Forest site productivity in temperate regions: empirical modelling in environmental and geographical space

Worldwide forests play an essential role in sustaining prosperity and well-being of human society thanks to the services they deliver like climate regulation, soil and water protection, provision of wood and non-wood forest products, recreational and spiritual significance and by hosting large part of the terrestrial biodiversity. In the last century, forests are experiencing an abiotic environment that changes much faster than during the past hundreds of years. Moreover, forests are susceptible to large pressures of different kinds, including biodiversity loss, pests and diseases, soil degradation, etc. In order to guarantee future provision of all currently provided ecosystem services forest management should continue to evolve into more sustainable and multifunctional directions. Accurate forest site productivity estimation, e.g. by means of site index, is one of the crucial elements of good forest resource management, since site productivity is a key indicator of forest ecosystem services like wood production and carbon sequestration. It is therefore crucial in estimating the future wood stocks, selecting appropriate locations for planting specific tree species or choosing the most appropriate management at a given location.In homogeneous even-aged forest stands, site index of the standing species can be directly inferred from measurements of tree height and age, using appropriate species-specific dominant height growth curves. But in mixed or uneven-aged stands, or in the case of stand conversion to other tree species, or afforestation of unforested land, or under site conditions which change with time, direct estimation is not possible. For such cases site index needs to be estimated indirectly from environmental factors like climate, topography and soil characteristics, using appropriate models. The overall aim of this research was to contribute to the development of a generic approach for multifactor modelling and spatial prediction of forest site productivity, and to identify the most important influential site quality variables for three important tree species of European temperate lowlands: pedunculate oak (Quercus robur L.), common beech (Fagus sylvatica L.) and Scots pine (Pinus sylvestris L.). We established in homogeneous, even-aged stands of each of these species an amount of research plots which were surveyed in detail for dendrometrical, topographical, litterfall, vegetation, humus and soil characteristics. Based on these data, a stepwise approach starting from the evaluation of the performance of different non-spatial modelling techniques to predict site productivity, revealed boosted regression trees (BRTs) and generalized additive models (GAMs) as the most appropriate empirical modelling techniques to predict forest site productivity in environmental space. Both their accurate predictive performance and good ecological interpretability make these techniques preferred in a wide range of situations. GAM predictions reached a higher predictive accuracy, whereas BRT models hadsome additional advantages with respect to interpretability of ecological processes. Besides, scale proved to be an important issue in successful forest site productivity modelling. Empirical site index models proved to be very scale-dependent and their applicability was limited to the scale of development.In a final step, empirical modelling was expanded from environmental space to geographical space, incorporating the geographical location of observations explicitly, in combination or not with environmental attributes, to end up with predictive maps estimating the site productivity for the entire study area of Flanders. Although hybrid regionalisation techniques, as regression-kriging or co-kriging, accounting for both spatial dependence and environmental contrasts would be expected to result in the best regionalisation approaches for forest site productivity predictions, this was not the case in all situations. Depending on the availability and the nature of the geodata, different spatial empirical modelling approaches were recommended for predictive mapping. Since no approach outperformed the others under all circumstances a decision tree was developed providing guidance in selecting the most appropriate technique considering the availability and nature of the geodata.Searching for the most influential site productivity variables, BRTs revealed that, although with different effects and in interaction with other co-variables, soil granulometric fractions and litterfall nitrogen concentrations were the most effective predictors of all three important tree species of a temperate lowland region: pedunculate oak, common beech, and Scots pine. Since Flanders is characterised by a pronounced north-south gradient of decreasing sand and increasing silt fractions, and since soil granulometry plays an important role in an optimal water holding capacity and a better nutrient retention, it is not surprising that for all species improved site productivity was recorded at sites with increasing silt fractions. More surprising was the negative effect of litterfall nitrogen concentrations for all species. Although many studies revealed a fertilising effect of increased nitrogen deposition, nitrogen saturation seemed to reduce species productivity in this region characterised by high nitrogen deposition. Tree ring analysis revealed moreover for common beech a long-term growth trend over the last century. The trend was characterized by an initial growth increase, reaching its maximum around the 1960 s, and followed by a recent growth decrease lasting until present. With an observed growth increase of maximal 19 to 24% and an overall growth increase over the 20th century of 12 to 18%, the observed long-term changes were very similar to the changes described in comparable studies of common beech in the temperate lowlands as well as those observed at beech s southern and elevation range edges. This consistency with observations elsewhere in Europe suggests an overall recent decreased vitality of common beech in Europe.Dankwoord Table of Contents List of Tables List of Figures Abbreviations and Symbols Abstract Samenvatting Introduction Chapter 1 Comparison and ranking of different modelling techniques for predicting site index in Mediterranean mountain forests Chapter 2 Evaluation of modelling techniques for forest site productivity prediction in contrasting ecoregions using Stochastic Multicriteria Acceptability Analysis (SMAA) Chapter 3 Predicting forest site productivity in temperate lowland from forest floor, soil and litterfall characteristics using boosted regression trees Chapter 4 Long-term growth changes of common beech in a temperate lowland region during the last century: a tree ring analysis comparing linear and non-linear mixed modelling approaches Chapter 5 Effects of scale and scaling in predictive modelling of forest site productivity Chapter 6 Comparison of location-based, attribute-based and hybrid regionalisation techniques for mapping forest site productivity Conclusions and perspectives References Appendix 1: supporting figures to Chapter 3 Appendix 2: Belgian drainage classification system as a function of the soil texture List of publicationsnrpages: 185status: publishe