- Email: [email protected]
Local spatial context measurements used to explore the relationship between land cover and land use functions
International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Contents lists available at SciVerse ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Local spatial context measurements used to explore the relationship between land cover and land use functions Anders Wästfelt a,∗ , Wolter Arnberg b,1 a b
Section of Agrarian History, Department of Economics, Swedish University of Agricultural Science, Box 7013, 750 07 Uppsala, Sweden Department of Physical Geography and Quaternary Geology, Stockholm University, Stockholm, Sweden
a r t i c l e
i n f o
Article history: Received 12 January 2012 Accepted 20 September 2012 Keywords: Land configuration Agricultural land use Spatial context Relative scale European landscape convention von Thunen
a b s t r a c t Research making use of satellite data for land change science has developed in the last decades. However, analysis of land use has not developed with the same speed as development of new satellite sensors and available land cover data. Improvement of land use analysis is possible, but more advanced methods are needed which make it possible to link image data to analysis of land use functions. To make this linking possible, variable which affect farmer’s long term decisions must be taken into account in analysis as well as the relative importance of the landscape itself. A GIS-based tool for the measurement of local spatial context in satellite data is presented in this paper and used to explore the relationship between land covers present in satellite data and land use represented in official databases. By the use of the developed tool, a land configuration image (LCI) over the Siljan area in northern Sweden was produced and used for analysis. The results are twofold. First, the produced LCI holds new information about variables that are relevant for the interpretation of land use. Second, the comparison with statistics of agricultural production shows that production in the study area varies depending on the relative land configuration. Villages consisting of relatively largescale arable fields and less diverse landscape are less diverse in production than villages which consist of smaller-scale and more heterogonous landscapes. The result is especially relevant for land use studies and policymakers working on environmental and agricultural policies. We conclude that local spatial context is an endogenous variable in the relation between landscape configuration and agricultural land use. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Relative spatial scale is a key variable in many types of land use decisions and is well known in agricultural economics (Federico, 2009). However, there are no commonly used methods to measure these variables in satellite data by the use of GIS. It is also well known that the spatial composition of landscapes affects biodiversity (Fahrig, 2003), and landscape ecologists have focused on the interrelations between changes in landscape metrics and ecological processes. The increased focus of policymakers on sustainable development is a driving force encouraging researchers to determine good indicators of landscape composition and changes. However, Dramstad (2009) has argued that there is a lack of communication between policymakers and researchers that results in less use of spatial indicators in policymaking. Most geographers,
∗ Corresponding author. Tel.: +46 709 742893. E-mail address: [email protected] (A. Wästfelt). 1 Wolter Arnberg died the autumn 2011 and never got the opportunity to see this paper finished, but he was convinced of its importance. 0303-2434/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jag.2012.09.006
ecologists, and planners agree that land uses and landscapes are inherently dynamic (Holmes, 2008; Kayhko and Skanes, 2006; Tscharntke et al., 2005). Rapid processes often exist parallel to slow ones, and a number of factors help catalyze them, which means that landscape changes are complex in character. Although there are some exceptions (Verburg et al., 2009; Bibby and Shepherd, 2000), most GIS and RS researchers have focused on the formal aspect of detecting change in absolute time-space (Boulila et al., 2011). Change detection has been based on per-pixel information reflected in images from different times. Less focus has been on the relative spatial composition that often follows different forms of land use (Boulila et al., 2011). To interpret and understand how land uses change, the inherent capacities of landscape itself need to be explored. One way to do this is to investigate the relationship between agricultural land use and the landscape’s spatial structure. There is an increased demand for systematic landscape studies both globally and regionally. One example is the Swedish ratification of the European Landscape Convention (ELC) in 2011, where the Swedish government agreed to honor Article 6a in the ELC (C.o. Europe, 2000), which includes the following requirements:
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
i) to identify its own landscapes throughout its territory; ii) to analyze the characteristics of those landscapes and the forces and pressures transforming them; and iii) to take note of any changes that may occur. The latter two requirements indicate that the demand for knowledge about land use analyzed from remotely sensed data will increase. Agriculture production can relatively easily be recognized in remote sensed data, estimations of for example yield can be done by use of NDVI (Funk and Budde, 2009). Beyond intensification, expansion of area and changes of spatial structure into large scale industrialized agriculture is common. Theoretically production specialization most often is explained by distance to market (Serneels and Lambin, 2001). The classical geographical explanation of intensity and specialization of production comes from von Thunens work (Nelson, 2002). In the contemporary western world where transportation costs in relative terms are decreasing over time, this basic model no longer to the same extent explain variations in land use and production specialization (Harvey, 2006). However, on a local scale level the transport costs are still important as an explaining factor for production. Different productions specializations are, over time, limited and dependent on former land use and changes, which means that there are path dependencies in land use in relation to the spatial structure, which implicitly means that knowledge about the interrelationship between spatial structures and land use can be used for interpretation of land use from spatial structure and vice versa. But this does not mean there is causality. The first question is if a relationship can be identified between the spatial structure and production specialization? To explore this relationship at a local scale there is a problem in what and how to measure. We propose a focus on relative scale and local spatial context inside a simplified spatial village model. In the following sections, a land configuration image (LCI) will be linked and compared with data about agricultural production to determine if there is a correspondence and if so, what the current characteristics are. The LCI consists of a systematic measurement of spatial context and the relative scale of different land covers within villages. The contextual information is collected from selected land covers derived from satellite data (selection is made in relation to a simple spatial village model and field interpretations). The appearance of each land cover class inside the village is measured by calculating the relative diversity and local scale. These different measurements are put together, thus producing the land configuration image. Based on the processing and initial selection of land covers, the village model appears in the resulting image as visible villages with different characteristics depending on their inherent characteristics.
235
of the feedback mechanisms that relate environmental patterns to social processes”. They have also indicated that it is essential to integrate patterns of land cover change with processes governing land use change. Changes in land use and spatial structure are linked to technological and socioeconomic capacities, but they are not necessarily correlated in a linear fashion. The relative importance of the spatial scale and configuration at play varies from one landscape and socioeconomic context to the next. A starting point could be that land use is not randomly organized spatially but rather obeys spatial organizational principles based on functional factors related to institutional forms of organization. Fundamentally, local spatial order is based on the principles of minimum transportation and labor, and these principles produce a concentric spatial layout on a theoretically flat landscape; the most labor- and manure-demanding cultivation occurs close to the farmsteads, and less labor- and resource-intensive forms of land use occur farther away (Chisholm, 1968). The degree to which the spatial organization of land approximates the most efficient spatial layout, depends, all other things being equal, on the different techniques which land-users choose to solve the problems of transportation and labor. Improved knowledge about the relationship between land use and land cover configuration is helpful in studies investigating changes in land use and land function (Verburg et al., 2009). 1.2. The study area The study area is located in the county of Dalarna in the northwestern part of the Swedish inland area (Fig. 1). Agricultural cultivation in villages in this region is known to have occurred since at least medieval times. The area is well known for its folklore culture. However, land use has changed dramatically since World War II. Today, many farms are abandoned, and the number of farmers is
1.1. Background Recently, the Land Use and Land Cover Change (LUCC) program (Liverman and Roman, 2008; Verburg and Veldkamp, 2005) and the emerging discipline of land change science (Rindfuss et al., 2004) have focused on investigating land use and changes in land use based on remote sensing data. According to Verburg et al. (2009), it is necessary to develop new methods of quantifying land use and land function dynamics in relation to remote sensing data. Their work indicates the difficulties that arise in linking land cover data obtained using remote sensing to land functionalities because there is no one-to-one relationship between land cover and land use. They have also suggested that more attention should be paid to land use and land use functions. Similarly, Nagendra et al. (2004) argue that “there is a strong need to better understand the dynamics
Fig. 1. Location map of the study area “Siljansbygden” in Sweden and a map of Sweden in Western Europe.
236
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 2. Landscape changes during the last three centuries (Skanes, 1996). The figure schematically shows the transformation of rural landscapes during the last 300 years.
decreasing. Many old meadows and grazing areas are overgrown, and others have been cleared and amalgamated into larger farming enterprises. However, a large number of smaller farms still exist. Within the study area are both traditional villages and modern villages, containing highly mechanized, large-scale farms (Antrop, 2005). This means that the study area is a good case as the aim is to study relative differences between villages with different landscape characteristics and land use.
1.2.1. Land use on farms and in hamlets and villages The hamlet/village in Europe has a history that predates medieval times. The hamlet was simultaneously an institutional and functional unit. Therefore, a village is a compound chorological concept that cannot be directly observed in a single pixel or a small group of pixels derived from satellite images (Angelstam et al., 2003). Instead, a village will visually manifest itself as a superstructure of pixels or land cover classes. The spatial organization of land cover is related to farm size and structure and may assist in revealing the structure of the village during the interpretation of an image. The growth of villages in Scandinavian countries was historically restricted by natural conditions and the capacity of meadows to produce winter fodder. Their growth was also restricted by the transport of manure to arable fields, which thereby determined the operational scale of the farm and the village in question. Today, many former meadows (hay production units) in Sweden have been transformed into either arable land or various types of transition land, while other meadows are now managed with the help of EU subsidies. However, most former meadows and pastures have been abandoned and covered with shrub. Hence, the relative amount of transition land is indicative of long-term changes. The balance between meadowland and arable land has essentially been abandoned in tandem with modern mechanized management because fossil fuel systems and artificial fertilizers are less limited by local ecological balances and constraints. In addition, mechanization creates a reason to transform small fields into larger ones so that the use of machinery is more efficient. As a result, we can expect relatively modernized villages and farms to be larger and remnants of traditional villages to consist of smaller fields and large overgrown
areas. In Sweden, as elsewhere in developed countries, laborsaving is the main force driving the reorganization of land into larger fields and farms. For scientists in landscape ecology, this means losses in biodiversity, whereas it is known that spatial diversity and smallscale landscapes featuring varied use produce more biodiversity (Cousins and Aggemyr, 2008) and are often promoted by subsidies. The historical transition to modern mechanized landscapes is familiar from earlier studies (Skanes, 1996) and has been documented using cadastral maps and inventories (Fig. 2). The reduction in spatial diversity that has accompanied the transition from a relatively small-scale, diversified agricultural landscape to a large-scale monoculture with only arable fields and forests is apparent in most parts of Sweden today and is similar to patterns appearing in other parts of the industrialized world. Current landscapes no longer contain villages in the traditional historical sense. Nonetheless, contemporary villages are to some extent influenced by the old structure. An aerial photograph of the village of Torrberg (Fig. 3) shows the general structure and the different zones of the village. The center of the village has become spread out due to the reallocation of property. Today, the old meadows and grazing areas are overgrown with deciduous trees along the border between the arable land and the forest.
1.2.2. Contemporary farms, villages and landscapes Many different factors may influence how a contemporary farmer runs his business, which in turn drives land use changes. Over time, changes in socioeconomic and political contexts will affect farmers and produce a persistent imprint on the landscape. The agricultural landscape is primarily a product of former land use. However, landscape itself also creates incentives for further land use and/or the cessation of land use. Farmers respond to the local spatial dimension in several ways. One such response is to shift or adapt production specialization Farmers prefer to use land within the vicinity of their farm buildings, usually within a radius of 10–30 km. Longer distances are not feasible due to the high cost of driving long distances with machinery. This factor gives shape to von Thünen’s land use zones, which are defined locally in terms of functions (Haggett, 2001, p. 459). This principal spatial model can provide an example of how different
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 3. Torrberg village. Torrberg village features a structure that is typical within the study area. Photo by Arkair.
237
(1983, p. 56), “Thus, contextual classification of any pixel can potentially involve, at least, the use of any other pixel (or group of pixels) from throughout the whole scene.” This condition exists because “first of all, remotely sensed data provide an alternative representation of geographical context to that given by maps. . . the context needs to be modeled as an endogenous variable,” i.e. an endogenous variable within the image (Liverman, 1998, pp. 1–5). This means that all types of image data yield information about context, but these data must be studied and extracted to make use of them (Wastfelt, 2009). Spatial context has been used to improve existing classifications in several applications and strategies (Arbia et al., 1999; Barnsley and Barr, 1996; Barr et al., 2004). The size of the neighborhood in which spatial context is studied depends on the scale. Cao and Lam (1997) uses the term ‘operational scale’ to indicate the most suitable area size to use to study a phenomenon. In our case, the operative scale is the largest farm/village size in the study area. In the interpretation of aerial photographs, both the detailed view and the overview are important, as noted in Campbell’s discussion of photomorphic regions (Campbell, 2002). The overview (the large window) will reflect the operational scale of the phenomenon of interest (for example, the agricultural space of a village). The detailed view (the small window) will be the size of the smallest specific field. In this paper, we use the term relative scale, which we define as the relationship between the overview and the detail. Relative scale is a local measure of association or autocorrelation of the class using the two differently sized neighborhoods centered at the same location and measuring the relative proportions between features of interest. This measure resembles a reading of a photographic image in which a detailed view and an overview are considered simultaneously (Lillesand et al., 2008, p. 191). This concept is consistent with Kabanza et al. (2001, p. 329): “Yet, photo-interpretation is inherently based on many different types of knowledge, which are generally represented differently”. This process can be understood as involving the simultaneous weighting of the information from narrow and wide fields of view (FOV).
Fig. 4. Historic village model showing the local spatial organization (Haggett, 2001).
2. Method patterns of village organization exhibit different land use intensities (Macmillan and Huang, 2008). This type of simplified spatial model of a village includes the following basic categories organized around the built-up village center: cultivated land, meadowland and woodland (Fig. 4). We argue that (1) distance to market as in the traditional von Thünen model has become less important in the contemporary western world where relative transport costs are constantly declining (Harvey, 2006) and that (2) landscape features with their inherited history affect the functional organization of land use from a local perspective. Farmers’ choice of crops and production is related to the relative size and spatial distribution of the usable land inside the potential farming area. These two factors are critical both for the interpretation of diversity in land use and the transformation of agricultural land across the globe. One of the aims of this paper is to prove whether there is a relationship and if so, what are its characteristics. 1.2.3. Contextual map/or image analysis Spatial context indicates the relative occurrence of one or several classes within a restricted range. The task of determining the local spatial context is a neighborhood operation that involves applying a moving circle to the study area and counting the number of classes within a specified radius (Fotheringham and Zhan, 1996). Unlike texture, which describes the spectral reflectance of a contiguous group of neighboring pixels, the spatial context is not dependent on contiguities according to Gurney and Townshend
2.1.1. Image pre-processing The preprocessing of the data includes geometric correction, radiometric normalization, in the case then the study is covering a number of satellite scenes mosaicking is necessary. Mosaik is produced by the use of relative radiometric normalization (Yang and Lo, 2000). The image covering the whole study areas are exhaustive clustering of the satellite images using unsupervised K-mean classification. Classes are selected from an initial set of 20 classes to create three new binary image layers showing cultivated land, transition land and woodland, as in the simple village model. This selection is based on extensive field work. These three model classes do not necessary cover the entire image area because some of the original classes cannot be assigned to one of the three model categories. 2.1.2. Spatial context We use the concepts of single class context and class context intensity to describe what we call spatial context, which is the basic element behind the LCI. In the following these two measurements are explained. Single class context measures the relative scale of land covers studied in remote sensed data. 2.1.2.1. Single class context and relative scale The single class context measures the spatial distribution of one single class within defined spatial range intervals and is defined as
238
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244 Table 1 The satellite images used in the study. Satellite
Scene
Date
RMS of georeferencing (m)
Landsat 5 TM Landsat 7 ETM+ Landsat 7 ETM+
194/18 195/17 196/17.5
2000–07–28 2001–08–15 2001–07–05
10.25 12.82 12.59
correspond to the categories specified by the simple village model. The outcome is displayed as an RGB color image. This color image is the LCI. To facilitate interpretation, the LCI is separated into a number of classes by use of unsupervised clustering. 2.2.1. LCI processing details The satellite data are processed in six steps:
Fig. 5. The single class context is the product of the class occurrences in the large (R) and the small (r) fields of view and measures the operational scale.
the division of occurrences in a small and a large window at the same location. The sizes of the two different windows are empirically determined based on interpretations of land covers in villages in the study area; hence, scaling is unnecessary (Hall and Arnberg, 2002; Hall et al., 2004; Chen and Stow, 2003). The single class context measures the number of occurrences in two centered moving windows with the specified radii r and R (Fig. 5). The formal description of the single class context (SCCc ) is as follows: SCCc =
INT(number in fovr/( × r 2 )) (number in fovR/( × R2 )) × 255
The calculation for each individual window is normalized with the maximum number possible for the respective size of the window ( × r2 ). The resulting SCC is multiplied by 255 to stretch the information so that in the following step, it can be used in the creation of an RGB image. 2.1.2.2. Class context intensity Class context intensity (CCIc ) can be defined as the relative intensity of specific land cover in a specified neighborhood. By controlling the range intervals in which the CCI is measured, one can link the operative scale of the farming activities in villages to the measurement of land covers. CCIc is implemented by applying a village distance to each SCCc image. Village distances are measured in the field and correspond to the size of the villages and the historical transition zones in the village model (Figs. 2 and 4). The CCI is calculated in a moving window with radius dc to each SCCc image and is divided by the theoretical maximum value for that window. Because of this process, pixels with no information in the initial SCCc image will reveal information about the SCCc values near the pixel (up to a distance of dc ).
CCIc =
dc 0
1. The preprocessing of the data includes geometric correction and radiometric normalization. The satellite images used in the study are listed in Table 1. Because the study area is located on the border between three different scenes, one single mosaic has been created by the use of relative radiometric normalization (Yang and Lo, 2000). 2. Moving filters are used to calculate the single class context and class context intensity. However, the size of the filter kernels must be set according to observations in the field and the interpretation of the visual image of the study area. The single class context for the model classes is calculated. The size of r and R is determined by manually measuring the average radii of circles fitting the smallest object and the largest object in the overall study area. If the large and small windows show the same proportion of the studied class, the SCCc has a high value, and the area must be relatively large and contextually homogenous. If the value of SCCc is lower, the single class context is smaller than the large window and represents either a small homogenous area or a small-scale matrix. The actual radii sizes are as follows: woodland areas, 5 and 15 pixels; cultivated land, 3 and 30 pixels; and transition land, 3 and 10 pixels. The pixel size is 25 m. 3. Because the “signals” of the woodland, transition land and cultivated land from step 3 are restricted to the original binary classes from step 2, but will influence land transformation, we need to calculate their influence for the analysis up to the size of the village model. Thus, a circular moving window filter with a radius of 250 m (agriculture) or 625 m (transition land) is used for two of the three SCC images (woodland class was not filtered in this step). The result is normalized using the maximum value to create a class context intensity image (CCI). 4. Merging the three images in step 3 (red – cultivated, blue – transition and green – woodland) into a single three-band image produces a land configuration image. It is possible to interpret such images visually, but we simplify the color range by dividing the image into 12 classes. 5. The land use configuration image and an analysis of the villages and the landscape in the study area can be analyzed using the LCI. 3. Results
SCCc MAX(SCCc ) × 255
2.2. Land configuration image The land configuration image is created by combining the three spatial class context intensity images. These subsets (CCI1-3)
The resulting image is the LCI, shown in Fig. 6 for Gärdsjö village and for the whole study areas in Fig. 7, Fig. 7 shows how each individual village in the area is individually configured in relation to the model and shows the individual characteristics of villages in the scene. Hence, the image can be used to investigate differences between villages and to perform an analysis of correlation between land configuration and land use. A subset of the ten
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
239
Fig. 6. Original image, selected land cover categories (red – cultivated, blue – transition and green – woodland), corresponding to land use in village model. Also shown is each step in the processing, first the single class context and then the class context intensity, finally the composite images are shown together with the land configuration image. In the last image a land use trajectory illustrates the spatial order for the village Gärdsjö.
processed images from each step are shown for the village Gärdsjö (Fig. 6): the original satellite image, the selected land cover categories corresponding to the village model, the single class context for each land cover in the model, the intensity measurement inside influence distances for two of three land covers, the Land Configuration Image in RGB mode shown, together with LCI delinated into 13 classes. The last image is a land use trajectory illustrating the spatial order for the village Gärdsjö.
Fig. 7 shows the whole study area and all villages. The LCI can be read as a generalized map of land configuration in villages. The LCI shows the compound chorological concept “villages” and the relative contextual differences between villages. The LCI is linked to the initial classes derived from the remote sensing data and the additional contextual information. The spatial appearance represents the spatial order of the classes and is illustrated by a transect in the detailed view for the Gärdsjö village in Fig. 6. The image classes are interpreted as follows:
240
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 7. Land configuration image for the Lake Siljan district. The villages included for further analysis are marked with circles.
LC 2: Small-scale forest. LC 3: Small-scale heterogeneous land, including forest. LC 4: Small-scale heterogeneous forest. LC 5: Forest. LC 6: Small-scale heterogeneous transition land. LC 7: Small-scale transition land. LC 8: Small-scale cultivated landscape. LC 9: Transition land. LC 10: Large-scale cultivated land. LC 11: Large-scale transition land. LC 12: Large large-scale forest.
The land configuration image can be interpreted at three different levels.
1. Each pixel in the LCI includes information concerning the following: (a) in which part of a village the pixel is located and can thus be directly interpreted; (b) the relative degree to which a pixel belongs to the classes in the simple village model; and (c) the diversity and the relative scale in the local single class context and the class context intensity in the neighborhood of each pixel.
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
241
Fig. 8. Diagrams for the four villages Våmhus, Boda, Siljansnäs and Stumsnäs showing cross matching between LCI classes and agricultural production classes. Production classes – 3: grain, 7: fallow, 9: fodder (arable), 10: fodder (meadow), 11: other, 12: other meadow, 13: silage.
2. The image shows “villages” as compound chorological concepts and illustrates to what degree their pattern matches the simple village model, which makes it possible to interpret and compare the overall configuration of the villages. 3. The study area as a whole can be interpreted as consisting of different transition and change processes.
The land configuration image (Fig. 7) shows the relative differences between the villages and the relative configuration of each village in the study area. It is possible to determine the center of the large villages (interpretation level 1) categorized as LC 10. All the pixels in this class are close to the respective center of each village compared with an ordinary map, and each location includes a relatively large and concentrated area of cultivated land. Other villages are categorized as LC 8, implying that the relative amount and relative size of arable land is smaller than in those categorized as LC 10 or that the arable land is heterogeneously distributed in space. The LC 8 villages are also associated with the appearance of large areas of former meadows (currently shrubs), and the images
reveal how homogeneous the meadows and/or shrub growth are in areas between forest and arable land. The LCI shows a continuum of differences between villages in terms of relative local scale and diversity. The villages in the study area can be divided into groups with respect to their relative scale and diversity: villages with relatively large scale arable fields and homogenous configurations and those that consist of smaller arable fields and more diverse configuration. Some villages appear to have characteristics of both types.
3.1. Comparison with agriculture production data In the beginning of this article we asked if there is a relationship between agricultural land use and spatial structure. If this is the case, how can one characterize the relationship? In the following section we will compare the LCI with actual agricultural production data to test to what extent there is such a relationship. The respective production specializations of four villages as reported by the county administrative board has been cross-tabulated (pixel by pixel) with the LCI; production
242
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Table 2 Table showing cross-tabulation between the actual production and the spatial characteristics of the villages of Boda and Stumsnäs. The figures are reported in hectares. The positive values show relative more area used for production in respective LCI class in Boda. The negative values show relative more area used for production in respective LCI class in Stumsnäs.
3 4 6 7 8 9 10 11
3
7
9
−15 −159 232 231 458 −47 277 −233
0 −29 −70 −201 272 0 267 0
0 −21 372 −87 60 306 −552 0
10 0 52 138 113 844 112 97 0
11 0 −77 −195 157 −174 0 222 0
12 0 −36 180 596 568 144 160 0
13 12 −54 −91 15 388 −20 −737 −250
specializations are identified, but not the yield. Two types of interpretations of the result are then possible: first, production specialization in different types of villages and, second, whether different types of production appear in different local contexts inside villages. The general results are that fodder production is dominant in villages with large-scale arable fields and homogeneous characteristics, whereas in villages with relatively smaller arable fields and more spatially diverse characteristics, production is also more diverse (Fig. 8). These results may not be surprising, but it is important to recognize that in this case study, the spatial structure is quantified on a local scale level and in relation to the functional unity of villages and farms, a scale level where the actual decisions and changes occur. The results mean that there is a correlation between relative local configuration and land use functions. The comparison can be investigated on a more detailed level. There are only a few types of production in villages with large scale arable fields. In the two villages of Stumsnäs and Siljansnäs, mainly grain, arable fodder and silage production occur in the village center classes LC 8 and 10. The maximum value for a particular production category also occurs in LC 10 areas, which means that production in these places is mostly concentrated in the relatively large-scale field in the village center. Boda, a village characterized by relatively small scale arable fields and high spatial diversity, is different from Stumsnäs and Siljansnäs. Here, the range of production types is more varied and relatively more widely distributed among the LCI classes. Most production here occurs in category 8, which means that most of the land use types are located in diverse, relatively small-scale areas. It is also possible to notice that in less diverse, homogenous villages, “fodder meadows” (production class 10) and “other meadows” (production class 12) are rare. Where these types of land use do exist, they are found mostly on the periphery of each village, which is consistent with general historical knowledge about where former grazing land was located. To further investigate these correlations, we used one diagram to compare two typical villages. In the diagram (Table 2 and Fig. 9), the values for Boda are in the upper half, whereas those for Stumsnäs are under the zero plane. The comparison proves that land users prefer to specialize more in the cultivation of fodder and silage in villages characterized by less spatial diversity and relatively large arable fields. When landscapes are more diverse, heterogeneous and small-scale, production is focused on fodder in old meadows and traditional grain production. Grain production is mostly associated with large scale enterprises but are here found in an area with relatively small scale landscape structure which is somewhat surprising. The differences can be quantitatively calculated. For example, there are 187 more hectares of silage production in large-scale areas such as Stumsnäs than in Boda, whereas there are 212 more hectares of fodder production in old meadows in Boda than in Stumsnäs. What emerges here is the complex
Fig. 9. Diagram comparing land use for the villages of Boda and Stumsnäs. Production classes – 3: grain, 7: fallow, 9: arable fodder, 10: fodder meadows, 11: other, 12: other meadow, 13: silage.
interrelationship between local spatial configurations and the effects of changes in land use originating and created by land users over a long time. The differences shown in this comparison are a result of the long-term historical development of landscape and land use. One hypothesis is that villages with relatively large scale arable fields have undergone a transition from a high level of variation in land use to more homogenous land use. A few farmers run the homogenous villages by themselves often by leasing many other landowners’ land, and the increasing homogenization of the landscape is apparent. This dynamic is consistent with the narrow base of production in those communities. In the more diverse villages, land use is largely based on small-scale production, and there are a larger number of active small-scale farmers who are still on their own land subdivisions. The relatively higher amount of grain production here is interesting, and a more in-depth analysis will be required to explain it, as it is counterintuitive. A hypothetical explanation is that farmers here are upholding more traditional farming. As previously noted, current land use depends on previous land use. In turn, current land use will affect future land use and the landscape. This relationship between spatial structure and land use function is manifested locally, but it is shaped within the broader social context of economics, markets, and national and international relations. The importance of these results and findings are that it indicates that the distance to market is not the most important factor for development production specialization. Instead the local village perspective seems to indicate that the local configuration has a connection to different values transforming both agricultural land use and landscapes. 4. Discussion Landscape metrics in the form of local spatial context measurements have been presented, applied and investigated in this paper. The explicit focus of the local spatial context analysis and local agricultural functions means that local spatial variables linking land use have been identified and measured in satellite data. Assessments are based on land configuration images (LCI). The approach presented provides a possible means of further investigating land configuration on a local scale and makes it possible to interpret land use on a local scale. The LCI method is semi-automatic and can be expanded to areas that are larger than those that have been considered in this study.
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
The LCI used in the present case study shows the land configurations inside and between villages, but this method can be used to study all types of agricultural landscapes given that the principles governing the organization of land use are similar. Villages are both spatial and social entities, and this simple fact has been used to interpret the relative differences in land configuration and land use. Similarly, the interpretation of LCIs can be guided by knowledge about land use. LCIs reveal different responses to the economy, markets, state policies, historical processes and property boundaries. This study aimed to explore the relationship between land use and landscape structure. LCIs can be used to determine land use in a specific, known socioeconomic context, LCIs can also be used as a foundation for land use research and future monitoring and planning. Policymakers and researchers need to be aware of this relationship if they are to improve their capacity to track and adequately respond to land use/landscape changes. It is important to bear in mind that there is no simple causality between land use and land configuration and that the proper interpretation of an LCI will always be dependent on the socio-economic and technological context. With regard to production, some economists argue that relatively small-scale family farms are the most effective in terms of the area used and not in terms of labor, whereas the opposite is often true for large-scale farms (Ellis, 1993). This study did not address this problem, but LCIs could be used in the future to quantify this type of relationship in different socioeconomic settings. When an increase in production is necessary, downsizing the farm in question may be an option. This is not the case when up-scaling is present, as in the studied area, which leads to less diverse production and a more homogenous landscape. As previously mentioned, this is an ongoing process, but it is not necessarily a long-term prediction of landscape change; the socioeconomic situation can change so that changes in the opposite direction become the norm. However, having generic knowledge about the relationship between spatial context and land use will be fundamental in the future.
5. Conclusions Over the last decade a number of studies have been presented demonstrating how to study land cover and land use. Most such studies operate in absolute time space. Also many studies still relate to von Thunen’s theoretical assumption that distance to market is the variable explaining production specialization. By adding a local spatial perspective and relative measurement of local spatial variables a link was established between spatial analysis of remote sensed data and actual land use. The result from this study shows that there is a local correlation between the agricultural production specialization and the relative scale and spatial context on a local level. Relative scale and spatial context are interrelated to values produced by the use of agriculture landscapes. These findings validate our assumption that considering the local spatial context and the local relative scale as endogenous variables. But, there is not necessarily causality between spatial configuration and land use, because such a relationship involves human actors responding in unpredictable ways. But, there seems to be a strong path dependency between landscape configuration and land use, but this must be verified by further studies. This LCI is a tool which makes satellite images more useful for land use studies in the future. The results presented in this paper is a first step toward the investigation of the changing trends in land use and there relative stability, and it can be seen as a small step in improving the study of land function dynamics (Verburg et al., 2009).
243
Further studies are necessary to validate the usefulness of LCI, but we argue that it has the potential to help policymakers assess land use and address environmental considerations such as land use, landscape characteristics and biodiversity (Dramstad, 2009). To explore this, explicit studies of other values coming from land use could be explored. Finally, by the explicit integration of the local village perspective, relative local scale and local spatial context, it was shown that local configuration has a correlation to land use and land use functions. Local spatial configuration of landscape itself seems to affect the direction of production. This result supports our assumption that local context have effect on production specialization, not only distance to market. Acknowledgements We wish to thank two anonymous reviewers for their constructive comments. Funding for this research was provided by MISTRA (The Foundation for Strategic Environmental Research) and by the Swedish National Space Board. We are grateful to Johanna Jansson and Anki Nordin at the County Administrative Board of Dalarna for their help with the fieldwork required for this study. Valuable comments by Professor Ulf Sporrong on earlier drafts of this manuscript are also gratefully acknowledged. References Angelstam, P., Boresjo-Bronge, L., Mikusinski, G., Sporrong, U., Wastfelt, A., 2003. Assessing village authenticity with satellite images: a method to identify intact cultural landscapes in Europe. AMBIO 32, 594–604. Antrop, M., 2005. Why landscapes of the past are important for the future. Landscape and Urban Planning 70, 21–34. Arbia, G., Benedetti, R., Espa, G., 1999. Contextual classification in image analysis: an assessment of accuracy of ICM. Computational Statistics and Data Analysis 30, 443–455. Barnsley, M.J., Barr, S.L., 1996. Inferring urban land use from satellite sensor images using kernel-based spatial reclassification. Photogrammetric Engineering and Remote Sensing 62, 949–958. Barr, S.L., Barnsley, M.J., Steel, A., 2004. On the separability of urban land-use categories in fine spatial scale land-cover data using structural pattern recognition. Environment and Planning B: Planning and Design 31, 397–418. Bibby, P., Shepherd, J., 2000. GIS, land use, and representation. Environment and Planning B: Planning and Design 27, 583–598. Boulila, W., Farah, I.R., Ettabaa, K.S., Solaiman, B., Ben Ghezala, H., 2011. A data mining based approach to predict spatiotemporal changes in satellite images. International Journal of Applied Earth Observation and Geoinformation 13, 386–395. C.o. Europe, 2000. European Landscape Convention , Council of Europe. Campbell, J.B., 2002. Introduction to Remote Sensing. Taylor & Francis, London. Cao, C., Lam, N.S.-N., 1997. Understanding the scale and resolution effects in remote sensing and GIS. In: D. Quattrochi, A., Goodchild, M.F. (Eds.), Scale in Remote Sensing and GIS. Lewis Publishers (Chapter 3). Chen, D.M., Stow, D., 2003. Strategies for integrating information from multiple spatial resolutions into land-use/land-cover classification routines. Photogrammetric Engineering and Remote Sensing 69, 1279–1287. Chisholm, M., 1968. Rural Settlement and Land Use, An Essay in Location, 2nd ed. Hutchinson University Library [S.l.]. Cousins, S.A.O., Aggemyr, E., 2008. The influence of field shape, area and surrounding landscape on plant species richness in grazed ex-fields. Biological Conservation 141, 126–135. Dramstad, W.E., 2009. Spatial metrics—useful indicators for society or mainly fun tools for landscape ecologists? Norsk Geografisk Tidsskrift/Norwegian Journal of Geography 63, 246–254. Ellis, F., 1993. Peasant Economics: Farm Households and Agrarian Development. Cambridge University Press, Cambridge. Fahrig, L., 2003. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology Evolution and Systematics 34, 487–515. Federico, G., 2009. Feeding the World: An Economic History of Agriculture, 1800–2000. Princeton University Press, Princeton, NJ. Fotheringham, A.S., Zhan, F.B., 1996. A comparison of three exploratory methods for cluster detection in spatial point patterns. Geographical Analysis 28, 200–218. Funk, C., Budde, M.E., 2009. Phenologically-tuned MODIS NDVI-based production anomaly estimates for Zimbabwe. Remote Sensing of Environment 113, 115–125. Gurney, C.M., Townshend, J.R.G., 1983. The use of contextual information in the classification of remotely sensed data. Photogrammetric Engineering and Remote Sensing 49, 55–64. Haggett, P., 2001. Geography: A Global Synthesis. Pearson Education Ltd., Harlow.
244
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Hall, O., Arnberg, W., 2002. A method for landscape regionalization based on fuzzy membership signatures. Landscape and Urban Planning 59, 227–240. Hall, O., Hay, G.J., Bouchard, A., Marceau, D.J., 2004. Detecting dominant landscape objects through multiple scales: an integration of object-specific methods and watershed segmentation. Landscape Ecology 19, 59–76. Harvey, D., 2006. Spaces of Global Capitalism. Verso, London/New York, NY. Holmes, J., 2008. Impulses towards a multifunctional transition in rural Australia: interpreting regional dynamics in landscapes, lifestyles and livelihoods. Landscape Research 33, 211–223. Kabanza, F., Bourdua, D., Benie, G., 2001. Intelligent image analysis for environment monitoring. Advances in Environmental Research 5, 327–335. Kayhko, N., Skanes, H., 2006. Change trajectories and key biotopes—assessing landscape dynamics and sustainability. Landscape and Urban Planning 75, 300–321. Lillesand, T.M., Kiefer, R.W., Chipman, J.W., 2008. Remote Sensing and Image Interpretation, 6th ed. Wiley, New York. Liverman, D.M., 1998. People and Pixels [Elektronisk resurs] Linking Remote Sensing and Social Science, National Academy Press, Washington, DC. Liverman, D.M., Roman, C., 2008. Rosa Maria Human interactions with the Earth system: people and pixels revisited. Earth Surface Processes and Landforms 33, 1458–1471. Macmillan, W., Huang, H.Q., 2008. An agent-based simulation model of a primitive agricultural society. Geoforum 39, 643–658. Nagendra, H., Munroe, D.K., Southworth, J., 2004. From pattern to process: landscape fragmentation and the analysis of land use/land cover change. Agriculture, Ecosystems and Environment 101, 111–115.
Nelson, G.C., 2002. Introduction to the special issue on spatial analysis for agricultural economists. Agricultural Economics 27, 197–200. Rindfuss, R.R., Walsh, S.J., Turner, B.L., Fox, J., Mishra, V., 2004. Developing a science of land change: challenges and methodological issues. Proceedings of the National Academy of Sciences of the United States of America 101, 13976–13981. Serneels, S., Lambin, E.F., 2001. Proximate causes of land-use change in Narok District, Kenya: a spatial statistical model. Agriculture, Ecosystems and Environment 85, 65–81. Skanes, H., 1996. Landscape Change and Grassland Dynamics: Retrospective Studies based on Aerial Photographs and Old Cadastral Maps during 200 Years in South Sweden. Stockholm University, Stockholm. Tscharntke, T., Klein, A.M., Kruess, A., Steffan-Dewenter, I., Thies, C., 2005. Landscape perspectives on agricultural intensification and biodiversity—ecosystem service management. Ecology Letters 8, 857–874. Verburg, P.H., Veldkamp, A., 2005. Introduction to the Special Issue on spatial modeling to explore land use dynamics. International Journal of Geographical Information Science 19, 99–102. Verburg, P.H., van de Steeg, J., Veldkamp, A., Willemen, L., 2009. From land cover change to land function dynamics: a major challenge to improve land characterization. Journal of Environmental Management 90, 1327–1335. Wastfelt, A., 2009. Land use qualities identified in remotely-sensed images. International Journal of Remote Sensing 30, 2411–2427. Yang, X.J., Lo, C.P., 2000. Relative radiometric normalization performance for change detection from multi-date satellite images. Photogrammetric Engineering and Remote Sensing 66, 967–980.
Contents lists available at SciVerse ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Local spatial context measurements used to explore the relationship between land cover and land use functions Anders Wästfelt a,∗ , Wolter Arnberg b,1 a b
Section of Agrarian History, Department of Economics, Swedish University of Agricultural Science, Box 7013, 750 07 Uppsala, Sweden Department of Physical Geography and Quaternary Geology, Stockholm University, Stockholm, Sweden
a r t i c l e
i n f o
Article history: Received 12 January 2012 Accepted 20 September 2012 Keywords: Land configuration Agricultural land use Spatial context Relative scale European landscape convention von Thunen
a b s t r a c t Research making use of satellite data for land change science has developed in the last decades. However, analysis of land use has not developed with the same speed as development of new satellite sensors and available land cover data. Improvement of land use analysis is possible, but more advanced methods are needed which make it possible to link image data to analysis of land use functions. To make this linking possible, variable which affect farmer’s long term decisions must be taken into account in analysis as well as the relative importance of the landscape itself. A GIS-based tool for the measurement of local spatial context in satellite data is presented in this paper and used to explore the relationship between land covers present in satellite data and land use represented in official databases. By the use of the developed tool, a land configuration image (LCI) over the Siljan area in northern Sweden was produced and used for analysis. The results are twofold. First, the produced LCI holds new information about variables that are relevant for the interpretation of land use. Second, the comparison with statistics of agricultural production shows that production in the study area varies depending on the relative land configuration. Villages consisting of relatively largescale arable fields and less diverse landscape are less diverse in production than villages which consist of smaller-scale and more heterogonous landscapes. The result is especially relevant for land use studies and policymakers working on environmental and agricultural policies. We conclude that local spatial context is an endogenous variable in the relation between landscape configuration and agricultural land use. © 2012 Elsevier B.V. All rights reserved.
1. Introduction Relative spatial scale is a key variable in many types of land use decisions and is well known in agricultural economics (Federico, 2009). However, there are no commonly used methods to measure these variables in satellite data by the use of GIS. It is also well known that the spatial composition of landscapes affects biodiversity (Fahrig, 2003), and landscape ecologists have focused on the interrelations between changes in landscape metrics and ecological processes. The increased focus of policymakers on sustainable development is a driving force encouraging researchers to determine good indicators of landscape composition and changes. However, Dramstad (2009) has argued that there is a lack of communication between policymakers and researchers that results in less use of spatial indicators in policymaking. Most geographers,
∗ Corresponding author. Tel.: +46 709 742893. E-mail address: [email protected] (A. Wästfelt). 1 Wolter Arnberg died the autumn 2011 and never got the opportunity to see this paper finished, but he was convinced of its importance. 0303-2434/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jag.2012.09.006
ecologists, and planners agree that land uses and landscapes are inherently dynamic (Holmes, 2008; Kayhko and Skanes, 2006; Tscharntke et al., 2005). Rapid processes often exist parallel to slow ones, and a number of factors help catalyze them, which means that landscape changes are complex in character. Although there are some exceptions (Verburg et al., 2009; Bibby and Shepherd, 2000), most GIS and RS researchers have focused on the formal aspect of detecting change in absolute time-space (Boulila et al., 2011). Change detection has been based on per-pixel information reflected in images from different times. Less focus has been on the relative spatial composition that often follows different forms of land use (Boulila et al., 2011). To interpret and understand how land uses change, the inherent capacities of landscape itself need to be explored. One way to do this is to investigate the relationship between agricultural land use and the landscape’s spatial structure. There is an increased demand for systematic landscape studies both globally and regionally. One example is the Swedish ratification of the European Landscape Convention (ELC) in 2011, where the Swedish government agreed to honor Article 6a in the ELC (C.o. Europe, 2000), which includes the following requirements:
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
i) to identify its own landscapes throughout its territory; ii) to analyze the characteristics of those landscapes and the forces and pressures transforming them; and iii) to take note of any changes that may occur. The latter two requirements indicate that the demand for knowledge about land use analyzed from remotely sensed data will increase. Agriculture production can relatively easily be recognized in remote sensed data, estimations of for example yield can be done by use of NDVI (Funk and Budde, 2009). Beyond intensification, expansion of area and changes of spatial structure into large scale industrialized agriculture is common. Theoretically production specialization most often is explained by distance to market (Serneels and Lambin, 2001). The classical geographical explanation of intensity and specialization of production comes from von Thunens work (Nelson, 2002). In the contemporary western world where transportation costs in relative terms are decreasing over time, this basic model no longer to the same extent explain variations in land use and production specialization (Harvey, 2006). However, on a local scale level the transport costs are still important as an explaining factor for production. Different productions specializations are, over time, limited and dependent on former land use and changes, which means that there are path dependencies in land use in relation to the spatial structure, which implicitly means that knowledge about the interrelationship between spatial structures and land use can be used for interpretation of land use from spatial structure and vice versa. But this does not mean there is causality. The first question is if a relationship can be identified between the spatial structure and production specialization? To explore this relationship at a local scale there is a problem in what and how to measure. We propose a focus on relative scale and local spatial context inside a simplified spatial village model. In the following sections, a land configuration image (LCI) will be linked and compared with data about agricultural production to determine if there is a correspondence and if so, what the current characteristics are. The LCI consists of a systematic measurement of spatial context and the relative scale of different land covers within villages. The contextual information is collected from selected land covers derived from satellite data (selection is made in relation to a simple spatial village model and field interpretations). The appearance of each land cover class inside the village is measured by calculating the relative diversity and local scale. These different measurements are put together, thus producing the land configuration image. Based on the processing and initial selection of land covers, the village model appears in the resulting image as visible villages with different characteristics depending on their inherent characteristics.
235
of the feedback mechanisms that relate environmental patterns to social processes”. They have also indicated that it is essential to integrate patterns of land cover change with processes governing land use change. Changes in land use and spatial structure are linked to technological and socioeconomic capacities, but they are not necessarily correlated in a linear fashion. The relative importance of the spatial scale and configuration at play varies from one landscape and socioeconomic context to the next. A starting point could be that land use is not randomly organized spatially but rather obeys spatial organizational principles based on functional factors related to institutional forms of organization. Fundamentally, local spatial order is based on the principles of minimum transportation and labor, and these principles produce a concentric spatial layout on a theoretically flat landscape; the most labor- and manure-demanding cultivation occurs close to the farmsteads, and less labor- and resource-intensive forms of land use occur farther away (Chisholm, 1968). The degree to which the spatial organization of land approximates the most efficient spatial layout, depends, all other things being equal, on the different techniques which land-users choose to solve the problems of transportation and labor. Improved knowledge about the relationship between land use and land cover configuration is helpful in studies investigating changes in land use and land function (Verburg et al., 2009). 1.2. The study area The study area is located in the county of Dalarna in the northwestern part of the Swedish inland area (Fig. 1). Agricultural cultivation in villages in this region is known to have occurred since at least medieval times. The area is well known for its folklore culture. However, land use has changed dramatically since World War II. Today, many farms are abandoned, and the number of farmers is
1.1. Background Recently, the Land Use and Land Cover Change (LUCC) program (Liverman and Roman, 2008; Verburg and Veldkamp, 2005) and the emerging discipline of land change science (Rindfuss et al., 2004) have focused on investigating land use and changes in land use based on remote sensing data. According to Verburg et al. (2009), it is necessary to develop new methods of quantifying land use and land function dynamics in relation to remote sensing data. Their work indicates the difficulties that arise in linking land cover data obtained using remote sensing to land functionalities because there is no one-to-one relationship between land cover and land use. They have also suggested that more attention should be paid to land use and land use functions. Similarly, Nagendra et al. (2004) argue that “there is a strong need to better understand the dynamics
Fig. 1. Location map of the study area “Siljansbygden” in Sweden and a map of Sweden in Western Europe.
236
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 2. Landscape changes during the last three centuries (Skanes, 1996). The figure schematically shows the transformation of rural landscapes during the last 300 years.
decreasing. Many old meadows and grazing areas are overgrown, and others have been cleared and amalgamated into larger farming enterprises. However, a large number of smaller farms still exist. Within the study area are both traditional villages and modern villages, containing highly mechanized, large-scale farms (Antrop, 2005). This means that the study area is a good case as the aim is to study relative differences between villages with different landscape characteristics and land use.
1.2.1. Land use on farms and in hamlets and villages The hamlet/village in Europe has a history that predates medieval times. The hamlet was simultaneously an institutional and functional unit. Therefore, a village is a compound chorological concept that cannot be directly observed in a single pixel or a small group of pixels derived from satellite images (Angelstam et al., 2003). Instead, a village will visually manifest itself as a superstructure of pixels or land cover classes. The spatial organization of land cover is related to farm size and structure and may assist in revealing the structure of the village during the interpretation of an image. The growth of villages in Scandinavian countries was historically restricted by natural conditions and the capacity of meadows to produce winter fodder. Their growth was also restricted by the transport of manure to arable fields, which thereby determined the operational scale of the farm and the village in question. Today, many former meadows (hay production units) in Sweden have been transformed into either arable land or various types of transition land, while other meadows are now managed with the help of EU subsidies. However, most former meadows and pastures have been abandoned and covered with shrub. Hence, the relative amount of transition land is indicative of long-term changes. The balance between meadowland and arable land has essentially been abandoned in tandem with modern mechanized management because fossil fuel systems and artificial fertilizers are less limited by local ecological balances and constraints. In addition, mechanization creates a reason to transform small fields into larger ones so that the use of machinery is more efficient. As a result, we can expect relatively modernized villages and farms to be larger and remnants of traditional villages to consist of smaller fields and large overgrown
areas. In Sweden, as elsewhere in developed countries, laborsaving is the main force driving the reorganization of land into larger fields and farms. For scientists in landscape ecology, this means losses in biodiversity, whereas it is known that spatial diversity and smallscale landscapes featuring varied use produce more biodiversity (Cousins and Aggemyr, 2008) and are often promoted by subsidies. The historical transition to modern mechanized landscapes is familiar from earlier studies (Skanes, 1996) and has been documented using cadastral maps and inventories (Fig. 2). The reduction in spatial diversity that has accompanied the transition from a relatively small-scale, diversified agricultural landscape to a large-scale monoculture with only arable fields and forests is apparent in most parts of Sweden today and is similar to patterns appearing in other parts of the industrialized world. Current landscapes no longer contain villages in the traditional historical sense. Nonetheless, contemporary villages are to some extent influenced by the old structure. An aerial photograph of the village of Torrberg (Fig. 3) shows the general structure and the different zones of the village. The center of the village has become spread out due to the reallocation of property. Today, the old meadows and grazing areas are overgrown with deciduous trees along the border between the arable land and the forest.
1.2.2. Contemporary farms, villages and landscapes Many different factors may influence how a contemporary farmer runs his business, which in turn drives land use changes. Over time, changes in socioeconomic and political contexts will affect farmers and produce a persistent imprint on the landscape. The agricultural landscape is primarily a product of former land use. However, landscape itself also creates incentives for further land use and/or the cessation of land use. Farmers respond to the local spatial dimension in several ways. One such response is to shift or adapt production specialization Farmers prefer to use land within the vicinity of their farm buildings, usually within a radius of 10–30 km. Longer distances are not feasible due to the high cost of driving long distances with machinery. This factor gives shape to von Thünen’s land use zones, which are defined locally in terms of functions (Haggett, 2001, p. 459). This principal spatial model can provide an example of how different
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 3. Torrberg village. Torrberg village features a structure that is typical within the study area. Photo by Arkair.
237
(1983, p. 56), “Thus, contextual classification of any pixel can potentially involve, at least, the use of any other pixel (or group of pixels) from throughout the whole scene.” This condition exists because “first of all, remotely sensed data provide an alternative representation of geographical context to that given by maps. . . the context needs to be modeled as an endogenous variable,” i.e. an endogenous variable within the image (Liverman, 1998, pp. 1–5). This means that all types of image data yield information about context, but these data must be studied and extracted to make use of them (Wastfelt, 2009). Spatial context has been used to improve existing classifications in several applications and strategies (Arbia et al., 1999; Barnsley and Barr, 1996; Barr et al., 2004). The size of the neighborhood in which spatial context is studied depends on the scale. Cao and Lam (1997) uses the term ‘operational scale’ to indicate the most suitable area size to use to study a phenomenon. In our case, the operative scale is the largest farm/village size in the study area. In the interpretation of aerial photographs, both the detailed view and the overview are important, as noted in Campbell’s discussion of photomorphic regions (Campbell, 2002). The overview (the large window) will reflect the operational scale of the phenomenon of interest (for example, the agricultural space of a village). The detailed view (the small window) will be the size of the smallest specific field. In this paper, we use the term relative scale, which we define as the relationship between the overview and the detail. Relative scale is a local measure of association or autocorrelation of the class using the two differently sized neighborhoods centered at the same location and measuring the relative proportions between features of interest. This measure resembles a reading of a photographic image in which a detailed view and an overview are considered simultaneously (Lillesand et al., 2008, p. 191). This concept is consistent with Kabanza et al. (2001, p. 329): “Yet, photo-interpretation is inherently based on many different types of knowledge, which are generally represented differently”. This process can be understood as involving the simultaneous weighting of the information from narrow and wide fields of view (FOV).
Fig. 4. Historic village model showing the local spatial organization (Haggett, 2001).
2. Method patterns of village organization exhibit different land use intensities (Macmillan and Huang, 2008). This type of simplified spatial model of a village includes the following basic categories organized around the built-up village center: cultivated land, meadowland and woodland (Fig. 4). We argue that (1) distance to market as in the traditional von Thünen model has become less important in the contemporary western world where relative transport costs are constantly declining (Harvey, 2006) and that (2) landscape features with their inherited history affect the functional organization of land use from a local perspective. Farmers’ choice of crops and production is related to the relative size and spatial distribution of the usable land inside the potential farming area. These two factors are critical both for the interpretation of diversity in land use and the transformation of agricultural land across the globe. One of the aims of this paper is to prove whether there is a relationship and if so, what are its characteristics. 1.2.3. Contextual map/or image analysis Spatial context indicates the relative occurrence of one or several classes within a restricted range. The task of determining the local spatial context is a neighborhood operation that involves applying a moving circle to the study area and counting the number of classes within a specified radius (Fotheringham and Zhan, 1996). Unlike texture, which describes the spectral reflectance of a contiguous group of neighboring pixels, the spatial context is not dependent on contiguities according to Gurney and Townshend
2.1.1. Image pre-processing The preprocessing of the data includes geometric correction, radiometric normalization, in the case then the study is covering a number of satellite scenes mosaicking is necessary. Mosaik is produced by the use of relative radiometric normalization (Yang and Lo, 2000). The image covering the whole study areas are exhaustive clustering of the satellite images using unsupervised K-mean classification. Classes are selected from an initial set of 20 classes to create three new binary image layers showing cultivated land, transition land and woodland, as in the simple village model. This selection is based on extensive field work. These three model classes do not necessary cover the entire image area because some of the original classes cannot be assigned to one of the three model categories. 2.1.2. Spatial context We use the concepts of single class context and class context intensity to describe what we call spatial context, which is the basic element behind the LCI. In the following these two measurements are explained. Single class context measures the relative scale of land covers studied in remote sensed data. 2.1.2.1. Single class context and relative scale The single class context measures the spatial distribution of one single class within defined spatial range intervals and is defined as
238
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244 Table 1 The satellite images used in the study. Satellite
Scene
Date
RMS of georeferencing (m)
Landsat 5 TM Landsat 7 ETM+ Landsat 7 ETM+
194/18 195/17 196/17.5
2000–07–28 2001–08–15 2001–07–05
10.25 12.82 12.59
correspond to the categories specified by the simple village model. The outcome is displayed as an RGB color image. This color image is the LCI. To facilitate interpretation, the LCI is separated into a number of classes by use of unsupervised clustering. 2.2.1. LCI processing details The satellite data are processed in six steps:
Fig. 5. The single class context is the product of the class occurrences in the large (R) and the small (r) fields of view and measures the operational scale.
the division of occurrences in a small and a large window at the same location. The sizes of the two different windows are empirically determined based on interpretations of land covers in villages in the study area; hence, scaling is unnecessary (Hall and Arnberg, 2002; Hall et al., 2004; Chen and Stow, 2003). The single class context measures the number of occurrences in two centered moving windows with the specified radii r and R (Fig. 5). The formal description of the single class context (SCCc ) is as follows: SCCc =
INT(number in fovr/( × r 2 )) (number in fovR/( × R2 )) × 255
The calculation for each individual window is normalized with the maximum number possible for the respective size of the window ( × r2 ). The resulting SCC is multiplied by 255 to stretch the information so that in the following step, it can be used in the creation of an RGB image. 2.1.2.2. Class context intensity Class context intensity (CCIc ) can be defined as the relative intensity of specific land cover in a specified neighborhood. By controlling the range intervals in which the CCI is measured, one can link the operative scale of the farming activities in villages to the measurement of land covers. CCIc is implemented by applying a village distance to each SCCc image. Village distances are measured in the field and correspond to the size of the villages and the historical transition zones in the village model (Figs. 2 and 4). The CCI is calculated in a moving window with radius dc to each SCCc image and is divided by the theoretical maximum value for that window. Because of this process, pixels with no information in the initial SCCc image will reveal information about the SCCc values near the pixel (up to a distance of dc ).
CCIc =
dc 0
1. The preprocessing of the data includes geometric correction and radiometric normalization. The satellite images used in the study are listed in Table 1. Because the study area is located on the border between three different scenes, one single mosaic has been created by the use of relative radiometric normalization (Yang and Lo, 2000). 2. Moving filters are used to calculate the single class context and class context intensity. However, the size of the filter kernels must be set according to observations in the field and the interpretation of the visual image of the study area. The single class context for the model classes is calculated. The size of r and R is determined by manually measuring the average radii of circles fitting the smallest object and the largest object in the overall study area. If the large and small windows show the same proportion of the studied class, the SCCc has a high value, and the area must be relatively large and contextually homogenous. If the value of SCCc is lower, the single class context is smaller than the large window and represents either a small homogenous area or a small-scale matrix. The actual radii sizes are as follows: woodland areas, 5 and 15 pixels; cultivated land, 3 and 30 pixels; and transition land, 3 and 10 pixels. The pixel size is 25 m. 3. Because the “signals” of the woodland, transition land and cultivated land from step 3 are restricted to the original binary classes from step 2, but will influence land transformation, we need to calculate their influence for the analysis up to the size of the village model. Thus, a circular moving window filter with a radius of 250 m (agriculture) or 625 m (transition land) is used for two of the three SCC images (woodland class was not filtered in this step). The result is normalized using the maximum value to create a class context intensity image (CCI). 4. Merging the three images in step 3 (red – cultivated, blue – transition and green – woodland) into a single three-band image produces a land configuration image. It is possible to interpret such images visually, but we simplify the color range by dividing the image into 12 classes. 5. The land use configuration image and an analysis of the villages and the landscape in the study area can be analyzed using the LCI. 3. Results
SCCc MAX(SCCc ) × 255
2.2. Land configuration image The land configuration image is created by combining the three spatial class context intensity images. These subsets (CCI1-3)
The resulting image is the LCI, shown in Fig. 6 for Gärdsjö village and for the whole study areas in Fig. 7, Fig. 7 shows how each individual village in the area is individually configured in relation to the model and shows the individual characteristics of villages in the scene. Hence, the image can be used to investigate differences between villages and to perform an analysis of correlation between land configuration and land use. A subset of the ten
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
239
Fig. 6. Original image, selected land cover categories (red – cultivated, blue – transition and green – woodland), corresponding to land use in village model. Also shown is each step in the processing, first the single class context and then the class context intensity, finally the composite images are shown together with the land configuration image. In the last image a land use trajectory illustrates the spatial order for the village Gärdsjö.
processed images from each step are shown for the village Gärdsjö (Fig. 6): the original satellite image, the selected land cover categories corresponding to the village model, the single class context for each land cover in the model, the intensity measurement inside influence distances for two of three land covers, the Land Configuration Image in RGB mode shown, together with LCI delinated into 13 classes. The last image is a land use trajectory illustrating the spatial order for the village Gärdsjö.
Fig. 7 shows the whole study area and all villages. The LCI can be read as a generalized map of land configuration in villages. The LCI shows the compound chorological concept “villages” and the relative contextual differences between villages. The LCI is linked to the initial classes derived from the remote sensing data and the additional contextual information. The spatial appearance represents the spatial order of the classes and is illustrated by a transect in the detailed view for the Gärdsjö village in Fig. 6. The image classes are interpreted as follows:
240
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Fig. 7. Land configuration image for the Lake Siljan district. The villages included for further analysis are marked with circles.
LC 2: Small-scale forest. LC 3: Small-scale heterogeneous land, including forest. LC 4: Small-scale heterogeneous forest. LC 5: Forest. LC 6: Small-scale heterogeneous transition land. LC 7: Small-scale transition land. LC 8: Small-scale cultivated landscape. LC 9: Transition land. LC 10: Large-scale cultivated land. LC 11: Large-scale transition land. LC 12: Large large-scale forest.
The land configuration image can be interpreted at three different levels.
1. Each pixel in the LCI includes information concerning the following: (a) in which part of a village the pixel is located and can thus be directly interpreted; (b) the relative degree to which a pixel belongs to the classes in the simple village model; and (c) the diversity and the relative scale in the local single class context and the class context intensity in the neighborhood of each pixel.
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
241
Fig. 8. Diagrams for the four villages Våmhus, Boda, Siljansnäs and Stumsnäs showing cross matching between LCI classes and agricultural production classes. Production classes – 3: grain, 7: fallow, 9: fodder (arable), 10: fodder (meadow), 11: other, 12: other meadow, 13: silage.
2. The image shows “villages” as compound chorological concepts and illustrates to what degree their pattern matches the simple village model, which makes it possible to interpret and compare the overall configuration of the villages. 3. The study area as a whole can be interpreted as consisting of different transition and change processes.
The land configuration image (Fig. 7) shows the relative differences between the villages and the relative configuration of each village in the study area. It is possible to determine the center of the large villages (interpretation level 1) categorized as LC 10. All the pixels in this class are close to the respective center of each village compared with an ordinary map, and each location includes a relatively large and concentrated area of cultivated land. Other villages are categorized as LC 8, implying that the relative amount and relative size of arable land is smaller than in those categorized as LC 10 or that the arable land is heterogeneously distributed in space. The LC 8 villages are also associated with the appearance of large areas of former meadows (currently shrubs), and the images
reveal how homogeneous the meadows and/or shrub growth are in areas between forest and arable land. The LCI shows a continuum of differences between villages in terms of relative local scale and diversity. The villages in the study area can be divided into groups with respect to their relative scale and diversity: villages with relatively large scale arable fields and homogenous configurations and those that consist of smaller arable fields and more diverse configuration. Some villages appear to have characteristics of both types.
3.1. Comparison with agriculture production data In the beginning of this article we asked if there is a relationship between agricultural land use and spatial structure. If this is the case, how can one characterize the relationship? In the following section we will compare the LCI with actual agricultural production data to test to what extent there is such a relationship. The respective production specializations of four villages as reported by the county administrative board has been cross-tabulated (pixel by pixel) with the LCI; production
242
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Table 2 Table showing cross-tabulation between the actual production and the spatial characteristics of the villages of Boda and Stumsnäs. The figures are reported in hectares. The positive values show relative more area used for production in respective LCI class in Boda. The negative values show relative more area used for production in respective LCI class in Stumsnäs.
3 4 6 7 8 9 10 11
3
7
9
−15 −159 232 231 458 −47 277 −233
0 −29 −70 −201 272 0 267 0
0 −21 372 −87 60 306 −552 0
10 0 52 138 113 844 112 97 0
11 0 −77 −195 157 −174 0 222 0
12 0 −36 180 596 568 144 160 0
13 12 −54 −91 15 388 −20 −737 −250
specializations are identified, but not the yield. Two types of interpretations of the result are then possible: first, production specialization in different types of villages and, second, whether different types of production appear in different local contexts inside villages. The general results are that fodder production is dominant in villages with large-scale arable fields and homogeneous characteristics, whereas in villages with relatively smaller arable fields and more spatially diverse characteristics, production is also more diverse (Fig. 8). These results may not be surprising, but it is important to recognize that in this case study, the spatial structure is quantified on a local scale level and in relation to the functional unity of villages and farms, a scale level where the actual decisions and changes occur. The results mean that there is a correlation between relative local configuration and land use functions. The comparison can be investigated on a more detailed level. There are only a few types of production in villages with large scale arable fields. In the two villages of Stumsnäs and Siljansnäs, mainly grain, arable fodder and silage production occur in the village center classes LC 8 and 10. The maximum value for a particular production category also occurs in LC 10 areas, which means that production in these places is mostly concentrated in the relatively large-scale field in the village center. Boda, a village characterized by relatively small scale arable fields and high spatial diversity, is different from Stumsnäs and Siljansnäs. Here, the range of production types is more varied and relatively more widely distributed among the LCI classes. Most production here occurs in category 8, which means that most of the land use types are located in diverse, relatively small-scale areas. It is also possible to notice that in less diverse, homogenous villages, “fodder meadows” (production class 10) and “other meadows” (production class 12) are rare. Where these types of land use do exist, they are found mostly on the periphery of each village, which is consistent with general historical knowledge about where former grazing land was located. To further investigate these correlations, we used one diagram to compare two typical villages. In the diagram (Table 2 and Fig. 9), the values for Boda are in the upper half, whereas those for Stumsnäs are under the zero plane. The comparison proves that land users prefer to specialize more in the cultivation of fodder and silage in villages characterized by less spatial diversity and relatively large arable fields. When landscapes are more diverse, heterogeneous and small-scale, production is focused on fodder in old meadows and traditional grain production. Grain production is mostly associated with large scale enterprises but are here found in an area with relatively small scale landscape structure which is somewhat surprising. The differences can be quantitatively calculated. For example, there are 187 more hectares of silage production in large-scale areas such as Stumsnäs than in Boda, whereas there are 212 more hectares of fodder production in old meadows in Boda than in Stumsnäs. What emerges here is the complex
Fig. 9. Diagram comparing land use for the villages of Boda and Stumsnäs. Production classes – 3: grain, 7: fallow, 9: arable fodder, 10: fodder meadows, 11: other, 12: other meadow, 13: silage.
interrelationship between local spatial configurations and the effects of changes in land use originating and created by land users over a long time. The differences shown in this comparison are a result of the long-term historical development of landscape and land use. One hypothesis is that villages with relatively large scale arable fields have undergone a transition from a high level of variation in land use to more homogenous land use. A few farmers run the homogenous villages by themselves often by leasing many other landowners’ land, and the increasing homogenization of the landscape is apparent. This dynamic is consistent with the narrow base of production in those communities. In the more diverse villages, land use is largely based on small-scale production, and there are a larger number of active small-scale farmers who are still on their own land subdivisions. The relatively higher amount of grain production here is interesting, and a more in-depth analysis will be required to explain it, as it is counterintuitive. A hypothetical explanation is that farmers here are upholding more traditional farming. As previously noted, current land use depends on previous land use. In turn, current land use will affect future land use and the landscape. This relationship between spatial structure and land use function is manifested locally, but it is shaped within the broader social context of economics, markets, and national and international relations. The importance of these results and findings are that it indicates that the distance to market is not the most important factor for development production specialization. Instead the local village perspective seems to indicate that the local configuration has a connection to different values transforming both agricultural land use and landscapes. 4. Discussion Landscape metrics in the form of local spatial context measurements have been presented, applied and investigated in this paper. The explicit focus of the local spatial context analysis and local agricultural functions means that local spatial variables linking land use have been identified and measured in satellite data. Assessments are based on land configuration images (LCI). The approach presented provides a possible means of further investigating land configuration on a local scale and makes it possible to interpret land use on a local scale. The LCI method is semi-automatic and can be expanded to areas that are larger than those that have been considered in this study.
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
The LCI used in the present case study shows the land configurations inside and between villages, but this method can be used to study all types of agricultural landscapes given that the principles governing the organization of land use are similar. Villages are both spatial and social entities, and this simple fact has been used to interpret the relative differences in land configuration and land use. Similarly, the interpretation of LCIs can be guided by knowledge about land use. LCIs reveal different responses to the economy, markets, state policies, historical processes and property boundaries. This study aimed to explore the relationship between land use and landscape structure. LCIs can be used to determine land use in a specific, known socioeconomic context, LCIs can also be used as a foundation for land use research and future monitoring and planning. Policymakers and researchers need to be aware of this relationship if they are to improve their capacity to track and adequately respond to land use/landscape changes. It is important to bear in mind that there is no simple causality between land use and land configuration and that the proper interpretation of an LCI will always be dependent on the socio-economic and technological context. With regard to production, some economists argue that relatively small-scale family farms are the most effective in terms of the area used and not in terms of labor, whereas the opposite is often true for large-scale farms (Ellis, 1993). This study did not address this problem, but LCIs could be used in the future to quantify this type of relationship in different socioeconomic settings. When an increase in production is necessary, downsizing the farm in question may be an option. This is not the case when up-scaling is present, as in the studied area, which leads to less diverse production and a more homogenous landscape. As previously mentioned, this is an ongoing process, but it is not necessarily a long-term prediction of landscape change; the socioeconomic situation can change so that changes in the opposite direction become the norm. However, having generic knowledge about the relationship between spatial context and land use will be fundamental in the future.
5. Conclusions Over the last decade a number of studies have been presented demonstrating how to study land cover and land use. Most such studies operate in absolute time space. Also many studies still relate to von Thunen’s theoretical assumption that distance to market is the variable explaining production specialization. By adding a local spatial perspective and relative measurement of local spatial variables a link was established between spatial analysis of remote sensed data and actual land use. The result from this study shows that there is a local correlation between the agricultural production specialization and the relative scale and spatial context on a local level. Relative scale and spatial context are interrelated to values produced by the use of agriculture landscapes. These findings validate our assumption that considering the local spatial context and the local relative scale as endogenous variables. But, there is not necessarily causality between spatial configuration and land use, because such a relationship involves human actors responding in unpredictable ways. But, there seems to be a strong path dependency between landscape configuration and land use, but this must be verified by further studies. This LCI is a tool which makes satellite images more useful for land use studies in the future. The results presented in this paper is a first step toward the investigation of the changing trends in land use and there relative stability, and it can be seen as a small step in improving the study of land function dynamics (Verburg et al., 2009).
243
Further studies are necessary to validate the usefulness of LCI, but we argue that it has the potential to help policymakers assess land use and address environmental considerations such as land use, landscape characteristics and biodiversity (Dramstad, 2009). To explore this, explicit studies of other values coming from land use could be explored. Finally, by the explicit integration of the local village perspective, relative local scale and local spatial context, it was shown that local configuration has a correlation to land use and land use functions. Local spatial configuration of landscape itself seems to affect the direction of production. This result supports our assumption that local context have effect on production specialization, not only distance to market. Acknowledgements We wish to thank two anonymous reviewers for their constructive comments. Funding for this research was provided by MISTRA (The Foundation for Strategic Environmental Research) and by the Swedish National Space Board. We are grateful to Johanna Jansson and Anki Nordin at the County Administrative Board of Dalarna for their help with the fieldwork required for this study. Valuable comments by Professor Ulf Sporrong on earlier drafts of this manuscript are also gratefully acknowledged. References Angelstam, P., Boresjo-Bronge, L., Mikusinski, G., Sporrong, U., Wastfelt, A., 2003. Assessing village authenticity with satellite images: a method to identify intact cultural landscapes in Europe. AMBIO 32, 594–604. Antrop, M., 2005. Why landscapes of the past are important for the future. Landscape and Urban Planning 70, 21–34. Arbia, G., Benedetti, R., Espa, G., 1999. Contextual classification in image analysis: an assessment of accuracy of ICM. Computational Statistics and Data Analysis 30, 443–455. Barnsley, M.J., Barr, S.L., 1996. Inferring urban land use from satellite sensor images using kernel-based spatial reclassification. Photogrammetric Engineering and Remote Sensing 62, 949–958. Barr, S.L., Barnsley, M.J., Steel, A., 2004. On the separability of urban land-use categories in fine spatial scale land-cover data using structural pattern recognition. Environment and Planning B: Planning and Design 31, 397–418. Bibby, P., Shepherd, J., 2000. GIS, land use, and representation. Environment and Planning B: Planning and Design 27, 583–598. Boulila, W., Farah, I.R., Ettabaa, K.S., Solaiman, B., Ben Ghezala, H., 2011. A data mining based approach to predict spatiotemporal changes in satellite images. International Journal of Applied Earth Observation and Geoinformation 13, 386–395. C.o. Europe, 2000. European Landscape Convention , Council of Europe. Campbell, J.B., 2002. Introduction to Remote Sensing. Taylor & Francis, London. Cao, C., Lam, N.S.-N., 1997. Understanding the scale and resolution effects in remote sensing and GIS. In: D. Quattrochi, A., Goodchild, M.F. (Eds.), Scale in Remote Sensing and GIS. Lewis Publishers (Chapter 3). Chen, D.M., Stow, D., 2003. Strategies for integrating information from multiple spatial resolutions into land-use/land-cover classification routines. Photogrammetric Engineering and Remote Sensing 69, 1279–1287. Chisholm, M., 1968. Rural Settlement and Land Use, An Essay in Location, 2nd ed. Hutchinson University Library [S.l.]. Cousins, S.A.O., Aggemyr, E., 2008. The influence of field shape, area and surrounding landscape on plant species richness in grazed ex-fields. Biological Conservation 141, 126–135. Dramstad, W.E., 2009. Spatial metrics—useful indicators for society or mainly fun tools for landscape ecologists? Norsk Geografisk Tidsskrift/Norwegian Journal of Geography 63, 246–254. Ellis, F., 1993. Peasant Economics: Farm Households and Agrarian Development. Cambridge University Press, Cambridge. Fahrig, L., 2003. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology Evolution and Systematics 34, 487–515. Federico, G., 2009. Feeding the World: An Economic History of Agriculture, 1800–2000. Princeton University Press, Princeton, NJ. Fotheringham, A.S., Zhan, F.B., 1996. A comparison of three exploratory methods for cluster detection in spatial point patterns. Geographical Analysis 28, 200–218. Funk, C., Budde, M.E., 2009. Phenologically-tuned MODIS NDVI-based production anomaly estimates for Zimbabwe. Remote Sensing of Environment 113, 115–125. Gurney, C.M., Townshend, J.R.G., 1983. The use of contextual information in the classification of remotely sensed data. Photogrammetric Engineering and Remote Sensing 49, 55–64. Haggett, P., 2001. Geography: A Global Synthesis. Pearson Education Ltd., Harlow.
244
A. Wästfelt, W. Arnberg / International Journal of Applied Earth Observation and Geoinformation 23 (2013) 234–244
Hall, O., Arnberg, W., 2002. A method for landscape regionalization based on fuzzy membership signatures. Landscape and Urban Planning 59, 227–240. Hall, O., Hay, G.J., Bouchard, A., Marceau, D.J., 2004. Detecting dominant landscape objects through multiple scales: an integration of object-specific methods and watershed segmentation. Landscape Ecology 19, 59–76. Harvey, D., 2006. Spaces of Global Capitalism. Verso, London/New York, NY. Holmes, J., 2008. Impulses towards a multifunctional transition in rural Australia: interpreting regional dynamics in landscapes, lifestyles and livelihoods. Landscape Research 33, 211–223. Kabanza, F., Bourdua, D., Benie, G., 2001. Intelligent image analysis for environment monitoring. Advances in Environmental Research 5, 327–335. Kayhko, N., Skanes, H., 2006. Change trajectories and key biotopes—assessing landscape dynamics and sustainability. Landscape and Urban Planning 75, 300–321. Lillesand, T.M., Kiefer, R.W., Chipman, J.W., 2008. Remote Sensing and Image Interpretation, 6th ed. Wiley, New York. Liverman, D.M., 1998. People and Pixels [Elektronisk resurs] Linking Remote Sensing and Social Science, National Academy Press, Washington, DC. Liverman, D.M., Roman, C., 2008. Rosa Maria Human interactions with the Earth system: people and pixels revisited. Earth Surface Processes and Landforms 33, 1458–1471. Macmillan, W., Huang, H.Q., 2008. An agent-based simulation model of a primitive agricultural society. Geoforum 39, 643–658. Nagendra, H., Munroe, D.K., Southworth, J., 2004. From pattern to process: landscape fragmentation and the analysis of land use/land cover change. Agriculture, Ecosystems and Environment 101, 111–115.
Nelson, G.C., 2002. Introduction to the special issue on spatial analysis for agricultural economists. Agricultural Economics 27, 197–200. Rindfuss, R.R., Walsh, S.J., Turner, B.L., Fox, J., Mishra, V., 2004. Developing a science of land change: challenges and methodological issues. Proceedings of the National Academy of Sciences of the United States of America 101, 13976–13981. Serneels, S., Lambin, E.F., 2001. Proximate causes of land-use change in Narok District, Kenya: a spatial statistical model. Agriculture, Ecosystems and Environment 85, 65–81. Skanes, H., 1996. Landscape Change and Grassland Dynamics: Retrospective Studies based on Aerial Photographs and Old Cadastral Maps during 200 Years in South Sweden. Stockholm University, Stockholm. Tscharntke, T., Klein, A.M., Kruess, A., Steffan-Dewenter, I., Thies, C., 2005. Landscape perspectives on agricultural intensification and biodiversity—ecosystem service management. Ecology Letters 8, 857–874. Verburg, P.H., Veldkamp, A., 2005. Introduction to the Special Issue on spatial modeling to explore land use dynamics. International Journal of Geographical Information Science 19, 99–102. Verburg, P.H., van de Steeg, J., Veldkamp, A., Willemen, L., 2009. From land cover change to land function dynamics: a major challenge to improve land characterization. Journal of Environmental Management 90, 1327–1335. Wastfelt, A., 2009. Land use qualities identified in remotely-sensed images. International Journal of Remote Sensing 30, 2411–2427. Yang, X.J., Lo, C.P., 2000. Relative radiometric normalization performance for change detection from multi-date satellite images. Photogrammetric Engineering and Remote Sensing 66, 967–980.