al. 2014;Eriksson et al. 2015). The 3-PG model component is designed to predict forest growth based on the total stand biomass increment, while the Heureka-Regwise component is designed to predict management activities and mortality. This model has seven major sub-models. A brief explanation of the basic concepts, functions and structure of the submodels is provided along with supplementary materials section S1. The 3PG-Heureka model can be used for long-term forest management planning under changing climatic conditions. The minimum time-resolution for simulations with the model is five years.

Data inputs
Three kinds of input data are required for simulations using the 3PG-Heureka model: • Climate data: monthly averages of daily maximum temperature (T max , K), and daily minimum temperature (T min , K), as well as the total monthly rainfall (ppt, mm), total monthly incoming solar radiation (R, Wm −2 ) and number of frost days per month (n frostdays , days month −1 ). • Site-specific data: input data required for site description including the site's coordinates (latitude and longitude), altitude (m), site index (m), distance from coast (km), dominant tree species, soil moisture content (dry, moist, mesic, mesicmoist or wet), maximum available soil water, soil texture and type of ground vegetation present. • Initial and time series data: inputs required to define the initial conditions of the stand, including diameter at breast height (DBH, mm), the species identity of individual trees and the time series data including DBH increment in the last five years (d 5 , 0.1 mm).

Datasets used
Initial tree data The initial tree data used for the 3PG-Heureka simulations reported in this work were measurements conducted at the 657 permanent National Forest Inventory (NFI) sample plots in Kronoberg county between the years 2008 and 2012 (Figure 1(b)). The NFI data was obtained from the Department of Forest Resource Management, Swedish University of Agricultural Sciences, Umeå. Each circular NFI plot has an area of 0.0314 ha (corresponding to a radius of 10 m). The DBH and diameter increment in the last five years (d 5, 0.1 mm) of each individual stem within the NFI plots were calculated by subtracting the DBH measured during the previous NFI (2003)(2004)(2005)(2006)(2007) from that for the most recent one (2008)(2009)(2010)(2011)(2012), during which the species identity of each tree in each plot was recorded and each stem was assigned a unique stem ID. Site-specific parameters such as coordinates, distance from coast, altitude, soil moisture, ground vegetation type, dominant tree species, plot area and tree age were also estimated or measured during the plot inventory (Fridman et al. 2014).

Climate data and scenarios
The climate data used in this study were obtained from Regional Climate Model Simulations developed by researchers at the Rossby Centre, Swedish Meteorological and Hydrological Institute (SMHI 2016). Three climate scenarios, two different future climate scenarios and one historic climate scenario were considered. Two different future climate scenarios adopted from the Intergovernmental Panel on Climate Change (IPCC) fifth assessment report (AR5) were considered, namely the RCP8.5 and RCP4.5 Representative Concentration Pathways (Moss et al. 2010;IPCC 2013). RCP8.5 is a worst-case scenario that predicts an increase in radiative forcing of more than 8.5 Wm −2 during the twenty-first century (Moss et al. 2010). Such an increase is projected to cause a 2.6-4.8°C rise in global mean surface temperature and a 0.45-0.82 m rise in the global mean sea level in this scenario during 2081-2100 relative to the period 1986-2005 (IPCC 2013). The atmospheric CO 2 concentration in this scenario will be around 936 ppm at the end of this century and the radiative forcing will rise without stabilization (IPCC 2013).
RCP4.5 is a more optimistic scenario that predicts that the increase in radiative forcing will stabilize at 4.5 Wm −2 towards the end of this century, without ever exceeding this value (Thomson et al. 2011). It is predicted to cause a 1.1-2 6°C rise in global mean surface temperature and a 0.32-0.63 m rise in the global mean sea level during 2081-2100, relative to 1986(IPCC 2013. The CO 2 concentration in this scenario will be around 538 ppm at the end of this century (IPCC 2013).
RCP scenarios simulated using a global ensemble that includes four different Global Circulation Models (GCMs) from the CMIP5 project (Taylor et al. 2012) were considered in this study. The four GCMs were CNRM-CM5 (Voldoire et al. 2013), EC-Earth (Hazeleger et al. 2010), IPSL-CM5A-MR (Hourdin et al. 2013) and MPIESM-LR (Block and Mauritsen 2013). All the GCMs were regionalized for Kronoberg county using the Rossby Centre regional atmospheric model (RCA4; Kupiainen et al. 2013). The historic climate scenario was based on observed climate data that were converted into gridded format by interpolating using the Mesoscale Analysis (MESAN) model (Hāggmark et al. 2000). An illustration of projected changes in climatic variables in the study area under different climate scenarios is provided in the supplementary section S2.
The climate data used in this study were gridded with a spatial resolution of 12 × 12 km 2 for all climate variables. There were a total of 176 grid cells covering Kronoberg county (Figure 1(c)). The climate data for each grid cell were determined at the cell's central point, and each NFI plot was assumed to have identical climate data to the cell whose central point was closest. n frostdays was calculated from the daily values of T min . Any day when T min < 273.15 K (0°C) was considered to be a frost day. The total number of frost days per month was summed to determine n frostdays . n vdays was the number of days in a month when the average temperature ≥ 5°C (SMHI 2015). Monthly global mean CO 2 concentrations in parts per million (ppm) were obtained from the ESRL Global Monitoring Division (Dlugokency et al. 2015).
The future scenarios considered in this study cover the period between 2010 and 2100 (91 years). Historic climate data were available for the period between 1989 and 2010 (22 years). Therefore, in the historic climate scenario, the climate and atmospheric CO 2 data for this 22-year period were replicated over the entire simulation period (91 years).

Growth simulation using the 3PG-Heureka model
To begin with, in the 3PG-Heureka model the initial heights of individual trees were estimated from DBH and site characteristics using the functions developed by Söderberg (1992). Thereafter, the whole tree biomass (tons ha −1 ) of individual trees and of their foliage (tons ha −1 ) was calculated using biomass functions for individual trees (Figure 2). A detailed description of the calculation of whole tree biomass and foliage biomass of individual trees is provided in the supplementary section S3. The biomass production sub-model is designed to work at the stand level. The stand level variables were calculated by aggregating the data for individual trees in a particular plot. The whole tree biomass (tons ha −1 ) and foliage biomass (tons ha −1 ) of a particular tree species in a NFI plot were calculated from the individual tree biomass values. Mean values of plot variables such as the mean DBH (mm), mean height (dm) and mean age (years) were calculated from the corresponding individual tree variables.
As noted above, the 3PG-Heureka model has a fiveyear time step. However, its biomass production submodel is designed to predict monthly NPP values. Therefore, the biomass production sub-model is designed to run 60 monthly iterations (i.e., 5 years) per time step of the 3PG-Heureka model ( Figure 2). As trees grow, they accumulate biomass, which is allocated to particular tree components such as the foliage, stem and roots. Therefore, the foliage biomass (which determines the leaf area index, LAI) change as the trees grow. Consequently, the biomass production also changes over time. In this model, the stand level whole tree biomass is updated on a monthly basis. The foliage biomass, which is estimated from the stand level whole tree biomass is updated on a yearly basis (Figure 2). A brief description of the estimation of stand level foliage biomass from total stand biomass is provided in the supplementary section S4.
The stand LAI at the beginning of the simulation was calculated from the initial foliage biomass and used to compute the stand level monthly NPP for each plot (Equation S8-S12). The monthly NPP was then added to the initial stand level whole tree biomass at the end of each month. Then the stand level whole tree biomass at the end of the fifth year was computed ( Figure 2). The stand level basal area at the end of each five-year period was calculated from the simulated stand level whole tree biomass at the end of the period using a linear function. This function links the 3-PG part and Heureka part of the hybrid model. A detailed description of this function is provided along with the supplementary materials (S5).
In parallel, the five-year diameter growth of individual trees was estimated using empirical growth functions (Elfving and Nyström 2010). The DBH increments of individual trees were adjusted to ensure that they summed to the stand level basal area calculated with 3-PG growth simulator at the end of a fiveyear period. During NFI survey the individual tree heights were measured only for the sample trees (tallest tree in the plot). The five-year height increments for these sample trees were estimated from their measured tree heights (NFI data) and the height growth functions for top-height trees. Later the five-year height increments for all other individual trees were then estimated from their basal area using the basal area height relationship of the top-height trees in the plot (Elfving and Nyström 2010).
The basal area growth is calculated based on physiological processes and climate data in the 3PG-Heureka model. The adjustment of individual tree heights based on individual tree basal area functions was recalibrated to allow the tree height to be computed as a function of physiological processes and climate data. This was done by simulating the tree diameter in the 3PG-Heureka model without climate change for two periods while simulating the same with the effect of climate change for one period. If the individual tree diameter predicted with climate change at end of first period (d CC1 ) was greater than the diameter of the same tree without climate change at the end of second period (d CU2 ) then the height corresponding to d CU2 was considered as the new height for that particular tree. If the d CC1 was greater than the d of the tree in the first period as estimated without the effect of changing climate (d CU1 ) and less than d CU2 (Equation 1), or d CC1 was less than d CU1 , (Equation 2) interpolation functions were used to estimate the tree's height in the 3PG-Heureka model: where h = individual tree height (dm), d = individual tree diameter (cm), CC1 = variable simulated with the effect of changing climate at the end of first period, CU0 = variable simulated without climate effect at the beginning of the simulation (period 0), CU1 = variable simulated using without climate effect at the end of first period and CU2 = variable simulated without climate effect at the end of second period. Tree mortality is modelled using a two-step approach (Fridman and Ståhl 2001). In the first step, average mortality for each stand was estimated using a logistic function whose important variables include the basal area of larger trees, soil moisture, vegetation type and thinning history. In the second step, the probability of mortality of individual trees was modelled using species-specific logistic functions whose important variables include the basal area, mean DBH, DBH of individual trees and thinning history (Fridman and Ståhl 2001).

Parameterization of the 3PG-Heureka model
The 3-PG component of the 3PG-Heureka model requires some species-specific parameters, namely the maximum, minimum and optimum temperature (°C) for growth, stand longevity (years), maximum canopy conductance (m s −1 ) and site productivity modifier (Table 1). Values for all these parameters except for site productivity modifier were obtained from previous publications (Landsberg et al. 2005;Subramanian 2010;Potithep and Yasuoka 2011;Subramanian et al. 2015). An unmanaged forest landscape was simulated in the 3PG-Heureka for a period of 22 years, to avoid the need to account for management activities that would influence growth and thus affect the hybrid model's parameterization. A detailed description of parameterization of site productivity modifier and evaluation of the 3PG-Heureka's predictability is provided in the supplementary material sections S6 and S7 respectively.
The coefficients of the age modifier function (Equation S10; Table 1) were modified slightly to avoid overestimation of growth in very old stands.

Management regimes simulated
Four different management regimes were considered in this study: (i) Business as usual (BAU); (ii) Business as usual + Storm (BAU+Storm); (iii) Promoting alternative tree species (PAS); and (iv) Shorter rotation length (SR).

Business as usual regime (BAU)
BAU is the reference management regime, which is based on the current forest management programme for stands of Norway spruce, Scots pine and other broadleaved species in southern Sweden. The parameter values for this regime were drawn from a previous study on management in Kronoberg county (Eriksson et al. 2015).
In accordance with the prevailing practice, this management regime involves the adoption of an even-aged clear felling management system across the entire landscape. Norway spruce was extensively regenerated in Kronoberg county and birch was considered as the secondary species. Pre-commercial thinning (PCT) was done in all stands. A detailed description of planting, PCT settings, harvesting cost and timber prices used in the simulation are provided in the supplementary section (S8).

Business as usual + Storm regime (BAU+Storm)
The management regime implemented for the BAU +Storm was similar to that for the BAU. However, the management regime BAU+Storm also accounts for damage caused by major storm events using the windthrow model, which estimates the probability of stormfelling in a landscape on the basis of historical storm data (Lagergren et al. 2012;Eriksson et al. 2015). The historical data used in this model's development were gathered between 1948 and 2013, and include information on all storms for which the volume of timber damaged was recorded at the landscape level (Nilsson 2008). Nine storm events of varying wind speeds occurred in Kronoberg county during this period, in years 10, 15, 20 25, 30, 35, 55, 60 and 65 (with 1948 being year 1). The historical storm events occurring over the 66 years between 1948 and 2013 were repeated at the landscape level between 2010 and 2075 with same intensity and frequency and then repeated again between 2076 and 2100. Therefore, the 91-year simulation period included 13 predetermined storm events. A detailed description of the storm module is provided in the supplementary section (S9) for further reading.

Promoting alternative tree species regime (PAS)
The PAS regime was considered as an alternative management strategy. In this regime, Scots pine was extensively regenerated instead of Norway spruce in Kronoberg county. The regeneration settings of the 3PG-Heureka model were changed to promote regeneration of Scots pine.

Shorter rotation length regime (SR)
The SR regime was considered as an alternative management strategy. In this regime, the stands remained un-thinned and final felling was conducted at an earlier stage to avoid storm damage. The final felled stands were regenerated with genetically improved seedlings.
All other management activities such as PCT, commercial thinning, final felling and fertilization implemented in the PAS and SR regimes were carried out in a similar way to those used in the management regime BAU. Harvesting costs and timber prices were also similar to those used in the BAU regime (supplementary section S8). Damage due to major storm events was considered in the PAS and SR regimes and the parameters for the wind throw module were similar to those used in the management regime BAU+Storm (Supplementary section S9). The sum of the simulated output variables over the complete simulation period was calculated by summing up the corresponding values for each five-year period. The simulated output variables scaled up for the whole of Kronoberg county under various management regimes and climate scenarios were then plotted as functions of time (in years since the start of the simulated period). The simulated variables under the management regime BAU +Storm and various climate scenarios were compared to those for the management regime BAU to analyse the effects of climate change and storm events on the forest landscape's productivity and the economics of its management. The results obtained for the SR and PAS management regimes under the different climate scenarios were compared to the results for the management regime BAU+Storm to analyse the potential benefits and drawbacks of adopting alternative management strategies.

Comparison of forest variables under various management regimes and climate scenarios for the whole simulation period
The total growth in Kronoberg county during the whole simulation period was highest under the BAU management regime for all climate scenarios ( Table 2). The total harvested volume was highest under the BAU+Storm management regime for all climate scenarios. The total storm-felled volume was highest under the BAU+Storm management regime followed by the PAS regime. The total growth, total volume harvested and total storm-felled volume were lowest under the SR management regime for all climate scenarios. Total net revenue was highest under the BAU+Storm regime in the historic and RCP4.5 climate simulations. However, the total net revenue was highest under the BAU management regime in the RCP8.5 climate scenario. The total net revenue obtained under the SR management regime was greater than under the PAS regime for all climate scenarios. The accumulated standing volume over the full simulation period was highest under the BAU regime in all climate scenarios followed closely by the SR regime. There was a reduction in total standing volume in the BAU+Storm and PAS management regimes by the end of the simulation period.

Annual volume increment under different management regimes and climate scenarios in Kronoberg county
The annual volume increment in Kronoberg county increased over time in all management regimes and all climate scenarios (Figure 3). The annual volume increment during the early phase of the simulation period was 3.5 Mm 3 yr −1 for all the climate scenarios. The increase in annual volume increment was about the same in all climate scenarios until 2030. Afterwards the growth increased more in future climate scenarios (RCP4.5 and RCP8.5) when compared to the historic climate scenario. By the final phase of the simulation period (2080-2100), the annual volume increments for the historic, RCP4.5 and RCP8.5 scenarios had increased to 5.3 Mm 3 yr −1 , 5.7 Mm 3 yr −1 and 6.4 Mm 3 yr −1 , respectively in the BAU regime. The average annual volume increments in the management regime BAU were 5.8% and 10.8% higher in future climate scenarios RCP4.5 and RCP8.5 respectively during the time period 2050-2070 when compared to the historic climate scenario, while at the end of simulation period 2080-2100 the corresponding values were 8.6% and 21.0% respectively for RCP4.5 and RCP8.5 scenarios.
The annual volume increment was lower in the BAU +Storm, PAS and SR management regimes than that under the BAU regime because of the storm events included in the former cases. There was at least one storm event in every five-year period in the analysis. The annual volume increment plateaued after 2060 in all management regimes with storm while it continued to increase in future climate scenarios in the management regime BAU (Figure 3). The reduction in annual volume increment as a result of storm was more predominant in the management regimes BAU +Storm and PAS when compared to the management regime SR (Figure 3). The storm incidents in the years 2065 and 2095 dramatically reduced the corresponding annual volume increments and this reduction was more pronounced in the future climate scenarios (RCP4.5 and RCP8.5) than in the historic climate scenario. The storm event during the year 2065 reduced the annual volume increment in the BAU+Storm regime approximately by 7%, 9% and 12% for the historic, RCP4.5 and RCP8.5 scenarios respectively (Figure 3), whereas the corresponding values during the storm event during the year 2095 were 10% and 5% for future climate scenarios RCP4.5 and RCP8.5 respectively. No reduction in annual volume increment was found in the historic climate scenario during the storm event in 2095.

Annual net revenues obtained under different management regimes and climate scenarios in Kronoberg county
The annual net revenue obtained from Kronoberg county increased over time in all four climate scenarios under the management regime BAU (Figure 4). The annual net revenue during the early phase of simulation was lower for the management regime BAU than other management regimes due to lower harvest levels and absence of storm damage. The annual net revenue obtained from Kronoberg county was similar under all climate scenarios considered until 2030 for all the management regimes. Afterwards the annual net revenue obtained in future climate scenarios (RCP4.5 and RCP8.5) was higher than historic climate. The average annual net revenue obtained during the mid-phase of the simulation period (2050-2070) was 10.9% and 18.6% higher in future climate scenarios RCP4.5 and RCP8.5 respectively than the historic climate scenario. By the end phase of the simulation period (2080-2100) the corresponding values were 14.9% and 34.7% higher in RCP4.5 and RCP8.5 scenario respectively than the historic climate scenario.
Annual net revenue was reduced by storm events in all climate scenarios under the BAU+Storm and PAS regimes. The storm events during the years 2065 and 2095 strongly reduced the corresponding annual net revenue, and the reduction was stronger in the future climate scenarios (RCP4.5 and RCP8.5) than in the historic climate scenario. The storm event in 2065 also reduced the annual net revenue obtained under the management regime BAU+Storm by 56.7% (RCP4.5) and 41.8% (RCP8.5) in the future climate scenarios than the preceding year. Similarly the annual net revenue under the PAS regime was reduced by 78.8% (RCP4.5) and 64.1% (RCP8.5) by the storm event in the year 2065 when compared to the net annual income obtained in the year 2060. Interestingly, however, the reduction in annual net revenue by storm event was much less under the SR management regime. The annual net revenue obtained in the SR regime was only reduced by 19.3% (RCP4.5) and 12.4% (RCP8.5) when compared to the BAU regime during the storm event in the year 2065. A similar trend was also found during other storm events.

Storm-felled volumes under different management regimes and climate scenarios
The greatest levels of storm-felling occurred during the storm event in the year 2065. The storm-felled volume was higher in the future climate scenarios (RCP4.5 and RCP8.5) than the historic climate scenario under all management regimes ( Figure 5). The storm-felled volume under the PAS management regime was lower than under BAU+Storm and that under the SR regime was substantially lower. The storm events during the years 2030 and 2095 were of the same intensity, but the storm-felled volume was higher in 2095 especially in the future climate scenarios ( Figure 5).

Storm-felled volume in management regime PAS
Most of the clear felled stands in the PAS regime were regenerated with Scots pine and therefore during the end of simulation period both Norway spruce and Scots pine equally dominated the forest landscape in Kronoberg county. The storm-felling in Norway spruce stands was higher than Scots pine stands during storm event in the near future (up to the year 2045; Figure 6). However, a strong storm event in the year 2065 (intensity similar to storm Gudrun in Sweden in January 2005) uprooted Norway spruce and Scots pine stands equally. More Scots pine stands were storm-felled than Norway spruce in future climate scenarios during the storm event at the end of simulation period (2095).

Impact of climate change on production of forest landscape
The annual volume increment predicted by the 3PG-Heureka model in the future climate scenarios RCP4.5 and RCP8.5 was higher under all management regimes than in the historic climate scenario; the greatest increase in the annual volume increment occurred in the RCP8.5 scenario. The impacts of climate change on forest production were visible only after the year 2030 ( Figure 3). However, even until 2030 the annual volume increment was increasing under all climate scenarios and management regimes. This was due to accumulation of growing stock in the forest landscape due to less harvest removals when compared to the growth. Currently only around 70% of the total annual growth is removed during harvesting operations in Sweden (Skogsstyrelsen 2014).the annual volume increment does not increased as drastically in the SR regime when compared to other management regimes due to frequent final fellings in the SR regime. Previous landscape-level simulation studies using climate change adapted Heureka-Decision Support System (DSS) has shown that the growth in southern Sweden would increase by 36% in the RCP8.5 scenario at the end of this century (Eriksson et al. 2015). Other landscape-level simulation studies using climate change adapted HUGIN-DSS and EFISCEN-DSS have also revealed that the growth of Swedish forests would increase by 33-35% in Special Report on Emission Scenario (SRES) A2-scenario at the end of this century (Pussinen et al. 2009;Poudel et al. 2011). The increase in growth as per the prediction by the 3PG-Heureka model is 21% in the RCP8.5 scenario, which is lower than the above-mentioned former studies. This is because in the cited studies, the effect of climate change was implemented indirectly using process-based models in the DSS such as Heureka, HUGIN and EFISCEN (Pussinen et al. 2009;Poudel et al. 2011;Eriksson et al. 2015). The site index input variable in the DSS was changed according to the average change in climate response (20-30-year period) on production estimated using the process-based models (Poudel et al. 2011;Eriksson et al. 2015). In the 3PG-Heureka model the empirically based stand level growth simulators in the Heureka-Regwise model were replaced using growth simulators of a process-based model 3-PG. Implementing climate response only by increasing or decreasing site index in the DSS could probably underestimate the adverse effects of future climate like summer frost and drought, because it is not possible to account for the extreme variations in future climate only by changing the site index parameter in the DSS software. This might underestimate the extreme values of future climate change. This could be one of the reasons for the variation in results in the two modelling approaches.

Impacts of storm events on the forest landscape's production and economy under changing climate
Under the BAU+Storm regime, the annual volume increment was much lower than that under the BAU regime because of the storm events included in the former case. The level of storm-felling varied with the intensity of the storms throughout the simulation period (Figure 6), and was consistently higher under the future climate scenarios than in the historic scenario with storm events. However, the simulated storm- felling in 2095 was considerably greater than that in 2030 under the BAU+Storm regime for all climate scenarios ( Figure 6). This could be due to an increased proportion of conifer monoculture and an accumulation of standing volume in the forest landscape that is promoted by the current management patterns in Swedish forests. Thus the results of this study confirm the findings of Drössler et al. (2014) and Hanewinkel et al. (2011) that the likelihood of storm-fellings will increase in the future.
The annual net revenue increased over time under the BAU management regime. However, storm events caused dramatic reductions in annual revenue during the years 2065 and 2095 under the BAU+Storm regime. These predictions are borne out by practical experience: in 2005, storm Gudrun reduced the value of storm-felled trees by 50% and increased the cost of harvesting operations in Kronoberg county by 100%, causing huge losses to forest companies. This demonstrated that storms could dramatically affect both the production of a forest landscape and also its economic output. A similar storm event in the future would cause huge damage to the forest landscape and huge losses to forest industries. It would therefore be very desirable to identify management strategies that will reduce the occurrence of storm-felling.

Alternative management measures for mitigating storm damage
Two management regimes intended to minimize the impact of storm damage were considered in this work: PAS and SR. The PAS regime resulted in slightly lower reductions in the annual volume increment and annual net revenue following storm events than the BAU +Storm regime (Figures 3-4). Meanwhile the stormfelled volume under the PAS regime was lower than under the BAU+Storm regime ( Figure 6).
The landscape was equally dominated by Norway spruce and Scots pine stands in the management regime PAS by the end of the simulation period. However, increase in Scots pine has not contributed to a reduction in storm-felling in the PAS regime. Norway spruce is generally considered to be more susceptible to storm damage than other conifer species of similar height such as Scots pine (Valinger et al. 2006;Albrecht et al. 2012). Accordingly, around 51.3 million m 3 of Norway spruce and 11.3 million m 3 of Scots pine were storm-felled after storm Gudrun in 2005 (Valinger et al. 2006). It should, however, be noted that Norway spruce dominated the forest landscape in southern Sweden and that may be why it was so much more heavily felled during this storm. The risk of wind throw increases with mean stand height (Valinger et al. 2006). Extensive regeneration of Scots pine stands was performed in under the PAS regime, with the final felled Norway spruce stands being replaced with Scots pine stands at most sites. Consequently, by the storm event in 2030, most of the Norway spruce stands in the county were quite old while the Scots pine stands were quite young. Old Norway spruce stands are taller and thus more prone to storm damage, which may be why they suffered more storm-felling. During the storm event in the year 2095, Scots pine stands were slightly more abundant than Norway spruce, resulting in a greater level of storm-felling of Scots pine stands.
Overall, the risk of storm-felling under the PAS regime was slightly lower than the BAU+Storm regime. Better results could potentially have been achieved by regenerating with other fast-growing tree species such as hybrid aspen or hybrid larch.
Under the SR management regime, the annual volume increment and annual net revenue obtained were not drastically affected by storm events (Figures 3-4). The stormfelled volume was also considerably lower than under the BAU+Storm or PAS regimes ( Figure 6). This regime leaves stands un-thinned, resulting in more frequent clear felling operations. Due to lack of thinning there was better mutual support among the trees against storm in the stand and due to frequent clear felling the mean height of the stands was not as high as in the BAU+Storm and PAS regimes, thus reducing storm-felling. Management activities such as felling appreciably enhance the risk of storm damage in forest stands; the risk of storm damage in a recently thinned conifer stand is high and the extent of damage is linearly related to the intensity of thinning (Wallentin and Nilsson 2014). Therefore, an SR regime with shortened rotation periods could represent an efficient landscape-level strategy for reducing the risk of storm damage. Due to reduction in storm-felling the annual net revenue obtained in this regime was not significantly affected by storm events when compared to the BAU+Storm and PAS regimes.

Trade-offs and synergies in alternative management regimes
There was a trade-off in total growth and total harvested volume in alternative management regimes when compared to the BAU+Storm regime. This was due to the frequent number of harvests due to lack of thinning operation in the SR regime and promoting slow-growing Scots pine in the whole landscape. The highest accumulated standing volume of all the management regimes with storm events was achieved in the SR regime (Table 2), allowing an increase in the annual harvesting rate under this regime and thus raising the net income obtained.
Increased harvesting would also reduce the risk of stormfelling. Since the landscape was unthinned under the SR regime, more frequent final felling were conducted in the landscape. This has increased the annual harvesting rate under this regime and thus raising the net income obtained. Increased harvesting would also reduce the risk of stormfelling. In this regime Norway spruce was extensively regenerated. Previous stand level studies have shown that a thinning-free regime with short rotation periods could increase the profitability compared to conventional management of Norway spruce stands (Subramanian et al. 2016a). However, it is difficult to conclude that such an approach would improve profitability at a landscape level because of the risks associated with a thinning-free management, such as increased risk of natural mortality and snow damage (Rössler 2006;Štefančík 2012). The SR strategy also presents few selection opportunities and thus little scope to tune future wood quality, and prevents forest owners from deriving income via thinning at an early stage. Therefore, the financial benefits of adopting a thinning-free approach such as the SR regime could be smaller than expected. In the SR regime the proportion of stands with an age less than Sweden's Legal Minimum Age of Final Felling (LAF) regulations (Skogsstyrelsen 2010) could increase due to frequent clear felling. It would not be legal to harvest these young stands until they reached LAF. This could result in building up of standing volume in the landscape in this regime, which could increase the risk of storm-felling in the future. Moreover, it would also result in a landscape with dense Norway spruce stands that could be less attractive for recreational purposes and would negatively affect the landscape's biodiversity.
Apart from storm events, climate change could also increase the risk of damage caused by pests and pathogens (Blennow et al. 2006). Major pests and pathogens found in Swedish forests are root and butt rot (Heterobasidion annosum [Fr.] Bref. and Heterobasidion pariviporum [Fr.]), the Spruce bark beetle (Ips typographus L.), and the Pine weevil (Hylobius abietis L.). Shade-tolerant species such as Norway spruce are more prone to Heterobasidion parviporum infection than shade-intolerant species such as Scots pine (Piri and Valkonen 2013). Under the PAS regime, the majority of Norway spruce stands were extensively regenerated with Scots pine after final felling. This regime thus has advantages beyond the reduction in storm-felling: it should also reduce the incidence of infections caused by Heterbasidion parviporum and spruce bark beetle attacks while also increasing the landscape's biodiversity.

Uncertainties associated with the model simulation
There is a high level of uncertainty in projection of stochastic events such as storms (Rummukainen 2012).
In this study, historical storm events that occurred between 1948 and 2013 were simply replicated over the simulated period, with intensities identical to those recorded during the real storms. However, it is not necessarily likely that storms with similar intensities will reoccur. According to Pryor et al. (2012), the wind density in the Baltic sea region (which includes southern Sweden) will be around 10-15% greater in the near future than it has been in the recent past. The occurrence of highintensity storms is predicted to increase over Western Europe during the early autumn as an effect of climate change (Haarsma et al. 2013). The conclusions of this study may need to be reviewed if the intensity and timing of storm events are expected to vary in future.
There are several other factors whose influence on the growth and development of forest landscape could increase as a result of climate change that were not considered in this study. For example, the intensity of root and butt rot infection, spruce bark beetle damage, pine weevil infestations, summer frost, forest fires and changes in tree phenology could all affect the growth and development of forest landscapes to a greater extent than they do at present. The alternative management strategies recommended in this study may not be effective against these risk factors. For example, a thinning-free short rotation management regime could reduce the adverse effects of a future storm event in a forest landscape. However, it could also increase the risk of forest fires and snow damage. Therefore, before implementing alternative management strategy, the associated uncertainties should be considered carefully.

Conclusions
The annual volume increment of the forest landscape in Kronoberg county is predicted to increase by 8.6% and 21.0% respectively in RCP4.5 and RCP8.5 scenarios by the end of this century under the management regime BAU.
• The annual volume increment was dramatically reduced by major storm events under the BAU +Storm and PAS management regimes relative to the BAU case. Under the BAU+Storm regime, the storm event during the year 2065 reduced the annual volume increment by 7%, 9% and 12% in the historic, RCP4.5 and RCP8.5 climate scenarios, respectively. • The total standing volume was reduced by major storm events under the BAU+Storm and PAS management regimes by the end of the simulation period.
• The storm-felled volume under the PAS management regime was slightly lower than that under the BAU+Storm regime, while that under the SR regime was substantially lower than in either of the other two cases. • The simulated annual net revenue obtained from Kronoberg county also increased over time due to climate change. However, storms dramatically reduced net revenue under the BAU+Storm and PAS regimes. The net revenue obtained under the SR management regime was not reduced much by storm events. The SR regime may thus represent a possible adaptive strategy for minimizing storm damage in southern Sweden. • The growth predictions of the 3PG-Heureka hybrid model were less than the predictions made by previous landscape-level models using the Heureka, Hugin and EFISCEN Decision Support Systems. The growth simulator in the 3PG-Heureka model is climate sensitive, However, in the previous studies the effects of climate change were implemented indirectly using an approximation approach resulting in underestimation of extreme variations in future climate (e. g., droughts, summer frosts). • There are many uncertainties associated with the simulations and conclusions presented in this work.
In particular, several factors that could influence the growth and development of forest landscapes were not considered. All these uncertainties should be considered carefully before implementing any alternative management strategy.