Keywords

3.1 Introduction

Some studies have quantified the correlation between marginal vegetation and the distribution pattern of bird diversity [1, 2] and the relative importance of each type of marginal vegetation for each bird species, confirming how these elements are useful surrogates of High Nature Value (HNV) farmlands [3]. Some species are more related to hedgerows, while others are related primarily to shrubs and others are even related to uncultivated features. Marginal vegetation and isolated patches can be studied easily at the level of landscape spatial composition and configuration by a geographical information system (GIS) approach [4]. These elements promote landscape heterogeneity and can be analysed and measured using several landscape metrics.

Quantification of landscape heterogeneity and other structural characteristics of landscape or habitats using landscape metrics is not a new topic (for example, see O’Neill et al. [5]). However, the research possibilities of metrics application are still evolving due to the ongoing developments in geoinformation technologies and spatial data availability, development of theoretical approaches to measuring of landscape structure and knowledge of possible misuse (see e.g. [6–12].

In general, landscape structure can be characterized by the composition and spatial configuration of the landscape elements (e.g. [13]). Composition refers to the diversity and frequency of patches within the landscape, irrespective of their spatial configuration. Composition can be expressed by so-called non-spatial metrics, which can be sorted into four basic groups: relative class frequency, richness, evenness and diversity. Landscape configuration, in contrast, is related to spatially explicit characteristics of land cover types. The principal aspects of a landscape’s spatial configuration are the size distribution and density, shape complexity, core area, isolation, contrast, dispersion, contagion, subdivision and connectivity of the landscape patches. Both composition and configuration metrics are usable not only in landscape studies but also in animal ecology, for example as explanatory variables in species distribution modelling [14–16].

Description of landscape structure, as well as habitat structure, can be performed at three hierarchical levels [13]: for each patch in the landscape mosaic (patch level); for each patch class in the landscape (class level); or for the landscape mosaic as a whole (landscape level). Some metrics are meaningful only at one level (e.g. landscape heterogeneity indices such as the Shannon diversity index (SHDI), Shannon equitability index (SHEI) and Simpson diversity index (SIDI) can be determined only at the landscape level because they are integrated over all patches and classes in the study area), but most are defined for multiple levels (e.g. patch density (PD) and edge density (ED)). For many research topics, including identification of HNV areas, combination of these levels can bring good results, even if only the simplest composition metrics are used (see Fig. 3.1 and the landscape metrics case study at the end of this chapter).

Fig. 3.1
figure 1

Example of landscapes with various combination of patch richness–Shannon diversity index–edge density (PR–SHDI–ED) values. (a), (b): Landscapes with similar ED and PR, but very different SHDI. (a) PR = 5, SHDI = 0.182, ED = 75 m/ha. (b) PR = 6, SHDI = 0.900, ED = 80 m/ha. (c), (d): Landscapes with similar SHDI and PR, but different ED. (c) PR = 8, SHDI = 1.083, ED = 121 m/ha. (d) PR = 9, SHDI = 1.176, ED = 263 m/ha. (d) Heterogeneity is increased not only by near-natural elements like forests and woody and herbaceous vegetation, but also by paved roads and built-up areas. Data and scale: Visual interpretation and manual vectorization of orthorectified aerial photographs (orthophotos), sample localities 1 × 1 km in the Czech Republic. Author: Petra Šímová

Full size image

The critical issue of calculating and applying landscape metrics is working with an appropriate scale of data and analysis. Most of the landscape metrics are scale dependent, and many authors have pointed out that the scale of the data (observations) and the scale of the analysis must be coherent in order to calculate and interpret these metrics correctly [11, 13, 14, 17–19]. The issue of scaling is related to the grain size (pixel size) or to the minimum mapping unit (MMU), the extent of the study area and the thematic resolution (see Fig. 3.3 for an example). The behaviours of landscape metrics in relation to scaling factors has implications especially for the fields of research and the operations within which these metrics are used as explanatory variables or indicators, for example for species distribution modelling [14–16]. As HNV farmlands are closely connected with the occurrence of hedgerows, tree lines, patches of tall herb vegetation, grassy belts, etc. (i.e. elements that usually create small near-natural patches or green veining within the agricultural matrix), the issue of scale is extremely important in studies focused on HNV landscape indicators. Using an inappropriate scale can lead to inappropriate research results and conclusions (see Šímová and Gdulová [11] for a review of scale effects on metrics behaviour).

Inasmuch as all types of metrics are computed using GIS approaches and a lot of data sources and software can be used, another important issue for obtaining appropriate results is knowledge of the input data, metrics formulas and the basics of the software functionality. Even if packages specializing in landscape metrics are used, the user should be familiar with the basics of spatial data representation and methods of spatial analysis. Some pitfalls for computing and interpreting landscape metrics in HNV farmland studies, resulting from the issues of scale, data representation and computational methods, are introduced below in this chapter. The possible errors are demonstrated especially for the simplest metrics usable for identification of HNV farmland.

3.2 Simple Metrics for Identification of HNV Farmland

It is not the purpose of this chapter to give a list and description of all existing landscape metrics, because many specialized extensive works on this topic are available. One of the most comprehensive overviews of the metrics is presented in the FRAGSTATS software documentation [13], where the reader is given a deep theoretical background for using metrics in landscapes and ecological studies, as well as their description, interpretation, limitations, formulas and domains.

As shown in the landscape metrics case study at the end of this chapter, the simplest way to identify types of farmland with different levels of natural value can be to employ a combination of areas of main habitat (land cover) types, density metrics and diversity metrics. This simple approach does not have to be fruitful in all types of landscape and at all scales or for all ecological processes in the landscape, and use of more complex metrics can be necessary in some cases; however, use of the area, density and diversity metrics appears to be logical. These categories cover the amount of near-natural habitats and the amount of agricultural matrix, as well as the variety of habitat types (diversity metrics) and their arrangement in smaller or bigger elements (density metrics) (see Fig. 3.1c, d). In addition, a strong correlation of species occurrence, abundance and diversity with the amount of preferred habitat, size of habitat patches and area of ecotones has been proved in hundreds of studies. An undeniable advantage of using the simple metrics is also their simple interpretation and predictable behaviour across scales [11].

Description of the metrics used in the landscape metrics case study (based on McGarigal [13]; adjusted).

Percentage of Landscape

refers to how much of the landscape is composed of a particular habitat type. The percentages of forests, non-forest woody vegetation (e.g. tree lines, hedgerows, groups of trees and shrubs), non-agricultural herbaceous vegetation (e.g. near-natural meadows, herbaceous edges of forests, roads, streams and waterbodies), urban areas (small villages in this case), agricultural grasslands and arable land were used.

Mean Patch Size (MPS)

is a function of the number of patches and the total area, which can be calculated at the landscape level and the class level. In the sense of GIS terminology, the total number of polygons is divided by the total area of the locality (at the landscape level), or the number of polygons of a certain class is divided by the sum of the polygon areas of the same class (at the class level). MPS at the landscape level was calculated in our landscape metrics case study; however, MPS at the class level could also be useful in the HNV context.

Edge Density (ED)

is also meaningful at both the class and landscape levels. It is calculated as the sum of the lengths of the edges of the patches (polygons) of a certain habitat class or as the sum of all polygon edges. Depending on the type of landscape and landscape elements contained in the input data, ED can express a measure of the size of the patches and the area of ecotones in the landscape, as well as the number of linear elements (Fig. 3.1). One of the known limitations is that the ED value is affected by the resolution of the input data. Generally, the greater the detail with which edges are delineated in the data, the greater the edge length.

Patch Density (PD)

is the inverse value of MPS, calculated as the total area of the locality divided by the number of polygons (at the landscape level) or the sum of polygon areas of the same class divided by the number of polygons of that class (at the class level). PD and MPS are usually highly (positively and negatively, respectively) correlated with ED, but not necessarily. In general, landscape elements of the same size can have greatly varying edge lengths. Hence, the correlation depends on the shapes of the landscape elements and the jaggedness of their edges.

Patch Richness (PR)

is the simplest measure of landscape heterogeneity, which states the number of classes (habitat types) present within the study area. Like all diversity metrics, it is computed only at the landscape level. PR is not affected by the area of each class; therefore, two landscapes of the same PR value may have very different structures (see Fig. 3.1).

Shannon diversity (SHDI)

is one of the most popular diversity indices. In origin, it was introduced by information theory [20] and further applied in ecology of communities [21]; however, it is also a frequently used measure of heterogeneity in landscape ecology. Shannon diversity is computed as

$$ \mathit{\mathrm{SHDI}}=-{\displaystyle \sum_{i=1}^n\left({p}_i\kern0.15em *\kern0.15em \ln {p}_i\right)} $$

where n is the number of classes (habitat types) present in the study area (i.e. PR), p i is the proportion of the study area occupied by each class (habitat type) i. Proportions p i are within the interval 0.1, hence the natural logarithm of p i is negative. Therefore, the whole expression is multiplied by (−1) to obtain a positive value, thus a higher heterogeneity is expressed by a higher SHDI. As follows from the formula, the value of the SHDI increases not only with the number of classes (PR), but also with the evenness of their proportions. From two landscapes with the same PR, the higher SHDI will apply to the landscape where the proportions of classes are more similar. For example, the SHDI of a landscape composed of five habitat types with the proportions 0.1, 0.1, 0.1, 0.1 and 0.5 equals 1.27, while a landscape with the same habitat types arranged in the proportions 0.2, 0.2, 0.2, 0.2, 0.2 reaches an SHDI equal to 1.61.

Simpson diversity (SIDI)

is another popular measure of heterogeneity borrowed from community ecology. Similarly to Shannon diversity, SIDI works with proportions of classes for the purposes of landscape ecology and as a result it combines richness and evenness of habitat classes. Its interpretation is quite intuitive: the SIDI value given the formula below represents the probability that any two random points would fall in different habitat types.

\( \kern12.3em SIDI=1-{\displaystyle \sum_{i=1}^n}{p}_i^2 \)

All three diversity metrics have several important limitations in terms of interpretation and the comparability of their values across landscapes. One of the usual criticisms is that the measures do not convey any information about habitat quality. Due to this fact, landscape heterogeneity can increase not only owing to the phenomenon of a rising proportion of near-natural areas and green veining in a uniform agricultural landscape, but, for example, also due to spatial development of urbanized or mining areas, to the detriment of near-natural habitats (see Figs. 3.1 and 3.2). Hence, HNV farmland may theoretically have lower values of diversity indices than farmland without any ecological value. From the scaling point of view, the metrics values may depend on the study area extent, and there can be more habitat types in larger areas of interest. Moreover, spatial data describing the same study area can be created in different ways and for various purposes, so the same locality can be represented by completely different data sets, varying not only in spatial resolution but also in thematic resolution (see Fig. 3.3). This means that close attention should be paid to the character and extent of the area under study and to the way in which the landscape is represented in the input spatial data.

Fig. 3.2
figure 2

Example of different reasons for changing Shannon diversity index (SHDI) values in farmland: changes in spatial heterogeneity in the Kremze basin (Czech Republic) between 1952 and 1974. (a) Map of heterogeneity changes. The raster values were computed by subtracting two grids of spatial heterogeneity expressed by SHDI: SHDI (1974) minus SHDI (1952). (b) Example of greatly increasing heterogeneity. The area was diversified due to building of a small fishpond with near-natural grassland at its edge and division of arable land by non-cultivated grassy belts. (c) However, heterogeneity expressed in this way can increase greatly also due to development of urban areas and to changing spatial patterns of agricultural areas. (d) Example of greatly decreasing landscape heterogeneity. Although in 1952 the plot was quite intensively managed, liquidation of scattered vegetation and transformation of surrounding grasslands into arable land caused a great decrease in heterogeneity. Authors: Petra Šímová, Eva Bártová

Full size image
Fig. 3.3
figure 3

Example of input data preparation and influence of thematic resolution. The input data for calculation of landscape metrics are categorical layers (b, c) of land cover (habitats). One of the methods to create categorical maps is visual interpretation and manual vectorization of aerial photographs (a) (combined with field mapping in this case). As the method is subjective, it is necessary to prepare an interpretation methodology in advance. The main part of the methodology should be an interpretation key, describing all the categories that will be distinguished. The thematic detail (thematic resolution) influences many landscape metrics. Pictures (b) and (c) document two vectorizations of the same landscape: (b) only the main land cover types were distinguished; (c) more types of near-natural elements were mapped. It is obvious that area, density and diversity metrics would give different results and could indicate another landscape for (b) and (c). Author: Petra Šímová

Full size image

3.3 Input Data for Calculation of Landscape Metrics

Generally, two types of input data can be taken as basic information sources for computing landscape metrics: data acquired by remote sensing technologies (aerial or satellite images) and various types of maps obtained by ground mapping. In any case, categorical raster or vector layers of land cover (habitat) types must be created prior to beginning GIS analyses or using specialized software such as FRAGSTATS, Patch Analyst or V-Late.

As landscape metrics are often scale dependent and important elements of HNV farmlands are often composed from small patches or linear elements, input data at a detailed scale should be used as basic input for HNV analysis. As Fig. 3.4 shows, commonly available multispectral satellite data such as Landsat and Sentinel images have too coarse a resolution (30 m and 10 m, respectively) for recognizing small-sized landscape elements; therefore, they can be useful only for work at scales of tens or hundreds of square kilometres, where small patches and lines are disregarded. Input data with spatial resolution fine enough to describe landscapes including small groups of bushes or trees, near-natural herbaceous patches, etc., or even individual trees or bushes, can be acquired by satellites with very high resolution (VHR), e.g. WorldView 2 and 3, or QuickBird, or by using unmanned aerial vehicles (UAVs, drones) equipped with a multispectral camera (see Fig. 3.5).

Fig. 3.4
figure 4

Freely available multispectral satellite images: Landsat 8 in natural (a) and false (b) colours, resolution 30 m. At this resolution, small-sized elements in the landscape are not depicted. Better resolution (10 m) for analysis of High Nature Value (HNV) farmlands is given by Sentinel 2A images (c, d). Methods of classification of multispectral images enable us to reach more detailed thematic resolution (i.e. obtain categorical maps distinguishing more habitat types) in comparison with visual interpretation methods. Authors: Tomáš Klouček, Jiří Prošek, Petra Šímová

Full size image
Fig. 3.5
figure 5

Multispectral images with very high resolution. Satellites such as WorldView 2 (pictures (a), (b), image in natural and false colours, respectively) with resolution of 2 m enables us to recognize small woody elements and various types of herbaceous habitats in the landscape under study. For more detailed resolution (10 cm in this case) it is possible to use unmanned aerial vehicles (drones) equipped with a multispectral camera. Picture (c) was obtained using a hexacopter with a Tetracam camera. Individual bushes and trees can be recognized, as well as a lot of types of herbaceous vegetation (better distinguishable on the false colour version (d)). The advantage of such data is combination of high spatial and spectral resolution. The disadvantage is the cost of the images. Authors: Tomáš Klouček, Jiří Prošek, Jan Komárek, Ondřej Lagner, Petra Šímová

Full size image

An important advantage of utilizing remote sensing multispectral data is that various classification methods (e.g. supervised or non-supervised classification [22]) can be used to extract the categorical land cover information. Such classifications are not as dependent on subjective perception by a person or operator as manual methods of data preparation. Moreover, more detailed thematic information can be obtained (more habitat types can be recognized) thanks to spectral resolution of the images and advanced classification methods than by using visual interpretation only. If panchromatic or colour (RGB) images (usually aerial photographs) are taken as an input, the spectral information is too limited to separate land cover classes by automatic methods [23]. Therefore, visual interpretation and time-consuming manual vectorization of aerial photos are usually used in practice [23–25], sometimes in combination with field surveys or with existing maps.

3.4 Landscape Metrics Case Study: An Example of Using Landscape Metrics to Identify HNV Farmland

This short case study shows an example of how landscape metrics (even the simplest) can help to identify HNV farmlands at a detailed scale of analysis. The study used 358 sample localities (squares of 1 × 1 km) randomly distributed within farmland in the Czech Republic. The samples varied from completely uniform arable land to highly diversified (mountainous) mosaics of meadows, pastures and patches of forests. The input categorical layers were prepared using visual interpretation and manual vectorization of orthorectified colour aerial photographs (orthophotos) according to a detailed vectorization methodology and interpretation key. The categories distinguished were the percentages of forests, non-forest woody vegetation (e.g. tree lines, hedgerows, groups of trees and shrubs), non-agricultural herbaceous vegetation (e.g. near-natural meadows, herbaceous edges of forests, roads, streams and waterbodies), urban areas (small villages in this case), water bodies, agricultural grasslands and arable land. Simple landscape metrics (the percentages of the categories, ED and MPS at the landscape level, and the diversity metrics PR, Shannon diversity and SIDI) were computed using the basic tools of ArcGIS and MS Excel. Principal component analysis (PCA) in Canoco 5 software was employed to explore whether the landscape metrics were able to distinguish some types of landscape. PCA showed interpretable gradients in an ordination diagram (Fig. 3.6). For each quadrant of the ordination diagram, examples of the best-fitting sample landscape are presented (Figs. 3.7, 3.8, 3.9 and 3.10).

Fig. 3.6
figure 6

Principal component analysis (PCA) with area, density and diversity metrics. Area metrics: percentages of forests (Forest), non-forest woody vegetation (Woody), non-agricultural herbaceous vegetation (Herb), urban areas (Urban), agricultural grasslands (Grass) and arable land (Arable). Density metrics: mean patch size (MPS) and edge density (ED). Diversity metrics: patch richness (PR), Shannon diversity index (SHDI) and Simpson diversity index (SIDI)

Full size image
Fig. 3.7
figure 7

Typical landscapes of the first quadrant (Forest, Shannon diversity index (SHDI), Simpson diversity index (SIDI)). The typical samples are characterized by a mosaic of forests, non-forest woody patches, agricultural grasslands and patches of herbaceous vegetation. The metrics identified High Nature Value (HNV) farmland with a patchiness pattern

Full size image
Fig. 3.8
figure 8

Typical landscapes of the third quadrant (patch richness (PR), edge density (ED), Woody). The typical samples are characterized especially by occurrence of linear elements of woody vegetation in agricultural matrix, with addition of small-sized woody and herbaceous patches. Hence, the metrics identified High Nature Value (HNV) farmland with a significant proportion of linear elements

Full size image
Fig. 3.9
figure 9

Typical landscapes of the second quadrant (mean patch size (MPS)). The typical samples are characterized especially by big patches of arable land combined with patches of forests or woody vegetation. MPS (strongly negatively correlated with edge density (ED)) helps to interpret samples of the second quadrant as a quite uniform coarse grain (in relation to the scale of analysis) agriculture landscape with occurrence of forest patches

Full size image
Fig. 3.10
figure 10

Typical landscapes of the fourth quadrant (Arable). Agricultural landscapes with an overwhelming majority of arable land. The uniformity of the sites is sometimes interrupted by rare scattered woods or linear elements

Full size image