A GIS Model to Estimate a Sustainable Potential of Forest Fuel for Energy Generation in the Municipality of Växjö, Sweden

(1)

A GIS Model to Estimate a

Sustainable Potential of Forest Fuel

for Energy Generation in the

Municipality of Växjö, Sweden

Gunnar Wohletz

Master’s of Science Thesis in Geoinformatics

TRITA-GIT EX 11-005

School of Architecture and the Built Environment

Royal Institute of Technology (KTH)

(2)

Abstract

Since the 1980s the municipality of Växjö has been increasingly focusing on using wood to produce energy for the region. A permanent and sustainable supply of wood material is therefore indispensable. The main source for this wood fuel is hereby harvested wood from forest which can be used as energy, so-called forest fuel. The objective of this thesis is to develop a model to estimate a sustainable potential of forest fuel supply until the year 2050 for the municipality using a geo-graphic information system (GIS). The model overall follows a top-down approach that consists of three sequential modeling steps which are generally applicable for biomass potential estimations: the theoretical, technical and the reduced technical potential. As input data the model uses georeferenced forest data (called kNN-Sweden) and topographic data about the study area to describe and narrow down the forest fuel potential by setting numerical or topographic (spatial) parameters for each modeling step. In this report forest data from 2005 has been used, which was obtained short before the storm Gudrun damaged great parts of the Swedish forest landscape. The result shows that the municipality of Växjö should be able to satisfy its demand for energy wood from harvested forest wood alone until around the year 2035, but might have shortages afterwards until the year 2050. This thesis concludes that for the next 40 years the Växjö municipality should not only rely on its annually available forest fuel capacity, instead, different wood resources have to be utilized or forest wood from surplus years have to be stored for future tighter years. Since the forest data was obtained before the storm in 2005, those results must me treated with care, and for more accurate results the modeling steps should be repeated with newer forest data. The report also concludes that the estimation of the forest fuel potentials in this study still lacks of accuracy and that it is advised to treat the numerical modeling results with caution. There’s still room left for further improvement, and therefore possible error sources and suggestions for the future work are listed.

(3)

Acknowledgement

At this point I’d like to express my gratitude to my external supervisor Prof. Dr. Thomas Brinkhoff, Department of Applied Photogrammetry and Geoinformatics, Jade University Oldenburg, who has given me the opportunity to write about this topic in his department. My special thanks here goes to my advisor J¨urgen Knies, who has supported me throughout several months with his suggestions and insights. Also I’d like to thank all my colleagues from the department of Applied Photogram-metry and Geoinformatics who have given me technical support.

Furthermore sincere thanks also go to Hans-Petersson and Mikael Egberth from the Swedish University of Agricultural Sciences (SLU) in Ume˚a, Göran Gustavsson from Energikontor Sydost in Växjö, Alexander Rosenberg from the Chamber of Agriculture Lower Saxony in Oldenburg and Markus Biberacher from the Research Studios Austria in Salzburg for their support and data or literature provisions.

(4)

List of Figures

1.1 The logo and main partners of the North Sea SEP project[8]. . . 3 1.2 The company logo of Energikontor Sydost. . . 4 1.3 The division of Swedish provinces into 4 separate calculation zones

(modified)[14]. . . 6 1.4 The recommended usage of different tree parts[14]. . . 8 1.5 V¨axj¨o’s fossil and renewable energy supply in 2009 as a pie chart

(large green section represents wood)[7]. . . 9 1.6 V¨axj¨o’s fossil and renewable energy supply in 2009 in absolute

num-bers (sharp rising green line represents wood)[7]. . . 9

2.1 Top-down modeling approach for the determination of the biomass potential, analogous to ”Energieregion Rhein-Sieg” report[13]. . . 11 2.2 The effective thermal value of wood depending on its moisture content[17]. 13 2.3 A spruce tree and its above-ground components[18]. . . 15 2.4 The major moments of Marklund’s study[18]. . . 16 2.5 The location of the sample compartments all over Sweden. The

dashed line marks the border between northern and southern Sweden[18]. 17 2.6 The simplest (top) and best (bottom) biomass function for the ”stem

over bark” component for pine trees[19]. . . 18 2.7 The biomass distribution of single tree components depending on the

tree’s diameter[18]. . . 19 2.8 The inventoried below-ground biomass according to Marklund (i) was

complemented by measuring all major axes (d). A proportion of outer roots was excavated and used for derivation of regression functions to estimate the biomass of all outer roots[20]. . . 20 2.9 Below-ground biomass functions for Pinus sylvestris (pine).[20]. . . . 20 2.10 Division of biofuel, wood- and forest fuels into sub-groups by origin

of raw material and integration of flow[21] . . . 21

3.1 Map of the municipality of V¨axj¨o including the approximate locations of the CHP-plant (big circle) and local heating stations (small circles). 26 3.2 Illustrated example of applying the kNN-interpolation-method on an

(8)

3.3 Standard error of kNN-estimates of the total wood volume in the

experimental area of Remningstorp (V¨astra G¨otaland)[28]. . . 34

3.4 kNN-Sweden original forest biomass data near Bra˚as, municipality of V¨axj¨o. . . 35

3.5 kNN-Sweden forest biomass data averaged within forest stands near Bra˚as, municipality of V¨axj¨o. . . 37

3.6 Energy supply of fossil and renewable fuels in V¨axj¨o from 1993 to 2009 (in green: wood)[7]. . . 40

4.1 A simplified overview over the complete workflow to model the forest fuel potentials. . . 42

4.2 Model of the forest classification in ArcGIS ModelBuilder. . . 44

4.3 Model to create congruent managed forest raster datasets. . . 47

4.4 A detailed but still simplified model of the thinning period of a forest: The black line corresponds to the assumed real available biomass, where in a certain interval the forest is thinned and therefore a share of its biomass is lost, though on a long-term view the forest still grows. The red line corresponds to the regression curve that will be computed in SPSS. It is an approximation to the real available biomass, since the degree of thinning (light or strong) and interval size (5 or 10 years) are unknown, but still assumed to be carried ou regularly. . . 49

4.5 Model to create the theoretical forest fuel potential . . . 52

4.6 An illustration of the idea to project the total yield from the com-mercial thinning onto each year within that period. The percentages of 30% and 70% were assumed here. . . 55

4.7 Model to create the technical forest fuel potential. . . 58

4.8 Proportions of biomass of different tree parts of the 2003 harvest in the country and how much is used for energy purposes[14]. . . 62

4.9 Proportions of different tree parts (spruce) depending on the diameter at breast height (DBH)[17]. . . 63

4.10 Verification of proportions of different tree parts (spruce) depending on the DBH (calculated and plotted in Matlab). . . 63

4.11 Model to create polygon-shapefile with ”blocked” forest. . . 67

4.12 Energy conversion process . . . 67

5.1 Natural habitats near Bra˚as, municipality of V¨axj¨o. . . 71

5.2 Slopes within the municipality of V¨axj¨o. . . 72

5.3 Complete classified forest cover in the municipality of V¨axj¨o. . . 73

5.4 Classified forest near Bra˚as, municipality of V¨axj¨o. . . 74

5.5 Age histogram of the different managed forest types . . . 75

5.6 Biomass histogram of the different managed forest types . . . 76

5.7 Curve fitting for the different managed forest types (age ≥ 10) . . . 76

5.8 Linear fitting for the different managed forest types (age ≥ 10) . . . 77

(9)

5.10 Theoretical potential of forest fuel in 2015 near Bra˚as, municipality of V¨axj¨o. . . 81 5.11 Constraint map within the technical potential model near Bra˚as,

mu-nicipality of V¨axj¨o. . . 84 5.12 Sample area to test the model for the determination of blocked forest

sections in ArcGIS. The orange-filled areas are buffer areas that the forestry machines are are able to cover (here: 300m). The orange-striped areas are the resulting areas that depict forest areas that are blocked by water from the access roads direction inside the buffer. . 85 5.13 Energy conversion of wood according to the energy balance of 2009

(10)

List of Tables

3.1 Coordinates of the maximum geographical extents of the municipality

of V¨axj¨o (in SWEREF99 TM). . . 28

3.2 The geographical rectangle’s border coordinates of the downloaded data (in SWEREF99 TM) . . . 29

3.3 Overview over the datasets which are contained in the kNN-Sweden dataset from 2005. . . 33

3.4 Overview of newly created raster datasets on a forest stand level. . . 36

3.5 Available and used biomass functions in kNN-Sweden for the most common tree types in Sweden. . . 38

4.1 Allocation of tree species to tree types. . . 44

4.2 Input datasets for the forest classification model. . . 45

4.3 Output datasets for the forest classification model. . . 45

4.4 Input parameters for the forest classification model. . . 45

4.5 Input datasets to model congruent managed forest raster datasets. . 46

4.6 Output datasets after modeling congruent managed forest raster datasets. 47 4.7 Input datasets for modeling the theoretical potential. . . 51

4.8 Output datasets from modeling the theoretical potential. . . 51

4.9 Input parameters for modeling the theoretical potential. . . 53

4.10 Input datasets for modeling the technical potential. . . 57

4.11 Output datasets from modeling the technical potential. . . 58

4.12 Input parameters for technical potential . . . 59

5.1 Roads types that were contained the topographic data including the results if they were classified as suitable or unsuitable as an access road for forest. . . 70

5.2 Amount of valid age/biomass data tuples that were used for the re-gression analysis. . . 74

5.3 Input datasets to find managed forest . . . 77

5.4 Default felling ages of the forest types set in the theoretical model. . 78

5.5 Default mean annual outtake of forest biomass during the commercial thinning period in the theoretical potential model. . . 78

(11)

5.7 Percentages of different tree parts at felling age . . . 82 5.8 Percentages of dividing stem wood . . . 83 5.9 Percentual share of the timber residue from the stem over bark

(12)

Chapter 1

Introduction

This master thesis in Geoinformatics is settled in the area of renewable energy planning. In the following sections this topic will be introduced particularly in reference to the problem dealt with during the thesis.

1.1 Swedish energy situation and ambitions

These years the Swedish energy policy is more and more affected by international decisions. Terms such as ”global warming” and ”greenhouse gas emissions” are more and more moving into the center of the attention on an international or European policy level. The Kyoto protocol agreement will end in 2012 and another following internationally binding agreement hasn’t been found yet (e.g. on the UN Climate Change Conference in Copenhagen in 2009). Even though (or maybe because of that) the European Union already passed several own directives and guidelines to its member states to accelerate the transformation towards a sustainable energy future. For example by the end of 2008 it passed the often referred to as ”20-20-20” directive (directive 2009/28/EC[1]), which aims to reduce greenhouse gases by 20%, to have a 20% energy share of regenerative energies and increase the energy efficiency by 20% in 2020 compared to 2005. It also included a renewable energy share of 10% in the transport sector. Since the member states had different starting points to reach those goals they have been assigned individual goals to reach. Under the directive each member state had to create a ”National Renewable Energy Action Plan” (NREAP) present its future energy goals.

(13)

of fossil fuels by 2030[4]. To reach those goals Sweden’s already passed several directives to promote the national electricity generation by using renewable sources, for example the ”H˚allbarhetskriterier f¨or biodrivmedel och flyttande biobr¨anslen[5]” (Sustainability criteria for biofuels and bioliquids).

V¨axj¨o: The Greenest City of Europe

When the British BBC visited Växjö in 2007 they awarded the city with the title ”The Greenest City in Europe”. In fact the municipality of Växjö was indeed a good example for a green city. Efforts towards a green city had already started in the 1970’s, and in the 1980’s district heating from renewable energies has already been introduced in the region. During the 1990’s Växjö already decided to become a fossil fuel free municipality in the future.

Therefore Växjö has even more ambitious goals for the coming future than the Swedish action plan schedules for the 2020 goals. In 2006 (revised in 2010) the city of Växjö decided several goals for 2015, more precisely for example to reduce carbon dioxide emissions by at least 55% per inhabitant compared to 1993 (already reached a 33% decrease by 2008) or to reduce the total energy consumption per inhabitant by 15% compared to 2008. Until 2030 it shall be a fossil fuel free city[6].

(14)

1.2 The North Sea Sustainable Energy Planning

Inter-reg Project

The North Sea Sustainable Energy Planning project (short: North Sea SEP1) is supported by the European Regional Development Fund (ERDF) and is a part of the North Sea Region Programme (Interreg IV B). Interreg is an initiative from the European union and by now is in it’s 4th running period (2007-2013). The North Sea SEP project started on the September 1st 2009 and will run until August 2012 (3 years) and has a budget of over 5 million euros. The Jade University in Oldenburg (Germany) has the lead beneficiary role in the project. In total there are 26 project partners (usually companies, universities or municipalities) in 6 countries (Germany, Netherlands, Belgium, UK, Sweden and Denmark), including the Jade University in Oldenburg as well as Energikontor Sydost (the Energy Agency for Southeast Sweden, see section below), which headquarter is situated in V¨axj¨o (see figure 1.1). The project aims to ”develop and promote a model for regional development centered on renewable energy and energy efficiency activities, and to meet the needs of local and regional authorities undertaking sustainable energy planning”[8].

Figure 1.1: The logo and main partners of the North Sea SEP project[8].

(15)

Energikontor Sydost (Energy Agency for Southeast Sweden)

Since 2007 the company2 (founded 1999) is ”jointly owned by by an association where regional councils, counties and municipalities in Blekinge, Kalmar and Kro-noberg are members”[9]. Originally it has been established ”as an EU project under the association of local and regional authorities in Kronoberg” due to the grow-ing focus on climate change and connected energy- or transport-relevant issues. Its function is to ”initiate, coordinate and implement projects aimed at improving the energy efficiency and increased supply of renewable energy in all sectors of soci-ety”. To its main clients belong the European Union, the Swedish Energy Agency (Energimyndigheten), the Swedish Road Administration and the Swedish National Agency for Education. The company operates in three Swedish provinces (Swedish: l¨ans): Kronoberg, Blekinge and Kalmar.

Figure 1.2: The company logo of Energikontor Sydost.

1.3 Biomasses for renewable energy production

1.3.1 Definition and study examples

At the beginning the definition of renewable energy should be clarified. Renewable energy (in contrast to fossil fuel) is energy that comes from renewable natural re-sources. Those energy sources are often categorized in slightly different ways, but are often categorized as follows[10]:

• water • wind • biomass • solar • geothermal

(16)

determine the potential of biomasses in certain regions. Usually they were concen-trated on biomass potentials on a national scale, for example in Spain [11], where the study concentrated on biomass from agro-industrial residues, or in Portugal [12], where the study concentrated on forest biomasses for the whole country. But even from those studies it was already noticeable that the estimation of biomass potentials is a very complex matter, and the possible biomasses sources are considerably.

One example of the estimation of more regional biomass potentials is the ex-tensive study ”EnergieRegion Rhein-Sieg”, which was performed by the Research Studios Austria (RSA) in Salzburg for a municipality in North Rhine-Westphalia in Germany. The final report was published in January 2008 and included an analysis of the energy situation, potential estimations of renewable energy sources (including biomasses) and estimations for energy consumptions in the municipality[13]. In the study a GIS system was used to model the biomass potentials for the region. The biomass potential of this study included the estimation of silvicultural biomasses (so including forest wood), agricultural biomasses and biomass waste. To estimate the biomass potentials the authors used a top-down approach, which will be further explained in section 2.1.

1.3.2 Forest fuel: forest wood as energy About Swedish forest

The country of Sweden is mostly covered by forest. To map the state and the change of the Swedish forest the Swedish University of Agricultural Sciences (SLU) in Ume˚a annually performs a National Forest Inventory (NFI). The forest areas, growth rates, damages, fellings, forest stand conditions & volumes are measured and statistics are computed and published3. The NFI divides Sweden into 4 calculations zones (Swedish: beräkningsomr˚aden, which are often being referenced to when estimating biofuel potentials (see figure 1.3. Växjö lies within Bo 4. This division into different calculation zones implies already that forest is different in different regions of the country, possibly because of different climate (due to Sweden’s large North-South extent), and that the forest might therefore grow differently within Sweden.

Wood utilization in V¨axj¨o

Since V¨axj¨o (and Sweden in general) owns a great supply of wood by its high forest coverage (around 63% are covered by forest), the wood industry is comparatively large too. Due to the low population there’s a surplus of wood that is exported. Traditionally the high-quality stem wood (which is the biggest part of a tree) is processed into timber which serves as a building material for example for furniture or building houses (see figure 1.4). Wood of less good quality (usually the upper stem) is used as pulpwood to make paper products. Until a few decades ago the rest of the wood, so called logging residues since they fulfilled no industrial purpose at that

(17)

(18)

time (branches with leaves or needles, but also stems and roots), usually remained in the forest, because the demand for that wood wasn’t there. But nowadays those logging residues already have become a most essential renewable energy source in Växjö, which is also since the consumption of wood as biofuel (so called wood fuel ) already started comparatively early in this region. Therefore today wood has already become the most important biomass resource for energy generation in Växjö. Since 1980 the input of wood into the Sandvik CHP-plant and the local heating stations is steadily increasing and equaled the total energy of 900.23 GWh in 2009 (see figure 1.6). In that year wood made up for 39.9% of the total primary energy supply from fossil and renewable energies combined in Växjö (see figure 1.5) or 71.3% of the total renewable energy supply. The demand of wood as an energy resource still has a tendency to rise. Unfortunately nuclear energy is not included in those statistics, so that the share of wood fuel compared to the total energy supply of the region cannot be determined.

The wood fuel is mainly used for district heating in the Sandvik CHP-plant (86.4%), for local heat stations (6.7%), or directly for households and or the industry in the form of pellets or firewood (7.0%)[7]. While the local heat stations only produce heat, the CHP-plant produces heat and electricity in a combined process. In 2009 79.4% of the primary energy was converted into heat, while 20.6% were converted into electricity.

1.4 Research objectives and motivation

(19)

(20)

Figure 1.5: V¨axj¨o’s fossil and renewable energy supply in 2009 as a pie chart (large green section represents wood)[7].

(21)

(22)

Chapter 2

Literature Review

2.1 Model biomass potentials

As mentioned before in 1.3, a very interesting approach to describe biomass poten-tials has been developed by the Research Studios Austria (RSA) in Salzburg, which for example has been applied in their extensive study ”EnergieRegion Rhein-Sieg” (Energy region Rhein-Sieg), which estimated many different biomass potentials for a municipality in the south of North Rhine-Westphalia in Germany[13]. Figure 2.1 describes the top-down approach that was defined by RSA to determine the biomass potentials, i.e. it’s generally applicable for any kind of biomass potential estimation.

Figure 2.1: Top-down modeling approach for the determination of the biomass po-tential, analogous to ”Energieregion Rhein-Sieg” report[13].

(23)

The technical potential limits the theoretical potential by applying technical factors such as topographic issues (slope, access roads, etc.) and competition about the resource (e.g. furniture industry vs. forest fuel industry). Finally the reduced technical potential describes the final energy that reaches the consumers, which always denotes losses during energy conversion and energy transport.

2.2 Important basic terms in forestry

In order to understand the biomass estimation of forest wood, there are some basic terms that are often used in forestry and will also be more or less frequently be mentioned in this report.

• Diameter at breast height (DBH): The DBH is one of the most common den-drometric measurements and a standard method to determine the diameter of the trunk of a standing tree. The breast height hereby refers to the height of an adults breast, and is therefore not a worldwide standardized term. While in Europe the diameter is measured at 1.3 meters above ground, in Africa the DBH is measured at 1.4 meters.

• Forest cubic meter (m3_{f o): the forest cubic meter is a standardized and}

im-portant unit in forestry for measuring the volume of a tree above stump level, so without stumps and roots, but also excludes branches (i.e. only includes the stem wood).

• Dry-weight wood biomass: Expresses the biomass of wood with a moisture content of 0%. Usually wood from a fresh tree in the forest has a moisture content between 45% and 60%, while for air-dried wood it lies somewhere be-tween 12% bis 20%[16]. The moisture content usually decreases the energy output of the wood, since the energy that is needed to evaporate the water needs to be subtracted (see figure 2.2). Usually the energy inside the evap-orated hot steam is lost, which therefore decreases the energy value of wood with a high moisture content (unlike dry wood). It is less dramatic (or even wanted) in a CHP-plant as for example in the Sandvik plant in V¨axj¨o the en-ergy in the flue gases of the burned wood is also used (by performing flue gas condensation and using the hot water directly in the district heating system). Therefore the output energy in this energy conversion process depends on the total energy conversion efficiency of the plant rather than the moisture content of the input wood. If wood doesn’t have to be dried before combustion it also saves the operators time and money.

(24)

energy the efficiency in a CHP-plant is much higher than in a regular power plant, because more energy can be used and therefore less energy is lost. • Site index: The site index is a measure of soil productivity of a forest and is

built by estimating the height of the worst trees in a forest when the stand is 100 years old. In reality the site index also influences the minimum final felling age of a forest stand (see 2.4.3).

Figure 2.2: The effective thermal value of wood depending on its moisture content[17].

2.3 Marklund’s biomass functions

2.3.1 Overview

(25)

In 1988 Marklund published the final report of the ”Biomass functions” project which included biomass functions for pine and birch trees[19]. The functions were developed based on 1289 dried sample trees in total (pine: 493, spruce: 551, birch: 242). For spruce and pine biomass function for the stumps and roots were also developed. The roots were measured down to a diameter of 5cm. The leaves of birch trees were not included.

2.3.2 Tree components

In Marklund’s study the trees were always divided into several components for which a separate biomass function was developed. The following tree components were distinguished:

• stem over bark – stem under bark – stem bark • living braches

– branches (Swedish: grenaxlar) – needles • dead branches • stump-root-system – stump – roots ≥ 5cm – roots <5cm

This categorization shows that most tree components were even further subdi-vided into smaller sub-components (for example stem over bark as stem under bark and stem bark ). Marklund usually developed functions for each main tree compo-nent and also for each of its single sub-compocompo-nents. Only for the ”branches” (a sub-component of the ”living branches”) there was no single function developed, probably because it wasn’t considered to be particularly useful. Figure 2.3 gives an illustration how a spruce tree was divided into its stem, living branches and dead branches.

2.3.3 Study methods

(26)

Figure 2.3: A spruce tree and its above-ground components[18].

the sample compartments where the sample plots have been marked and the sample trees were collected from all over Sweden.

Within the sample plots the sample trees were measured and felled. A complete list of all measured or recorded variables can be found in Marklund’s report from 1987[18]. Afterwards the different tree components were gathered together on sepa-rate piles and weighted (”fresh weight”). To determine the moisture content of the wood some samples were selected and weighted in the laboratory. Not all the wood was weighted since the author assumed that there must a best a strong correlation between dry-weight wood and the fresh wood, so it was time-saving not to determine the weight of all samples. In the laboratory the samples were then further divided (e.g. stem wood and bark was separated) and then dried for 48 hours at 105◦ C to estimate their dry weights.

At the end of the study the regression analysis took place. First the diameter at breast height (DBH) was correlated to the dry weight of the tree components, since the DBH generally proved to have the highest correlation coefficient and therefore became the basic independent variable in the regression analysis. In order to further improve the regression functions the author introduced more independent variables (other dendrometric measurements from the sample trees) additionally to the DBH. In this way the standard error of the regression functions could be reduced, though the functions became more complex too. The ”best” function for spruce trees for example was obtained by using the model in equation (2.1), where DW is the dry-weight biomass, d is the DBH, h is the tree’s height and t is the tree’s age. β0 up to

(27)

(28)

(29)

ln(DW ) = β0+ β1∗

d

d + 14+ β2∗ h + β3∗ ln(h) + β0∗ ln(t) (2.1) The author created as well best-fitting functions (which often include a variety of measured variables), but also created simplified functions, which on the one hand yield a larger standard error, but on the other hand were easier to use since not as many variables had to be known. The ”simplest” function always used the DBH as the only independent variable. An example for the simplest function (1 independent variable) and the best function (8 independent variables) of the stem over bark biomass for pine can be found in figure 2.6.

(30)

An important fact that was already explained by Marklund is that the total biomass distribution of a tree on its single components depends on the DBH of a tree. In figure 2.7 this phenomena is illustrated for the Norway spruce. For example while with an increasing DBH the ”stem over bark” share compared to the total biomass also increases, the ”needles” share decreases. It is very important to keep this phenomena in mind, since the DBH of a tree should also increase through time, the composition of a tree’s components also changes through time.

Figure 2.7: The biomass distribution of single tree components depending on the tree’s diameter[18].

2.3.4 Revisions and extensions by Petersson and St˚ahl

In 2006 Hans Petersson and G¨oran St˚ahl from the SLU revised and extended the set of Marklund’s biomass functions by another set of functions for below-ground biomasses for spruce, pine and birch trees[20]. Until then the below-ground biomasses for the annual NFI were estimated by applying Marklund’s below-ground biomass functions for spruce and pine. Bu the authors claimed that Marklund’s functions were underestimating the below-ground biomass since a lot of roots were left in the ground when Marklund performed his experiment. The goal of Petersson’s and St˚ahl’s study was to calibrate Marklund’s data against new data by measuring roots down to 2mm diameter.

The new biomass data was collected in 2002. 12 forest stands (with 3 sample plots each) over Sweden were selected and trees were measured and felled according to Marklund’s methodology. Root samples were taken and dried and weighted in the laboratory. In total there were samples from 31 spruce trees, 34 pine trees and 14 birch trees (far less than in Marklund’s study).

(31)

Figure 2.8: The inventoried below-ground biomass according to Marklund (i) was complemented by measuring all major axes (d). A proportion of outer roots was excavated and used for derivation of regression functions to estimate the biomass of all outer roots[20].

create new below-ground biomass functions for spruce and pine, as well as (for the first time) below-ground biomass functions for birch trees. For each tree type they created functions for two cases: measuring the roots down to a 5mm- and down to 2mm-diameter. For spruce and pine they adapted Marklund’s model of creating different functions of different complexities (and therefore also different standard errors). Figure 2.9 yields the results for the below-ground biomass functions for pine. The new functions for birch trees only include the DBH as an independent variable.

(32)

2.4 Forest fuel and sustainable forest management

2.4.1 Definition of forest fuel

In the final report of the ”Swedish Lithuanian Wood Fuel Development Project” (a bilateral co-operation project between the Swedish Forest Administration and the Forest Department in Lithuania) the term forest fuel has been well-defined (see figure 2.10). Forest fuel is hereby defined as a sub-category of biofuel, while even forest fuel is further sub-divided into primary forest fuel, which are logging residuals, stem wood from final cuttings, pre-commercial and commercial thinning and wood without industrial demand or use, and by-products and reject from the wood industry (for example sawdust from sawmills). This definition was believed to be very suitable for this project, since it on the one hand clearly delimits forest fuel from other biofuels and on the other hand further explains how forest fuel can be sub-categorized to strengthen its definition.

Figure 2.10: Division of biofuel, wood- and forest fuels into sub-groups by origin of raw material and integration of flow[21]

2.4.2 Problem definition

(33)

biomass had to be modeled over time and Marklund’s biomass functions seemed to be an obviously suitable way to calculate the biomasses for all important tree species. The problem was that his formulas were all dependent on the diameter of a tree (DBH), but not on the age of a tree, which was essential if the biomass should be modeled over time. Since the kNN-Sweden data didn’t contain any information about the DBH of the trees, but only on their age, it seemed clear that it wasn’t possible to use Marklund’s functions directly. The first most obvious solution was to find a correlation between the DBH and the age of a tree, so that the biomass of a tree could be linked to its age. According to the correlation tables in Marklund’s study[18], the DBH showed the best correlation to the biomass of a tree (average by 0.83 for spruce), second came the height of a tree (average by 0.74 for spruce) and only third came the age of a tree (average only by 0.41 for spruce). Disregarding the fact that Marklund’s biomass functions couldn’t be used in any case since they are all based on the DBH but not on the age, the correlation table already gave an impression about how low their correlation was in reality. There might be many reasons for the low correlation between the DBH and the age of a tree: climate or soil conditions (e.g. nutrient balance), slopes, water/sunshine availability or the spatial distribution of the trees could all be reasons for a forest to grow inconsistent over time at different locations. Most of those unknown factors couldn’t even be modeled because the data wasn’t available (e.g. soil conditions). All these factors can influence the tree growth over time.

Therefore the idea was embraced to find simple and well-fitting regression curves by plotting the age (independent variable) and biomass (dependent variable) against each other. In order to do so it was important to find out how a managed forest is growing over time, i.e what happens to the forest over time and therefore its biomass. This will be explained in the following section.

2.4.3 Typical life cylce of managed forest

(34)

Regeneration

Year 0-10: In this period the natural regeneration is promoted by leaving some seeding trees behind when the forest is felled. If that is not possible, then reaf-forestation is necessary, which can be very expensive and therefore is not advised. The longer the forest owner waits to do reafforestation, the more expensive and difficult it will be to perform[22]. Another way is to grow new trees already years before the trees from the last generation are felled. This would even shorten the generation change of the forest (which made the forest economically more valuable), also because the trees can grow better in this way than on bare land. Within the around first 10 years the forest culture should be regularly checked to promote the forest. Within this time period no biomass is extracted from the forest.

Pre-commercial thinning / Cleaning

Year 10-25: This time period describes that some trees are either selectively or schematically (or both) chosen to be removed in order to promote the growth of other trees in the forest[22]. It is performed when it is economically reasonable, but usually when trees have reached a certain height. Shrubs are removed and trees are thinned out of the forest in order to keep trees within a distance from each other. Preferably bad, sick or bended trees are removed. Generally it is also advisable to remove trees who ”rob” to much light of other trees. Biomass is indeed extracted from the forest, but is is usually left in the forest because it is uneconomical to remove them because the costs are too high and the benefit is too low since the trees are still comparatively small and yield only few biomass.

Commercial Thinning

Year 25+: Commercial Thinning is performed around once every 5 to 10 years[22]. This operation aims to promote certain trees or tree species in the forest (usually those which are economically valuable and have a good height/diameter ratio). The promoted trees will therefore reach a higher thickness earlier. The credo is not to leave the forest growth to nature, but to ”interfere into the war among the trees[22]” and decide it for them.

Final felling

(35)

for forest stands[23], though they are rather unsuitable to identify a default value for the felling age of the forest:

• forest stands with at least 50% coniferous trees: 45-100 (depending on the province and site index, in V¨axj¨o: 45-90)

• forest stands with at least 50% are ash trees (deciduous): 50 • forest stands with at least 50% are beech trees (deciduous): 80

• forest stands with at least 50% are birch, aspen or alder trees (deciduous): 35 2.4.4 Sustainable forestry

(36)

Chapter 3

Study Area, Data Description

and Preparation

3.1 Study area

The study area for developing the GIS-tool was the municipality of Växjö which is located in the Swedish province (Swedish: län) of Kronoberg in the southern part of Sweden (see figure 3.1). The maximum spatial extents of the area are about 70km in North-South-direction and 45km in East-West-direction. The municipality had a total population of 83,005 people by the end of 2010[24], while most of the population lived in the city of Växjö (55,600 in 2005[25], around 67%), but also in several smaller cities like Rottne (2,224 inhabitants), Ingelstad (1,655 inhabitants), Bra˚as (1,567 inhabitants) or Lammhult (1,503 inhabitants). Around 63% of the municipal area is covered by forest, which therefore poses a huge renewable energy resource for the region. Already during the 1980’s a district heating system which was connected to a local CHP plant was established on the fringes of the city of Växjö, and its heating network is being extended until today to include most parts of the city. Since 2010 VEAB, which owns the power plant, even started to supply the city by district cooling. The local CHP plant named Sandvik runs several boilers which almost run on biofuel alone (mostly wood). Furthermore the small cities Bra˚as, Rottne, Lammhult and Ingelstad each run a near heating power station to supply local heat.

(37)

(38)

3.2 Data description and preparation

For any GIS-analysis it is most essential to have useful data at hand. That’s why at the beginning possible data sources and data availability needed to be researched. The goal was to preferably find data that was freely available and accessible. Also the legal framework had to be clarified, since it should be guaranteed that the data could be used for the project which has a non-commercial use.

Before any analysis could take place, some elementary decisions had to be made. The reference system of the project had been determined as SWEREF99. In gen-eral any reference system would have been possible to choose, but for practical rea-sons SWEREF99 was chosen since the digital maps from Lantm¨ateriet were already in that particular reference system (so no transformation had to be done). Also SWEREF99 (which has mostly replaced the older RT90 in Sweden) is almost equiv-alent to the worldwide geodetic reference system WGS84, and is the most widely used reference system in Sweden today. Second of all the cell size for the raster data had to be determined. Since the kNN-Sweden raster datasets already were in the size of 25x25 meters and that size was perceived as a reasonable resolution for the analysis, the same size was set as the default raster cell size for the project. This resolution provided a sufficient amount of detail, while also not being too coarse. It was important to keep the approximate size of the study area in mind, because the files also shouldn’t grow too large (which would make them difficult and long-winded for a GIS to handle), but on the other hand provide fine an acceptable resolution to yield a significant result. For the raster analysis it was also important that all raster datasets would lie exactly above each other, since that way new rasters would be exactly aligned with the old ones, and raster calculations would be kept most accurate.

In the following sections the data sources and the data preparation steps will be discussed.

3.2.1 Topographic Data (GSD: Geografiska Sverigedata)

Lantm¨ateriet1(the Swedish mapping, cadastral and land registration authority) pro-vides digital maps of Sweden in different levels of detail. For students and university stuff members in Sweden the data can be downloaded for educational and research purposes from Lantm¨ateriet’s Digital Map Library2 (Digitala Kartbibliotheket) af-ter an account has been regisaf-tered. It is only possible for researchers, teachers and students from an affiliated Swedish university (with an email address from a Swedish university) to register and download any data. To create an account the user must use a computer that is connected to a valid network, i.e. the user must be in the network of a Swedish university[26].

Also the data must not be used for any commercial purposes. The data volume that can downloaded per user and per month is limited. The user can decide to

1

http://www.lantmateriet.se/

(39)

Direction Border value North 6,342,810.554m South 6,273,789.196m East 514,395.462m West 468,073.236m

Table 3.1: Coordinates of the maximum geographical extents of the municipality of V¨axj¨o (in SWEREF99 TM).

download specific map layers in different formats, so for example topographic maps can be downloaded as ESRI shapefiles (vectorized format) or TIFF (raster format), or landuse maps are available as ESRI GRIDs (raster format).

On Lantm¨ateriet’s website one can find Swedish documentations of the vector data in pdf-format. English documentations (also as pdf) were formerly deleted from website, but are still available and can be requested via phone or mail from Lantm¨ateriet.

Since the project was going to be performed on a very local scale, a very detailed topographic map that includes all necessary details such as access roads would be most suitable, but also height data or landuse maps might come in handy. Also the data should be downloaded with a reasonable buffer around the study area, so that for example roads that in fact are located outside of the study area but still might influence the study area by its proximity would also be contained in the dataset. Another reason for having a buffer is that forest stands that partially lie outside of the study area will also be fully covered by the data (since they don’t usually end at administrative borders). So the buffer should be chosen large enough so that the whole study area will be covered (since some forest stands might be rather longish and can lie further outside of the study area than expected).

The following datasets were downloaded from Lantm¨ateriet’s Digital Map Li-brary3 and used for this project:

• Topographic map 1:50,000 (Swedish: Terr¨angkarta) • Height data (Swedish: H¨ojddata)

Before downloading the data the maximum geographical extents of the munic-ipality of V¨axj¨o had to be determined. Table 3.1 contains the coordinates in the SWEREF99 TM coordinate system. These coordinates describe a geographic rect-angle with a north-south-extent of 69.021km and an east-west-extent of 46.322km.

(40)

Direction Border value

North 6,344,000m (Buffer: 1,189.446m) South 6,272,000m (Buffer: 1,789.196m) East 516,000m (Buffer: 1,604.538m) West 467,000m (Buffer: 1,073.236m)

Table 3.2: The geographical rectangle’s border coordinates of the downloaded data (in SWEREF99 TM)

border values would at least include that minimum buffer of 1000 meters. Those new border values can be found in table 3.2. This time a geographic rectangle with a north-south-extent of 72km and an east-west-extent of 49km was created.

While the height data and land cover data could be downloaded as one single file from the server, the topographic map had to be downloaded in 6 ”stripes” because the file size of the whole extent exceeded the maximum size for a single download-inquiry. That’s why the area was divided into 6 equal rectangular zones (along north-south direction) by using the following intermediate values to limit the extents in their North and South direction: 6,272,000m (maximum southern extent), 6,284,000m, 6,296,000m, 6,308,000m, 6,320,000m, 6,332,000m, 6,344,000m (maximum northern extent). Later in the project the necessary layers were merged back together.

Useful data layers

Not all the layers in the downloaded data were useful to the project. The following list is an overview of which data layers were used in the project and why they were considered as important:

• Administrative borders (from ’Topographic map’). The administrative borders of the municipality of V¨axj¨o were most crucial since they shaped the spatial borders of the study area.

• Roads (from ’Topographic map’). Roads could be used to find out how good the forest is accessible for performing forestry. Unfortunately the dataset of Lantm¨ateriet mostly only contained public roads, but not specifically forest roads (which are often created or used specifically for forestry inside the forest). Since that data was probably missing this might have lead to the assumption that plenty of forest couldn’t be accessed over roads, which is very likely not even the case (cause usually all larger managed forest areas should already have been made accessible for forestry machines).

(41)

there is to be protected and left unharmed. For the analyst however those areas also yield information about in which areas the forest grows naturally (because that forest is not managed), and therefore should be excluded from the forest dataset during the analysis of the managed forest’s growth.

• Water bodies and rivers/streams (from ’Topographic map’). Water is an ob-stacle for forestry machines when accessing the forest, so this data can be used to find out which forest parts are blocked by water and are therefore hard or impossible to access.

• Heights (from ’Height data’). The height data, also called Digital Eleva-tion Model (DEM), is very useful data to derive slopes for the study area. Those slopes can describe which areas are difficult or impossible to access with forestry machines.

The following sections explain the steps of some basic data preparation that was carried out with the layers mentioned above.

Extract municipal administrative borders of V¨axj¨o

1. Borderlines that didn’t mark the municipal border of Växjö had to be deleted. Note: This step wouldn’t be necessary if no other closed polylines shapes existed rather than the ones marking the municipal borders of Växjö, i.e. no other polygon than the municipality of Växjö could be built from it.

2. Merge all the polylines into one single polyline. Note: Since each file (of the 6 ”stripes”) carried the same name they had to be renamed first.

3. Convert polylines into a polygon. The result was only one polygon that covered the study area.

Extract forest access roads

(42)

Extract natural habitats

The topographic data also included a polyline layer which marked the borders of natural habitats. Those areas should be identified and extracted. Many different kinds of natural habitats were described in the documentation of the file, but at last all areas of that particular layer were assumed to describe unmanaged forest and were therefore selected.

Starting with the polylines first of all polygons were built to have geometries that cover those areas (see sample in figure 5.1). Afterwards those features were converted into a raster dataset (always with the default resolution of 25x25m) and later reclassified to a single value (default: 1), so that all the areas would be all treated as uniform.

Compute slopes from height data

The height data (DEM) from Lantm¨ateriet was used to derive slopes for the region. For the project the preferred unit for the slopes was percent, since this unit is mostly used in reality.

3.2.2 Forest coverage data: kNN-Sverige (kNN-Sweden)

The kNN-Sweden forest dataset is a result of a method developed by the Remote Sensing Laboratory (Forest Resource Management) at the Swedish University of Agricultural Sciences (SLU) in Ume˚a which is using the k- Nearest Neighbor (kNN) algorithm to interpret satellite data into information about the forest. The outputs are ”continuous estimates of specified forest parameters, such as total wood volume, wood volume by tree species, stand age, and above-ground tree biomass”[27]. There are datasets for 2000 and 2005 freely available to download on the SLU website4.

kNN-Sweden was a 2-year-project that was started after several smaller projects using SPOT satellite images to create forest maps for several Swedish provinces. In the project field data plots of theNational Forest Inventory (NFI) (an annually created countrywide forest inventory based on field samples of forest variables which are based on a systematic grid across the country consisting of sample plots with classified locations), satellite images (LANDSAT 5 and LANDSAT 7 for the data from 2000, SPOT 4 and SPOT 5 for the data from 2005) and data from digital maps (from Lantmäteriet, see 3.2.1) were combined[28]. A ”forest mask” is created from the using Lantmäteriet’s 1:100 000 topographic map (Swedish: vägkarta) by choosing three different classes that all count as forest: forest (the only class available in Växjö), mountain forest (not available in Växjö) and forested wetland (also not available in Växjö). This mask determines the areas that need to be interpolated using the kNN-method.

The kNN-method then interpolates for every pixel on the satellite image un-der the forest map mask using the variables from the NFI-plots. The values are

(43)

calculated through weighting the k- nearest sample plot by their inverse euclidean distance (see figure 3.2).

Figure 3.2: Illustrated example of applying the kNN-interpolation-method on an unknown forest-pixel by using the 5 closest NFI-plots[28].

There have been different studies that proposed different numbers for k. The higher k is, the more the pixel-level results will average towards the mean. In the countrywide estimates for Sweden, k = 15 has been chosen[28].

The output dataset yields forest variables for the whole forest coverage of Sweden. Only small data samples might be missing due to clouds on the satellite image.

Table 3.3 gives an overview over the different variables that are contained in the kNN-Sweden dataset from 2005, which has been used in this project. The raster datasets have a resolution of 25x25 meters per pixel.

The biomass values (which have been added to the dataset only by 2010) are given in kilogram per hectare, and are the sum of three different variables that are estimated by the Swedish National Forest Inventory: Dry weight of stem over bark, Dry weight of branches including needles, Dry weight of stump and roots over 2mm. Those were calculated by Marklund’s and Petersson’s biomass functions (see 2.3).

The volumes are given in forest cubicmeter (m3f o) per hectare (Swedish: skogsku-bikmeter (m3sk)), which describes the solid volume of a standing tree, though only includes the stem over bark component of the trees, and excludes branches and nee-dles as well as the stump-root-system. The volumes were calculated according to the tree volume functions by Manfred N¨aslund from 1947[29]. The age of the forest is given in whole years.

(44)

Description Data Type Unit

Age Raster years

Biomass Raster kg/ha

Height Raster m

Beech volume Raster m3f o/ha Birch volume Raster m3f o/ha Concorta volume (not available for V¨axj¨o) Raster m3f o/ha

Oak volume Raster m3f o/ha

”Other coniferous forest” volume Raster m3f o/ha

Pine volume Raster m3f o/ha

Spruce volume Raster m3f o/ha Total volume Raster m3f o/ha

Forest stands Polygon

-Table 3.3: Overview over the datasets which are contained in the kNN-Sweden dataset from 2005.

the preferred size was aimed to be near the mean area of a clearcut in the particular region, which in V¨axj¨o should be around 2.5 ha. After defining the stands geometry the mean values of all the kNN-pixels that falls inside each segment for each variable have been calculated.

Apart from economic or management reasons, the clustering of the forest into forest stands still has another benefit: it additionally increases the accuracy of the forest variables. The accuracy of that data on the pixel-level is very low, while the accuracy on a stand-level is significantly better[28]. In an experimental area the mean error in the variable ”total volume” levels out between 5 and 10 percent when the investigated area reaches more than 100 ha (see figure 3.3).

In March 2011 the kNN-Sweden dataset from 2005 has been updated once again and the files were corrected due to ”problems with clear cuts” (citation from kNN-Sweden website), which probably refer to damages caused by the Gudrun storm in 2005, since the Swedish NFI underestimated the harvests from the field plots. This corrected dataset has not been used for the project since it was published after the start of this thesis research. The modeling results using the uncorrected data would even though correspond to a regular growth scenario.

(45)

Figure 3.3: Standard error of kNN-estimates of the total wood volume in the exper-imental area of Remningstorp (V¨astra G¨otaland)[28].

Transform data to SWEREF99

Since the whole kNN-Sweden datasets were only available in the old RT90 reference system, they had to be transformed to SWEREF99 first. The following steps were performed:

1. Due to some inconsistencies in the kNN-data the projection for all raster datasets had to be defined as RT90 to make sure that ArcGIS would rec-ognize all datasets to be in the same reference system, so that they could all be transformed to SWEREF99 in the same way later. Some of them were in the Bessel 1941 system which uses the same ellipsoid with RT90, so the projection didn’t result into any geometric change of the data.

2. Re-import the raster datasets with a transformation from RT90 to SWEREF99, which could be done automatically by a pre-defined transformation method in ArcGIS (SWEREF99 TO RT90 method; in the opposite case parameters will be automatically inverted). Note: a complete list of all geographic transforma-tions that are already included into ArcGIS can be found in the ArcGIS instal-lation directory (ArcGIS/Desktop10.0/Documentation/geographic transforma-tions.pdf). Because of the difference of both systems there cannot be an exact transformation between RT 90 and SWEREF99. The transformation yields a mean error of 10-15 cm and a maximum error of 15-25cm ([30]), which is fortunately not significant for the accuracy of this project.

(46)

Extract kNN-datasets for the municipality of V¨axj¨o

Since the data so far covers the whole province of Kronoberg, the areas that covers the municipality of V¨axj¨o still had to be extracted. In order to do so the kNN-data was intersected with the polygon that covers the municipality (from topographic data in 3.2.1). Since the forest stands (polygon features) were usually not ending at the municipal borders, that procedure meant that all features that somewhere intersected with the polygon of the municipality were selected. This meant that also forest stands that partially lay outside of the municipality were selected. Since those polygons could have been either included or not included, it was decided to include them in the data. It should therefore be noted that this operation slightly overestimates the results for the forest wood potentials later. Figure 3.4 shows a map of a small sample area of the region that displays the biomass variable in the original 25x25 meter raster resolution with overlapped forest stand borders.

Figure 3.4: kNN-Sweden original forest biomass data near Bra˚as, municipality of V¨axj¨o.

Compute mean values for forest stands

(47)

Mean age Raster years

Mean biomass Raster kg/ha

Mean height Raster m

Mean beech volume Raster m3f o/ha Mean birch volume Raster m3f o/ha Mean oak volume Raster m3f o/ha Mean ”Other coniferous forest” volume Raster m3f o/ha Mean pine volume Raster m3f o/ha Mean spruce volume Raster m3f o/ha Mean total volume Raster m3f o/ha

Stand area Raster ha

Table 3.4: Overview of newly created raster datasets on a forest stand level.

”forest stands” has been created and added to the kNN-Sweden data package. Also in order to make the model more applicable for real-life application (since forests are usually not managed on a pixel level), it was decided that all the forest variables should be built and regarded on a stand level. To create those raster datasets which contained mean values for each forest stand the following number of steps has been carried out:

1. The polygons from forest stands shapefile that intersect with (so not necessarily completely lie within) the municipality of V¨axj¨o were extracted.

2. The Zonal Statistics tool in ArcGIS was used to ”rasterize” the forest stand polygons which contains one statistical value that is built from the input raster dataset for each of the overlaying forest stands (”zones”). As overlaying zones the forest stand polygon shapefile was selected, which was used together with each raster dataset from the kNN-data to deliver a different statistical value for the new ”rasterized” forest stands, for example the mean pine volume of the forest stand. Therefore a raster dataset for each variable was created (see table 3.4).

3. Furthermore the area of each forest stand polygon is computed (unit: hectare) and exported as a raster in the same resolution.

A sample of that result can be viewed in figure 3.5, where the biomass variable is displayed for the area around Bra˚as.

About the biomass values

(48)

(49)

Marklund 1987 Marklund 1988 Petersson 2006

Spruce above stump ! !1 %

below stump % ! !2

Pine above stump % ! %

below stump % ! !2

Birch above stump % ! %

below stump % % !3

Table 3.5: Available and used biomass functions in kNN-Sweden for the most common tree types in Sweden.

1 _{revised formulas (no correction factor needed unlike 1987)} 2

revised formulas (former underestimation assumed)

3

not used in kNN-Sweden (due to large bias), instead used ”spruce below stump” functions

the kNN-Sweden the biomass functions from Lars Gunnar Marklund (see 2.3) and Petersson & St˚ahl (see 2.3.4) were used. Table 3.5 gives a simple overview over which biomass functions are available and which functions were used in the Swedish NFI (and therefore the kNN-Sweden). It shows that Marklund’s biomass functions from 1988 were used for above-ground biomasses for all three tree species in the kNN: spruce functions were used for spruce trees, pine functions for pine trees and birch functions for all broadleaved species. For below-ground biomasses Petersson’s biomass functions (down to 2mm root diameter) were used. Again spruce functions were applied to spruce and pine functions were applied to pine. For birch (and all other broadleaved trees) the below-ground functions for spruce were used, since the functions for birch were only based on 14 sample trees (see 2.3.4), which was considered as too inaccurate since the bias would have been too large if it was extrapolated.

(50)

Forest cover

Another small operation which was performed with the kNN data was to create a raster dataset that yielded the total forest cover of the study area. In order to do this one of the rasters that has been created by the Zonal Statistics tool before was used (to guarantee that the pixels of the new dataset would exactly correspond to the raster files that are later used in the model), for example the mean total volume, and reclassified (default: 1) to get the total forest cover of V¨axj¨o. This was important so that the forest share of study area could be computed.

3.2.3 Energy balance for V¨axj¨o

(51)

(52)

Chapter 4

Methodology

4.1 Modelling the forest fuel potential

Figure 4.1 gives an overview over the planned workflow to model the forest fuel potentials. External data is hereby colored red/orange, processes are colored in green, created output data is held in blue and process/model parameters are kept in purple.

In total there are five processes or modeling steps. The kNN-data hereby takes over the leading role over the whole project, since the first process describes the classification of the forest data into three forest types (coniferous, mixed and decid-uous), which will be discussed in detail in 4.2. Afterwards the forest growth over time will be modeled, since the goal of the study is to predict forest fuel potentials into the future. The process to obtain the forest growth rates will be explained in 4.3. Next the first of the three biomass potentials, the theoretical potential, will be modeled. Every biomass potential model will also contain a set of parameters, which all have certain values by default, but can be modified at will. Please refer to 4.4 for an extensive description of the modeling process. The technical potential, which uses some of the topographic data to evaluate the data from the theoretical potential against topographic features, is modeled next and is presented in 4.5. Lastly the reduced technical potential describes the final energy by using efficiency values from the energy balance for the region. Please refer to 4.6 for details.

(53)

(54)

time, but rather to estimate timber production or to optimize the forest management under different requirements. All those models are naturally more complex than the model developed in this thesis, because they have a long development history and are used professionally. But their accuracy in the prediction of forest fuel over time is unknown, since it is a new application field and those systems were not originally designed for this purpose.

The forest biomass study in Portugal [12], which estimates wood logging residuals for the whole country, already follows a similar approach than the method from RSA. It differentiates between the theoretical biomass potential and the available biomass potential and gives results for both of these potentials. But the available biomass potential in this study is still completely different from the technical potential in this study, since it the only technical restriction is the distance of the forest to a power plant that can utilize the wood. Even though this study also uses NFI forest data to estimate the potentials, and aggregates the results for each Portuguese provinces. This implicitly assumes that the forest in the area is evenly distributed.

4.2 Forest classification

As the first step towards modeling the forest fuel potential the decision was made to simplify the model by categorizing the forest into three classes: coniferous, mixed and deciduous forest. This simplification was made because forest stands then could be allocated into one of these three classes, which would be easier to handle. RSA also used this classification in the ”EnergieRegion Rhein-Sieg” study[13]. Also it was assumed at that time that forest would best be categorized in those classes because the forestry methods for those forests types might differ. But while this classification method on the one hand simplified the model, it also made it less accurate on the other hand because the classification also meant a generalization of the data. The problem that had to be solved was how to define the thresholds for those forest classes. Skogsstyrelsen itself refers to as coniferous forest if at least 7 out of 10 trees are conifer (so 70%)[23]. Influenced by this value the default border for a coniferous forest was finally set at 75% (three quarters).

A model was created with the ArcGIS ModelBuilder which used the zonal kNN-datasets (see 3.2.2) as input kNN-datasets to reclassify the forest into the three defined categories depending on the default threshold value of 75%. The threshold is set as a parameter, which means it can be modified by the user to another value in case the user would like to change it. The model basically checks, according to the volumes of all tree species on a pixel according to the allocation table 4.1, what percentages a forest type on each pixel holds in total and then decides which forest type that pixel belongs to. Since all pixels within one forest stand have the same value, always whole forest stands are being categorized to belong to one specific forest type. After the classification the model also creates datasets which carry the biomass values of each stand.

(55)

Description Forest Type

Beech deciduous

Birch deciduous

Oak deciduous

”Other coniferous forest” deciduous

Pine coniferous

Spruce coniferous

Table 4.1: Allocation of tree species to tree types.

while table 4.2 contains the input datasets of the model and table 4.3 lists all the output datasets of the model. Finally table 4.4 holds the parameters of the model, which are merely the two threshold values for the classification of forest into conif-erous and deciduous forest (all other forest is treated as mixed forest).

Figure 4.2: Model of the forest classification in ArcGIS ModelBuilder.

4.3 Modeling forest growth

(56)

Mean biomass Raster kg/ha

Mean beech volume Raster m3f o/ha Mean birch volume Raster m3f o/ha Mean oak volume Raster m3f o/ha Mean ”other coniferous forest” volume Raster m3f o/ha Mean pine volume Raster m3f o/ha Mean spruce volume Raster m3f o/ha Mean total volume Raster m3f o/ha

Table 4.2: Input datasets for the forest classification model.

Description Data Type Default Value / Unit Coniferous forest coverage Raster 1

Mixed forest coverage Raster 1 Deciduous forest coverage Raster 1 Coniferous forest biomass Raster kg/ha Mixed forest biomass Raster kg/ha Deciduous forest biomass Raster kg/ha

Table 4.3: Output datasets for the forest classification model.

Description Parameter type Default value Minimum share of conifer trees for coniferous

forest

Classification 75%

Minimum share of conifer trees for coniferous forest

Classification 75%

(57)

Description Data Type Default Value / Unit Coniferous forest stands cover Raster 1

Mixed forest stands cover Raster 1 Deciduous forest stands cover Raster 1 Natural habitats Raster 1 Forest age Raster years Forest biomass Raster kg/ha

Table 4.5: Input datasets to model congruent managed forest raster datasets.

4.3.1 Determine managed forest

For this study it was assumed that managed forest refers to all the forest that is not in a natural habitat zone. On the one hand this assumes that all other forest within the study would be managed (which might not be the case in reality), and on the other hand also assumed that forest within a natural habitat wouldn’t be managed (which also might not be true). Nevertheless it was decided that for the study this was a good approach to extract the managed forest of the region.

The above definition meant that all the forest that lies inside a natural habitat had to be excluded from the forest datasets to create datasets that only contained the ”managed” forest. In 3.2.1 a raster dataset which includes all natural habitats has been already created. ArcGIS was used again to create a model in the ModelBuilder (see figure 4.3) which contained several input datasets (see table 4.5) and output datasets (see table 4.6). The goal of this model was to create datasets for each forest type which contained the age and the biomass variable. The model checks which pixels of the forest areas for each forest type do overlay with a pixel that belongs to a natural habitat and excludes those pixel from the forest dataset. Those datasets could then be exported as text-files (in ASCII format) to be read into a statistical analysis software such as SPSS.

The input datasets were the raster dataset containing the natural habitats, the kNN-datasets containing the age and biomass information as well as the forest stands coverage datasets for each forest type. The output datasets contained the age and biomass values as well as the plain forest type coverage information for each managed forest type.

A GIS Model to Estimate a Sustainable Potential of Forest Fuel for Energy Generation in the Municipality of Växjö, Sweden