Archetypical patterns and trajectories of land systems in Europe

116 Pages·2015·14.44 MB·English

Manuscript Click here to view linked References

Archetypical patterns and trajectories of land systems in Europe
1
2
3
4
5
6
7 Authors: Christian Levers 1,*, Daniel Müller 1,2,3, Karlheinz Erb 4, Helmut Haberl 3,4, Martin
8
9 Rudbeck Jepsen 5, Marc J. Metzger 6, Patrick Meyfroidt 7,8, Tobias Plieninger 9, Christoph Plutzar 4,
10
11
12 Julia Stürck 10, Peter H. Verburg 10, Pieter J. Verkerk 11, Tobias Kuemmerle 1,3
13
14
15
16
17
1 Geography Department, Humboldt-Universität zu Berlin
18
19
20 Unter den Linden 6, 10099 Berlin, Germany.
21
22
23 2 Leibniz Institute of Agricultural Development in Transition Economies (IAMO)
24
25
Theodor-Lieser-Str. 2, 06120 Halle (Saale), Germany.
26
27
28
3 Integrative Research Institute on Transformations of Human-Environment Systems (IRI THESys),
29
30
31 Humboldt-Universität zu Berlin
32
33
Unter den Linden 6, 10099 Berlin, Germany.
34
35
36 4 Institute of Social Ecology Vienna, Alpen-Adria Universität Klagenfurt, Wien, Graz
37
38
39 Schottenfeldgasse 29, 10 1070 Vienna, Austria.
40
41
42 5 Section of Geography, Department of Geosciences and Natural Resource Management, University
43
44 of Copenhagen
45
46
Øster Voldgade 10, DK-1350, Copenhagen K, Denmark.
47
48
49
6 School of GeoSciences, University of Edinburgh, Edinburgh
50
51
52 Drummond Street, Edinburgh EH8 9XP, Scotland, UK.
53
54
55 7 Georges Lemaître Centre for Earth and Climate Research, Earth and Life Institute, Université
56
57
catholique de Louvain
58
59
60 Place Louis Pasteur 3, B-1348 Louvain-La-Neuve, Belgium.
61
62
1
63

64
65
8 Fonds de la Recherche Scientifique – FNRS
1
2 Rue d’Egmont 5, B - 1000 Brussels, Belgium.
3
4
5 9 Department of Geosciences and Natural Resource Management, University of Copenhagen
6
7
Rolighedsvej 23, 1958 Frederiksberg C, Denmark.
8
9
10
10 Institute for Environmental Studies, VU University Amsterdam
11
12
13 De Boelelaan 1087, 1081 HV Amsterdam, The Netherlands.
14
15
16 11 European Forest Institute (EFI), Sustainability and Climate Change Programme
17
18
Yliopistokatu 6, 80100 Joensuu, Finland.
19
20
21

22
23
24 * Corresponding author. Unter den Linden 6, 10099 Berlin, Germany; tel.: [phone]; fax:
25
26
27 [phone]; e-mail address: [email protected] (C. Levers).
28
29
30
31
32
Email addresses of co-authors:
33
34
35
DM: [email protected], KHE: [email protected], HH: [email protected],
36
37
38 MRJ: [email protected], MJM: [email protected], PM: [email protected],
39
40
TP: [email protected], CP: [email protected], JS: [email protected],
41
42
43 PHV: [email protected], PJV: [email protected], TK: [email protected]
44
45
46
47
48
49 Short title: Archetypes of land systems in Europe
50
51
52 Word count abstract: 260
53
54
55 Word count manuscript: 6785
56
57
58
59
60
61
62
2
63

64
65
1
2 1 Abstract
3
4
5 2 Assessments of land-system change have dominantly focused on conversions among broad
6
7
3 land-use categories, whereas intensity changes within these categories have received less attention.
8
9
10 4 Considering both typically results in diverse patterns and trajectories of land-system change,
11
12
5 requiring approaches to reduce this complexity. Using Europe as a case study, we applied a
13
14
15 6 clustering approach based on self-organising maps and 12 land-use indicators to map (i) land-system
16
17
7 archetypes for the year 2006, defined as characteristic patterns of land-use extent and intensity, and
18
19
20 8 (ii) archetypical change trajectories, defined as characteristic changes in these indicators between
21
22
9 1990 and 2006. Our analysis identified 15 land-system archetypes, with low-intensity archetypes
23
24
25 10 dominating (ca. 55% coverage) followed by high-intensity archetypes (ca. 26%). In terms of change,
26
27 11 we identified 17 archetypical change trajectories, clustered in four broad categories. Stable land
28
29
12 systems were most widespread (ca. 40% of the EU27), followed by land systems characterised by
30
31
32 13 land-use conversions (ca. 26%), de-intensification trends (ca. 18%), and intensification trends (ca.
33
34
14 15%). Intensively used and intensifying land systems were particularly widespread in Western
35
36
37 15 Europe, whereas low-intensity and de-intensifying land systems dominated in Europe’s east.
38
39
16 Comparing our archetypes with environmental and socioeconomic factors revealed that favourable
40
41
42 17 accessibility, topographic, climatic, and soil conditions characterised intensively managed areas.
43
44 18 Intensification was also most common in these areas, suggesting an ongoing polarisation of
45
46
19 intensification in favourable areas and de-intensification and abandonment trends in more marginal
47
48
49 20 areas. Our archetypes could become useful templates for assessing ecosystem services, for exploring
50
51
21 the trade-offs of land-system changes, as well as for developing context-specific land-management
52
53
54 22 policies to steer European land systems onto desired pathways.
55
56
57 23 Keywords: land-system change; land-use intensity; Europe; land management; landscape; automated
58
59 24 clustering
60
61
62
3
63

64
65
25 1 Introduction
1
2
3 26 Humans have affected more than 75% of the earth’s ice-free surface by either land management
4
5
27 or land conversions (Luyssaert et al. 2014, Ellis et al. 2010) making land use a major force of global
6
7
8 28 environmental change (Haberl et al. 2007). Land use can change due to given demands for land-
9
10
29 based products in two general ways: (i) conversions among broad land-cover/use categories (e.g.,
11
12
13 30 deforestation, farmland abandonment, urban expansion), leading to changes in the extent of these
14
15
31 classes, and (ii) changing management practices (e.g., changes in mechanisation, fertiliser
16
17
18 32 application, or wood harvesting), resulting in intensity changes within these broad land-cover/use
19
20 33 categories. Both types of changes are commonly interlinked, resulting in complex patterns and
21
22
34 trajectories of land-system change. Understanding this complexity is important to design and
23
24
25 35 implement context-specific, effective policy measures (Rounsevell et al. 2012, Foley et al. 2011), but
26
27
36 also to assess the impacts of land-system changes on biodiversity and ecosystem services.
28
29
30
37 The characterisation and mapping of typical land-change patterns and trajectories that consider
31
32
33 38 both changes in extent and intensity across sectors and at the level of the land system as a whole is a
34
35
39 powerful tool for understanding the complexity of land-system changes. Such an approach allows for
36
37
38 40 identifying “syndromes” or “archetypes” of land-system change, which describe unique land-change
39
40 41 patterns or processes that occur repeatedly across space. Archetypes facilitate a more integrative
41
42
42 understanding of land-system changes, and, when combined with driving factors of land change,
43
44
45 43 provide deep insights into change trajectories, some of which may remain uncovered if area extent,
46
47
44 intensity, and driving factors are studied in isolation from each other (Müller et al. 2014).
48
49
50
45 Despite calls for such an integrative analysis (Verburg et al. 2009) and the growing recognition
51
52
53 46 of the importance of land-use intensity (Luyssaert et al. 2014, Erb et al. 2013, Lambin et al. 2000),
54
55
47 most studies to date focussed on individual land-change processes, often conversions only, thereby
56
57
58 48 neglecting feedbacks between land-use sectors (e.g., Estel et al. 2015, Kuemmerle et al. 2015,
59
60
61
62
4
63

64
65
49 Hansen et al. 2013, Hatna and Bakker 2011, Kaplan et al. 2009, Feranec et al. 2007). Furthermore,
1
2 50 most studies focus on local scales but assessments of land-system changes from landscape to
3
4
51 regional to continental scales are also important because these are the scales predominantly targeted
5
6
7 52 by policy making in agriculture, forestry, and nature conservation sectors (e.g., in the European
8
9
53 Union). The few existing studies that have taken a more holistic approach to mapping land systems
10
11
12 54 on broad scales have been restricted to one point in time (Václavík et al. 2013, van Asselen and
13
14
55 Verburg 2012, Ellis and Ramankutty 2008), thereby neglecting land change, or solely focussed on
15
16
17 56 land-cover change, but did not include information on land-use intensity (Stellmes et al. 2013, Hill et
18
19 57 al. 2008). What is needed are analyses that (i) jointly consider area and intensity changes, (ii) include
20
21
58 multiple sectors (e.g., agriculture, forestry), and (iii) at spatial resolutions and extents relevant for
22
23
24 59 policy-making. Unfortunately, such an assessment has so far not been carried out for any region in
25
26
60 the world.
27
28
29
61 Europe is an interesting case to study land-system changes as its large environmental, political,
30
31
32 62 and socio-economic heterogeneity resulted in a diversity of land systems and multi-facetted land-
33
34
63 change pathways (Vos and Meekes 1999, Jepsen et al. 2015, Fuchs et al. 2015). Europe has also
35
36
37 64 experienced a period of marked land-use change recently, including both changes in the extent and
38
39
65 intensity of agriculture and forestry (Jepsen et al. 2015, Rounsevell et al. 2012). For example, the
40
41
66 breakdown of the Soviet Union and the subsequent eastward expansion of the EU triggered
42
43
44 67 widespread land-use change (Munteanu et al. 2014, Kuemmerle 2008), both in agriculture (Griffiths
45
46
68 et al. 2013b, Müller et al. 2009) and forestry (Griffiths et al. 2013a, Ellis et al. 2010, Kuemmerle et
47
48
49 69 al. 2007). Furthermore, changes in policy instruments (e.g., the Common Agricultural Policy)
50
51
70 markedly influenced Europe’s land system (Donald et al. 2002). Unfortunately, the spatial patterns
52
53
54 71 and trends of these land-system changes remain only partly understood.
55
56
57 72 Our goal was to identify and map Land-System Archetypes (LSA) as well as Archetypical
58
59 73 Change Trajectories (ACT) of land systems on a 1 km2 grid for the EU27 between 1990 and 2006.
60
61
62
5
63

64
65
74 Throughout this study, we understand land systems as a combination of land cover and land-use
1
2 75 intensity patterns where the elements are linked through systemic interactions (following van
3
4
76 Asselen and Verburg 2012). Our definition of land systems assumes that the co-occurrence of
5
6
7 77 recurring, distinguishable combinations of land cover and land-use intensity patterns reflects their
8
9
78 systemic interactions, though this cannot be directly observed. Within this, we define land use as the
10
11
12 79 socioeconomic activities with which humans utilise land cover (Lambin et al. 2006) by management
13
14
80 practices that can be characterised by different degrees of intensity. For the purpose of this paper, we
15
16
17 81 define our archetypes as a regularly appearing and distinguishable combination of land cover and
18
19 82 land-use intensity patterns (LSAs) or changes (ACTs) that are linked through systemic interactions.
20
21
83 We used self-organising maps (SOMs), an unsupervised clustering technique, to map LSAs and
22
23
24 84 ACTs. SOMs reduce high-dimensional data by grouping observations based on their similarity in
25
26
85 terms of features and locations and are hence highly suited for our approach. We used the resulting
27
28
29 86 archetypes for assessing land-system change and to compare observed changes with a range of socio-
30
31
87 economic and environmental factors (hereafter referred to as “explanatory factors”) that are known to
32
33
34 88 drive land-system change. Our study goes beyond existing EU-wide typologies that focused on
35
36
89 characterising landscapes or environmental conditions (van Eupen et al. 2012, Brus et al. 2012,
37
38
39 90 Hengeveld et al. 2012, Duncker et al. 2012, Hazeu et al. 2011, Mücher et al. 2010, Westhoek et al.
40
41 91 2006, Metzger et al. 2005, Meeus 1995) by explicitly focussing on land-system change and by
42
43
92 incorporating land-use intensity metrics. Specifically, we ask the following research questions:
44
45
46
93 1. Which are archetypical patterns and change trajectories in Europe’s land system?
47
48
49 94 2. How do land-system patterns and changes relate to each other?
50
51
95 3. What characterises archetypical patterns and change trajectories in terms of key explanatory
52
53
54 96 factors of land change?
55
56
57
58
59
60
61
62
6
63

64
65
97 2 Material and methods
1
2
3 98 We compiled a set of 12 indicators representing the extent of broad land-use categories and the
4
5
99 management intensity within these categories pertaining to agriculture and forestry for the entire
6
7
8 100 EU27 and the years 1990 and 2006 (section 2.1 and Table A1). We used self-organising maps
9
10
101 (Kohonen 2001) to derive LSAs for the year 2006, using indicators of land-use extent and intensity
11
12
13 102 from that year, as well as ACTs between 1990 and 2006 (see section 2.3 for the calculation of
14
15
103 indicator change values, Figure 1). Subsequently, we reviewed, refined, and labelled the outcomes of
16
17
18 104 the SOM clustering in an expert workshop for both, LSAs and ACTs. We overlaid the resulting
19
20 105 archetypes with 14 explanatory factors of land change (section 2.2 and Table A2). Whenever
21
22
106 possible, we gathered data at a spatial resolution of 1 km2; otherwise we relied on data at the NUTS-
23
24
25 107 3 level (Nomenclature des unités territoriales statistiques; i.e., Nomenclature of Territorial Units for
26
27
108 Statistics).
28
29
30
109 << Figure 1 approximately here >>
31
32
33
34
110 2.1 Land-use indicators
35
36
37
38 111 2.1.1 Extent of broad land-use categories
39
40 112 We used a harmonised dataset on land use in Europe on a 1 km2 grid for the time period between
41
42
113 1990 and 2006 (see Plutzar et al. 2015). Specifically, our dataset was generated using CORINE
43
44
45 114 (Coordination of Information on the Environment) land-cover maps, sub-national forest data, and
46
47
115 CAPRI (Common Agricultural Policy Regionalised Impact) data on biomass production in NUTS2
48
49
50 116 regions related to cropping, grazing, and forestry in an additive, closed-budget approach. From these
51
52
117 datasets, we derived information on the extent of (i) built-up and infrastructure, (ii) cropland (arable,
53
54
55 118 permanent, and fallow), (iii) forests and other wooded land, as well as (iv) grazing land (e.g.,
56
57
119 meadows, pastures) for both time steps. See Text A1 in the Supplementary Material A for details.
58
59
60
61
62
7
63

64
65
120 Fallow farmland and farmland abandonment is not covered well by CORINE. To characterise
1
2 121 fallow farmland, we used a series of maps generated from Moderate Resolution Imaging
3
4
122 Spectrometer (MODIS) satellite images for the years 2001-2012 at a spatial resolution of
5
6
7 123 approximately 250 m (Estel et al. 2015), depicting the extent of fallow land annually. From these
8
9
124 maps, we derived dominantly unmanaged farmland for the year 2006 by identifying pixels with at
10
11
12 125 least four fallow years between 2001 and 2006 and at least ten fallow years between 2001 and 2012
13
14
126 following the definitions by Estel et al. (2015).
15
16
17
127 2.1.2 Intensity of broad land-use categories
18
19
20 128 Regarding land-use intensity, we used metrics for both, input and output intensity (Erb et al.
21
22
129 2013, Kuemmerle et al. 2013). We used nitrogen application rates to assess the input intensity for
23
24
25 130 croplands. Crop-specific nitrogen application rates for 1990 and 2006 were stratified into three
26
27
131 classes: (i) low intensity with 0-50 kg N ha-1, (ii) medium intensity with 50-150 kg N ha-1, and (iii)
28
29
30 132 high intensity with > 150 kg N ha-1 (Temme and Verburg 2011). To estimate input intensity on
31
32
133 grasslands, we relied on stocking densities for cattle, sheep, and goats for the years 1990 and 2006
33
34
35 134 that were measured in livestock units (LSU) and classified into four classes: (i) 0-25 LSU km-2, (ii)
36
37 135 25-50 LSU km-2, (iii) 50-100 LSU km-2, and (iv) >100 LSU km-2 (Neumann et al. 2009). We
38
39
136 combined the two middle classes into a medium intensity class.
40
41
42
137 To assess the output intensity for croplands and grasslands for the years 1990 and 2006, we
43
44
45 138 utilised data from a recent HANPP assessment for Europe (Plutzar et al. 2015). Specifically, we used
46
47
139 the amount of biomass [tC km-2 yr-1] harvested on arable cropland, permanent cropland, and
48
49
50 140 grassland, and the spatial coverage [%] of the respective land-use categories to derive a harvesting
51
52 141 intensity indicator [gC m-2 yr-1]. To avoid inflated harvest intensity values due to high yields on very
53
54
55 142 small areas, we used a maximum value of 1000 gC m-2 yr-1 (Haberl et al. 2007). To assess output
56
57 143 intensity in forestry, we used spatially disaggregated data on sub-national wood production [m³ ha-1
58
59
144 yr-1] (Verkerk et al. 2015). See Text A2 in the Supplementary Material A for details.
60
61
62
8
63

64
65
145 2.2 Explanatory factors of land use change
1
2
146 For the selection of explanatory factors of land-use change, we built upon reviews and meta-
3
4
5 147 analyses of land change (van Vliet et al. 2015, Geist et al. 2006, Lambin and Geist 2006). We
6
7
148 selected seven variables as location factors at a 1 km2 resolution that were assumed to influence land-
8
9
10 149 use change (Table A2): accessibility, aridity index, environmental zones, growing degree days,
11
12
150 population density, soil organic carbon, and terrain ruggedness. Furthermore, we selected seven
13
14
15 151 potential underlying drivers that were for the most part only available on NUTS-3 level (Table A2):
16
17
152 economic activity, farm characteristics (capital input, economic size, labour input, land under
18
19
20 153 agricultural use, subsidies), and protected area extent. We gathered data only for 2006, because most
21
22 154 variables were not available for the entire study area for 1990 (e.g., because some countries were not
23
24
155 in the EU then).
25
26
27
28
156 2.3 Methods
29
30
31
32 157 2.3.1 Data preparation
33
34
158 All data were harmonised to the same extent and projection (Lambert Azimuthal Equal Area
35
36
37 159 projection). Our set of 12 land-use indicators consisted of binary and continuous variables. To avoid
38
39
160 problems arising from using binary data for the calculation of changes as well as in the clustering, we
40
41
42 161 aggregated all indicators to 3x3 km2 cells by calculating the mean for continuous indicators and the
43
44 162 relative area share of certain classes for binary indicators. Likewise, we aggregated the explanatory
45
46
163 factors to the 3x3 km2 grid by calculating the majority value for the environmental zones layer, the
47
48
49 164 area share of protected areas, and mean values for the remaining factors. Data that were available for
50
51
165 administrative units only (i.e., NUTS3-level with mean = 4,423 km2 and s.d. = 5,882 km2) were
52
53
54 166 rasterised to the same 3x3 km2 resolution, assuming homogeneous distribution of values. To quantify
55
56
167 changes in land-use extent and intensity, we calculated absolute differences for all 12 indicators
57
58
59
60
61
62
9
63

64
65
168 between 1990 and 2006. Subsequently, we z-transformed the resulting differences to zero mean and
1
2 169 unit standard deviation to make indicators comparable.
3
4
5
170 2.3.2 Self-organising maps
6
7
8 171 We used self-organising maps (SOMs), an automated clustering technique based on an
9
10
172 unsupervised, competitive learning algorithm (Kohonen 2001). SOMs can be used to visualise high-
11
12
13 173 dimensional data and reduce their complexity to fewer (often two) dimensions by grouping
14
15 174 observations based on their similarity. In comparison to traditional cluster techniques such as k-
16
17
175 means or hierarchical clustering, SOMs depend less on expert rules or supervised threshold selection
18
19
20 176 and are not restricted by the number of input features (Václavík et al. 2013). Furthermore, SOMs
21
22
177 preserve the typology of the input data by weighing more similar observations stronger in the
23
24
25 178 clustering process (Ripley 1996). SOM-based algorithms have been widely applied in different
26
27
179 fields, including geographic information science (Agarwal and Skupin 2008, Kohonen 2001) and
28
29
30 180 land-system science (Václavík et al. 2013).
31
32
33 181 The parameterisation of SOMs requires – similar to k-means clustering – an a-priori definition
34
35
182 of the desired number of clusters, which are typically organised in an output grid. Choosing an
36
37
183 insufficient number of clusters in respect to the variability of the input data will result in clusters that
38
39
40 184 are not well separated and inhomogeneous, while choosing too many clusters will result in splitting
41
42
185 up homogeneous clusters. Hence, identifying the desired number of clusters is a key step to generate
43
44
45 186 a meaningful cluster map.
46
47
48 187 To identify the ideal SOM parameterisation, we performed a sensitivity analysis with varying
49
50
188 cluster dimensionalities ranging from 2x2 to 6x5. We determined both parameters by finding the
51
52
53 189 natural breakpoint in the mean distance of the samples to their cluster centroid (Maulik and
54
55
190 Bandyopadhyay 2002) and by evaluating the Davies-Bouldin cluster validity index that relates intra-
56
57
58 191 and inter-cluster variability (Davies and Bouldin 1979). For LSAs, these indices suggested an
59
60
61
62
10
63

64
65