Towards robust projections of future forest dynamics: why there is no silver bullet to cope with complexity

Abstract

Forest ecosystems are of key importance for providing a broad range of ecosystem goods and services. Yet, forest ecosystems are subject to manifold pressures and considered to be particularly sensitive to rapid climate change. A particular uncertainty remains in larger scale assessments (i.e., regional to national scale) due to the effects of local-scale drivers of climate change impacts (e.g., related to site-, species- and stand-specific responses), which is particularly important for forest management and planning. Multiple dynamic vegetation models (DVMs) have been developed to analyze forest ecosystems and their dynamics. Among the various types of DVMs, forest gap models (FGMs) have been introduced to represent stand-scale dynamics as upscaled from individual tree behavior, thus allowing to represent uneven-aged, mixed-species stands. This renders them principally suitable to assess the future development of today’s forest stands under changing environmental conditions while accounting for site-specific factors. Even though many ecological models strive for generality and realism, so far few FGMs have been applied across large areas (e.g., the national to continental scale). This may be due to parameterization problems and insufficient representation of underlying mechanisms. To overcome these issues, many models are (re-)calibrated locally or adapted structurally to reflect local conditions. Yet, such strategies tend not to increase a model’s general applicability across a broad range of conditions. Furthermore, model developments aiming at structural improvements typically focus on one single process. However, due to unaccounted process interactions, such efforts are at risk of ‘getting the right patterns for the wrong reason’ (i.e., missing ecological reality). Thus, recent pleas have been made to jointly revisit the ecological processes at the core of DVMs to increase the consistency of process interactions. Using the state-of-the-art FGM ForClim, this PhD thesis aims to enhance the robustness of the projections for various fields of FGM applications (e.g., natural and managed forest stand dynamics) across a range of spatial and temporal scales. A particular focus was on the consistency of the simulated processes and their interactions. Specifically, these efforts were geared towards providing locally accurate climate impact assessments over large areas, representing a current research frontier in FGMs and DVMs in general. In Chapter 1, I conducted a sensitivity analysis of the ForClim model to analyze model behavior and identify model components that deserve particular attention for reducing uncertainties of parameter estimates and/or improving process representation. Since the relative parameter influence may not be constant over time and likely varies with site conditions, stand structure and species composition, I analyzed the model’s parameter sensitivity at 30 representative sites across Europe and compared results for monospecific and mixed stands at two system states in time (i.e., early and late succession). Key parameters causing the largest variability in model outputs were related to tree establishment, the water and light regimes, growth and temperature, whereby the relative parameter influence of the latter strongly varied with local climate conditions. Moreover, model sensitivity differed between monospecific and mixed stands as well as between early and late successional stages, reflecting the differential influence of ecological processes with stand structure. In addition, model application at a European scale (i.e., far beyond the geographical range ForClim was originally developed for) pointed at model shortcomings. Since ForClim proved to be highly sensitive to water-related parameters across most of the European continent (Chapter 1) and accurately representing water limitations in DVMs is of utmost importance in view of climate change, I revisited ForClim’s water module in Chapter 2. Most DVMs explicitly model water availability based on a water balance with potential evapotranspiration (PET) as the main driver. I assessed the performance of ForClim’s water module, which includes the PET formulation by Thornthwaite and Mather (1957), by confronting simulated with observed monthly AET at forested FLUXNET sites. Further, I included alternative PET formulations in the comparison, particularly the one by Priestley and Taylor (1972) featuring a higher degree of mechanistic detail. The performance of the water module as applied in ForClim depended primarily on climate type independent of the PET formulation applied. I thus conclude that increasing the complexity of the PET formulation will hardly improve the estimates of water deficiencies at an annual scale in ForClim. Rather, more attention should be payed to forest-specific features in the context of the water balance, such as the representation of belowground and phenological processes, because most water balance models have been developed for agricultural applications. In Chapter 3, I scrutinized the representation of ForClim’s core processes and the consistency of their interplay. I developed a set of alternative process and parameter representations for the core processes light availability, tree establishment, growth and mortality, based on ecological theory and diverse sources of empirical data. I applied a pattern-oriented modeling (POM) approach to test all combinations of the standard and alternative formulations (yielding 504 model versions) against a comprehensive set of empirical patterns for diverse model applications and a wide range of site conditions. I found that adapting one process in isolation can improve model performance for one specific application. However, the best model versions typically included more than one alternative process or parameter formulation. Thus, simultaneously considering multiple core processes is key for revealing internal inconsistencies in the model framework and model improvements. In this context, POM proved to be highly suitable to bridge various fields of model application and to compare model outputs with a broad set of patterns comprising diverse forest characteristics at different temporal and spatial scales. I conclude that the forest ecology community should make good use of the ever-increasing data availability and the POM framework to challenge the core processes of DVMs in a holistic manner. Finally, the increasing impacts of climate change on forest ecosystems have triggered multiple model-based impact assessments for the future, which however feature considerable uncertainty regarding local impacts over larger areas (i.e., regional to national scales). In Chapter 4, I aimed at bridging this gap by analyzing the climate change sensitivity until the end of the 22nd century for 71 typical managed Swiss forests. To account for various sources of uncertainty in the projections, the effect of eight different model versions (developed in Chapter 3) as well as alternative soil types and climate change scenarios were considered. The simulations showed substantial changes in basal area and species composition, with dissimilar responses to climate change across and within elevation zones. I identified the following stands as being most prone to negative climate-induced impacts: (1) stands in the sub-montane and low montane elevations zones and (2) stands located on poor soils in the high montane and subalpine elevation zones. The introduction of additional, more climatically adapted species partly mitigated the negative impacts of climate change, suggesting that admixing drought-tolerant species is advisable across all elevations to increase the resistance and resilience of forest stands to climate change. Yet, the large influence of site conditions and the choice of the forest model on some of the projections indicates that uncertainty sources other than the climate change scenarios need to be considered in impact assessments. By considering diverse sources of uncertainty, including structural and parameter-related uncertainties of the model, I was thus able to demonstrate their key relevance for an improved, evidence-based decision support in forest management under climate change. Throughout the thesis, I presented an approach to improve the robustness, accuracy and generality of FGMs, and demonstrated how to upscale climate-change impact assessments from the local to the national scale, which I believe are important steps in advancing the frontier of large-scale applications of FGMs. The insights from this thesis are furthermore relevant for other DVMs, especially for those that feature strong structural similarity with ForClim.