incorporating landscape metrics into satellite

5 downloads 0 Views 734KB Size Report
metrics into satellite analyses of land-cover change .... COHDEFOR gave out new contracts for sawmills to log in the region. Under successive contracts, the ...
Landscape Research, Vol. 27, No. 3, 253–269, 2002

Fragmentation of a Landscape: incorporating landscape metrics into satellite analyses of land-cover change

JANE SOUTHWORTH, HARINI NAGENDRA & CATHERINE TUCKER

ABSTRACT The relationship between trajectories of forest-cover change and the biophysical and social characteristics of the landscape in the mountains of Western Honduras is addressed. Metrics of land-cover change were used to infer patterns of land-use change, using Landsat TM imagery from 1987, 1991 and 1996. With 15–20% of the land cover changing across each two-date period, the study landscape was very dynamic. Areas of reforestation were signiŽcantly larger than areas of deforestation, across all dates. Patch size was a good indicator of economic activity. Stable patches of forest and agriculture were fewer and larger, compared to forest regrowth and clearing. Small patches of swidden agriculture were found close to roads, at lower elevations and on more gradual slopes between 1987 and 1991. Between 1991 and 1996, expansion of export coffee production resulted in forest clearings on steeper slopes and at higher elevations. Results highlight the importance of landscape metrics in monitoring landcover change over time. KEY WORDS: landscape metrics, reforestation, satellite imagery, Honduras, landuse change

Introduction The relationship between human behaviour and forest change poses a major research challenge for development projects, policy makers and environmental organizations that aim to improve forest management. Knowledge of the areal extent of forest cover and the processes of change represents an integral step, but in many areas of the globe this knowledge is still relatively scarce. The most frequently used technique for the mapping of tropical forests or rates of change is the visual and digital analysis of satellite data (Hall et al., 1991). On-theground Želd studies are too costly to use for such an analysis. Remote sensing from satellites allows the production of maps at a greater spatial extent and over frequent time steps, while saving time needed in the Želd for map veriŽcation, Jane Southworth, Department of Geography, 3141 Turlington Hall, University of Florida, PO Box 117315, Gainesville, FL 32611–7315, USA. Email: [email protected].edu. Harini Nagendra & Catherine Tucker, Center for the Study of Institutions, Population, and Environmental Change, Indiana University, 408 N. Indiana Ave, Bloomington, IN 47408, USA. 0142-639 7 Print/1469-971 0 Online/02/030253-1 7 Ó 2002 Landscape Research Group Ltd. DOI: 10.1080/0142639022014951 1

254

J. Southworth et al.

and for additional data collection, e.g. interviews, vegetation analysis. In addition, satellite imagery can also be considered an excellent tool for the analysis and measurement of landscape patterns due to its spatially explicit and extensive overview which, by its nature, includes a digital mosaic of the spatial arrangement of land cover (Chuvieco, 1999). Several studies have used remote sensing to map patterns of deforestation and to analyse the rates of forest-cover change in the tropics and elsewhere (Hall et al., 1991; Roughgarden et al., 1991; Woods & Skole, 1998). These studies have proven useful for interpreting the causes of deforestation and the impact of such changes on the region. Monitoring of change (be it deforestation or reforestation) is frequently perceived as one of the most important contributions of remote sensing technology to the study of global ecological and environmental change (Apan & Peterson, 1998; Roughgarden et al., 1991). Many researchers believe that the integration of remote sensing techniques within analysis of environmental change is essential if ecologists are to meet the challenges of the future, speciŽcally issues relating to global change; however, in practice, this integration has so far been limited (GrifŽths & Mather, 2000; Luque, 2000; Riitters et al., 2000). Some research has analysed landscape patterns using remotely sensed information (Chuvieco, 1999; Duncan et al., 1999; Peralta & Mather, 2000; Zaizhi, 2000). Patterns of reforestation and deforestation most often directly relate to human impacts, and are extremely complex due to changes occurring across multiple temporal and spatial scales (Duncan et al., 1999). This complexity makes it difŽcult to establish direct causal links between patterns of change and the processes lying behind them. The research reported in this paper aims to initiate discussion on the importance of directly monitoring and analysing change trajectories when land-cover change is the subject of investigation. The analysis of spatial pattern within a landscape, or landscape ecology analysis, is frequently divided into three components (Chuvieco, 1999): (1) landscape pattern, (2) landscape function and (3) landscape change. Remotely sensed imagery can be used to provide information in all of these areas. This research will focus on methodologies to incorporate information for components (1) and (3), which can indirectly be used to infer information about component (2). Such spatial analysis is an excellent way to analyse underlying ecological relationships within a landscape (Pino et al., 2000; Turner, 1990). While direct loss of forest area is a primary concern, forest fragmentation issues also assume vital signiŽcance in the context of maintaining the ‘natural’ variability in the size, shape and distribution of the mosaic of patches which exists within a landscape with little human inuence (Riitters et al., 2000). This variability is believed crucial in affecting the ow of species and materials within a landscape (Forman, 1995). It therefore becomes important to quantify changes in landscape pattern, in addition to estimates of percentage change in area over time. This study will address the use of landscape metrics to study spatial patterns of forest fragmentation and change, by calculating for each class a range of metrics such as mean patch size, mean patch shape, edge density, interspersion–juxtaposition, contagion, etc., from satellite-based land-cover classiŽcations and analyses of land-cover change. Analyses of how these patterns change over time will allow us to identify relationships between landscape patterns observed using land-cover classiŽcations, and the processes driving the changes (Brown et al., 2000). In addition, the direct linkage of geographical information system (GIS) technologies with remote sensing and landscape

Fragmentation of a Landscape

255

ecology research allows us to integrate spatial land-cover patterns and ecological processes in a manner which is essential to the understanding of processes of change (Forman, 1995; Turner, 1990). This paper highlights the importance of studying change within a landscape, as a valuable addition to the static land-cover classes which are more commonly studied by researchers when incorporating remotely sensed analyses into landscape studies (Chuvieco, 1999; Zaizhi, 2000). Results emphasize the need to measure change quantitatively, to incorporate both landscape pattern and landscape change from satellite image analysis, and to integrate GIS information on the biophysical structure of the landscape. Our analysis utilizes a study area in Western Honduras for which we have extensive ground data, socio-economic information (Tucker, 1996, 1999a,b) and satellite-derived classiŽcations of land cover from three dates (Southworth & Tucker, 2001). Information on biophysical and vegetation characteristics of the landscape allows us to infer trends in pattern at the landscape ecological level, and to associate this with information on the social processes leading to reforestation and deforestation in this complex, dynamic landscape. Materials and Methods Site Description The research site is composed of the municipio (county) of La Campa and an adjacent area that includes Celaque National Park, a cloud forest and the highest point in Honduras (Figure 1). The analysis draws on extensive ethnographic data from the area, particularly La Campa. The site is located in Western Honduras, a region characterized by mountainous topography with slopes averaging over 30 degrees. Soils tend to be thin and rocky; 97% of the region is considered unsuitable for intensive agriculture. Pine–oak forests occur as the dominant, natural vegetation, but large areas have been transformed into open pasture (Cha´vez Borjas, 1992; Pineda Portillo, 1984). The research area falls in one of the poorer parts of Honduras, a nation recognized as one of the poorest nations in Latin America. The study site has experienced important gains in the past two decades: transportation infrastructure has improved, new schools and health centres have been built, and the infant mortality rate has declined. Through better ties to markets and food-distribution networks, hunger and severe malnutrition have been reduced. Yet education and health care remain inadequate to meet the needs of the growing population. Most of the people are descendants of Lenca Indians, and many households depend primarily upon subsistence production of maize and beans for their livelihood (Fonseca et al., 1999). In the past decade, however, better-off farmers in the region have expanded coffee production for national and international markets. Coffee has been a factor in increasing social heterogeneity; some coffee growers have made major gains in income and material possessions due to years of good coffee prices. In 1996, approximately 55% of the study site’s land was forested. Most of this is secondary succession of varying age, except for the cloud forest of Celaque National Park. Logging did not begin in the area until the 1960s, and initially had little impact. In 1973, the Honduran legislature passed a new law, Decree 103, which created the Honduran Corporation for Forest Development (COHDEFOR) and nationalized all of the nation’s trees. While trees became

Figure 1. Study area and its location within Meso-America.

256 J. Southworth et al.

Fragmentation of a Landscape

257

government property, the land on which they stood remained in owners’ hands. COHDEFOR gave out new contracts for sawmills to log in the region. Under successive contracts, the prime pine forests in the region were heavily logged. Celaque, which was not declared a national park until 1987, experienced minimal logging due to its relative inaccessibility. By the 1980s, local people had become aware of environmental degradation related to inadequate logging practices by sawmills, and COHDEFOR lacked the resources to monitor and sanction sawmills for numerous violations. In La Campa, residents formed a grass-roots association that expelled COHDEFOR and ended logging in the county in April 1987. The logging ban has been enforced and it continues to receive wide support within the county (Tucker, 1999a,b). Throughout the study site, logging declined with the exhaustion of mature pine forests and the insolvency of regional sawmills. The social processes occurring in the study site have similarities to those in a number of other rural areas in the developing world. Poverty, population growth and market integration have often been linked to environmental degradation (Jansen, 1998; Kaimowitz & Angelsen, 1998). Prior research (Tucker, 1996, 1999a,b) has suggested that population pressure has led to agricultural intensiŽcation, including the use of fertilizer and ox-driven ploughs. Market integration is also linked to agricultural change and expanded production of commercial crops. While these trends would generally be predicted to motivate deforestation, our previous research addressing forest-cover change found that reforestation exceeded deforestation in this area. Between 1987 and 1996, 9.77 km2 of land was reforested and only 7.48 km2 was deforested, as determined by satellite image analysis (Southworth & Tucker, 2001). This reversal in the dominant trend of deforestation in Latin American countries, including Honduras, relates to the current institutional, biophysical and socio-economic context. In the study site, the presence of reforestation suggests a challenge to the dominant theories of land-cover change. Moreover, the patterns of this land-cover change suggest complex socio-economic and biophysical processes that must be studied in a spatial context. To date, this region has not been studied extensively by researchers, in part due to the lack of importance attached to natural pine–oak dry tropical forest types, when compared to the more attractive moist tropical forest types. The La Campa region is important to study, in order to understand the possibly anomalous processes of reforestation occurring here. Such research will also increase our knowledge of naturally occurring dry tropical forests, which cover a greater area than moist tropical forests, but have had few studies of their functions, structure and processes, or signiŽcant impacts related to human activity (Whighman et al., 1990). Image Analysis Pre-processing Landsat TM images were obtained for March 1987, 1991 and 1996, as this month corresponds to the end of the dry season when fallow agricultural lands can be easily distinguished from forests. The images were cut to exclude cloud cover in the southern half, as this region was not relevant for our analysis. Geometric rectiŽcation was carried out using 1: 50 000-scale maps and the nearestneighbour resampling algorithm, with a root-mean-square (RMS) error of less than 0.5 pixels ( , 15 m). Using a similar procedure, the rectiŽed 1996 image

258

J. Southworth et al.

served as the basis to rectify the 1987 and 1991 images via image-to-image registration. An overlay function veriŽed that the images overlapped exactly across the three image dates. Following rectiŽcation, calibration procedures corrected for sensor drift and other differences due to variations in the solar angle and atmospheric conditions. Without such calibration, any analysis to detect change could be evaluating differences at the sensor level rather than changes at the Earth’s surface. Hence, all the images underwent radiometric calibration, atmospheric correction and radiometric rectiŽcation (Jensen, 2000). Image ClassiŽcation, Change Detection and GIS Analysis Training sample data were used to determine the land-cover classes on the ground and then train the satellite image to recognize them. Classes for agriculture, young fallows (approximately 1–3 years), cleared areas, bare soil, water and urban areas were aggregated in this analysis to create a non-forest class. Forest was deŽned as having a canopy closure of 25% or greater, based on forest plots from Želdwork. In addition, this canopy-closure threshold indicates areas that function as forest both physically and socially for the communities who use these areas. Land-cover maps of forest and non-forest cover for 1987, 1991 and 1996 were derived by independent supervised classiŽcation of the Landsat images, using a Gaussian maximum-likelihood classiŽer. Only two cover classes were used to simplify the analysis of change and these images of forest and non-forest will also be used in the single-time-point analyses. Our classiŽcations were subsequently veriŽed based on additional Želd information. With classiŽcation accuracies exceeding 85% for all three dates, classiŽed images generally agreed visually with actual land cover. Following classiŽcation, change-detection analysis was undertaken. Change detection is a technique used to determine the change between two or more time periods for a particular land cover, by providing quantitative information on spatial and temporal distribution. It offers an important tool for monitoring and managing natural resources (Macleod & Congalton, 1998). Four aspects of change detection are important when monitoring naturally occurring or human-induced phenomena: (1) detecting the changes that have occurred; (2) identifying the nature of the change; (3) measuring the areal extent of the change; and (4) assessing the spatial pattern of the change (Macleod & Congalton, 1998). Estimation of change requires the acquisition of images for the same area over two or more time periods. The forest/non-forest classiŽcations were overlaid using a GIS (ARC/INFOÔ software) in order to calculate the rates and types of changes across each image. Changes in land cover across the three dates were detected using an image-addition technique resulting in eight possible classes (Table 1). Within the image-addition technique the three separate images (1987, 1991, 1996) are summed together to create one single image. In this ‘change image’ each pixel (cell of information) now includes information on land cover for all three dates (Mertens & Lambin, 1997, 2000). In order to describe the changes listed in Table 1, we have made the following assumptions: · patches changing from forest to non-forest means deforestation; · patches changing from non-forest to forest means reforestation; · cleared patches are due to some form of human activity on the landscape; and

Fragmentation of a Landscape

259

Table 1. Land-cover-change classes from the three-date change image: 1987–1991– 1996 Category

1987

1991

1996

Land-cover classa

1 2 3 4 5 6 7 8

Forest Forest Forest Forest Non-forest Non-forest Non-forest Non-forest

Forest Non-forest Non-forest Forest Non-forest Forest Forest Non-forest

Forest Non-forest Forest Non-forest Non-forest Non-forest Forest Forest

Stable forest Older, more permanent forest clearing Old forest clearing with regrowth Recent forest clearing Stable agriculture Forest regrowth with new clearing Older, more permanent forest regrowth Recent forest regrowth

a

Following Mertens and Lambin (2000).

· size and shape of patches indicate their economic function, with smaller patches of forest to non-forest representing shifting cultivation, larger patches of non-forest to forest representing areas of abandoned logging and larger abandoned agricultural areas. For example, a pixel which was forest cover in 1987, 1991 and 1996 is considered a ‘stable forest’ class, and a pixel which was forest in 1987 and 1991 and non-forest in 1996 is a ‘recent forest clearing’ (Table 1). In addition, change images were also created between the dates 1987–1991 and 1991–1996 to determine if different processes had been occurring across different time steps. Thus we created three change images: 1987–1991, 1991–1996 and 1987–1991–1996. For each of these we calculated a change matrix, i.e. percentage or area changing across dates for each land-cover class. In addition to the creation of image-based products, a number of GIS surfaces were created to help explain the spatial patterns of forest/non-forest cover and their associated classes across the change images. Using a 1: 50 000 digital elevation model and 1: 50 000 maps of the study region, coverages of roads, towns, elevation and slope were created, and analyses of frequency of spatial association between the land-cover classes and the GIS coverages were undertaken. Landscape Metrics Landscape metrics were calculated using the software Fragstats 2.0 (McGarigal & Marks, 1994) for each individual image classiŽcation (1987, 1991, 1996), in addition to all the change images. Fragstats provides a very comprehensive set of spatial statistics and descriptive metrics of pattern at the patch, class and landscape levels (Haines-Young & Chopping, 1996). The particular metrics we chose to use to measure structure at the patch level were size (area in hectares) and the patch shape index (computing the complexity of patch shape, compared with a square patch of identical area, taking a value of 1 when most compact and increasing without limit as the patch becomes more irregular) (Forman, 1995; McGarigal & Marks, 1994). We could not use Fragstats to compute nearestneighbour distance as the number of patches exceeded the maximum possible for this computation: 14 624 (McGarigal & Marks, 1994). For single-date and change image analyses, one-tailed Mann–Whitney U Tests (Sokal & Rohlf, 1981)

260

J. Southworth et al.

were used to assess whether patch size differed signiŽcantly (p , 0.05) across forest and non-forest classes, and between categories of land-cover change. Tests of signiŽcance for differences in patch shape were also carried out using the shape index; however, our results parallel the analysis for patch size and so are not reported separately here. At the class level, our interest is in comparing descriptive metrics of land-cover pattern between forest and non-forest classes, and across categories of land-cover change. These metrics can be grouped into categories of area, shape, core, diversity and contagion/interspersion (Haines-Young & Chopping 1996). To simplify interpretation, the following metrics were used (Forman, 1995; GrifŽth et al., 2000; Riitters et al., 1995). (a) Percentage land cover (% LAND): percentage of total area occupied by each class. (b) Largest-patch index (LPI): area of the largest patch in each class, expressed as a percentage of total landscape area. (c) Number of patches (NP): total number of patches in this class. (d) Mean patch size (MPS): average patch size for the class, in hectares. (e) Edge density (ED): sum of length of all edge segments for the class, divided by total landscape area. (f) Mean shape index (MSI): average complexity of patch shape for a class (the index is 1 when square, and increases without limit as the patch becomes more irregular). (g) Interspersion–juxtaposition index (IJI): degree of interspersion of patches of this class, with all other classes (this index takes values from 0, when the class is found adjacent to only one other class type, and increases to 100 as the patch type becomes increasingly interspersed with other class types). The indices of % LAND, LPI, NP and MPS correspond to area metrics. Together with ED, these provide indications of the degree of fragmentation for different land-cover types and change images. MSI and IJI provide metrics of shape and contagion/interspersion. This analysis does not include measures of core (we Žnd no ecological basis for deŽning core distance in this landscape) or diversity (as the number of classes is constant across time, diversity indices do not vary appreciably). Complete descriptions of these metrics, and equations for their calculation, are provided in McGarigal & Marks (1994). Results and Discussion Single-date Land-cover ClassiŽcations and Landscape Analyses Land-cover classes of forest and non-forest across all three dates (1987, 1991 and 1996) show that forest cover is the dominant class (Figure 2). However, between 1987 and 1991 total forested area decreased by 0.9% of the total area of the region. Between 1991 and 1996 the area of forest cover increased by 1.5% of the region. The composition of the forested area across the landscape in terms of patch numbers and sizes (Figure 2) shows that the number of forested patches has decreased from 1987 to 1991 to 1996, but that the size of these patches has increased from a mean of 8.4 hectares in 1987 to 8.9 hectares in 1991, and to 9.6 hectares in 1996. The areas of non-forest followed a more irregular trend with the number of patches decreasing from 7572 in 1987 to 6139 in 1991, and patch

Figure 2. A comparison of patch dynamics between forest and non-forest land cover classes. The upper row of boxes describes the largest patch index (LPI), number of patches (NP), mean shape index (MSI), mean patch size (MPS) and percentage cover for the area of landscape under forst cover for 1987, 1991 and 1996. The lower row of boxes similarly provides landscape level statistics for non-forest cover during 1987, 1991 and 1996. The boxes in grey describe the percentage area of the landscape that remains as forest and as non-forest, as well as the percentage area changing from forest to non-forest and vice versa, from 1987 to 1991, and 1991 to 1996.

Fragmentation of a Landscape 261

262

J. Southworth et al.

Table 2. Results of a one-tailed Mann–Whitney analysis of patch-level differences in patch size (p , 0.05) Coverage

Differences in patch size across classes in ascending order

1987 1991 1996 1987–1991 1991–1996 1987–1991–1996

Forest , non-forest Forest , non-forest Non-forest , forest Deforestation , reforestation , forest , non-forest Deforestation , forest , reforestation , non-forest Forest regrowth with new clearing , older, more permanent forest clearing/ recent forest clearing/old forest clearing with regrowth , older, more permanent forest regrowth , stable agriculture , stable forest/recent forest regrowth

size increasing from 6.1 to 7.9 hectares. From 1991 to 1996, however, the number of patches increased from 6139 to 8788, with a corresponding decrease in mean patch size from 7.9 to 5.2 hectares. A patch-level analysis (Table 2) shows that in both 1987 and 1991 the size of patches was signiŽcantly different (at 0.05 conŽdence level) between forest and non-forest areas, with non-forest or clearings being larger. By 1996, however, the size of patches was again signiŽcantly different (at 0.05 conŽdence level), but for this time period forest patches were larger. This result is indicative of the changes in land use within this study region, where there has been a shift in the agricultural base. Since 1987 there has been a decrease in swidden agricultural clearings due to the process of agricultural intensiŽcation, which has led to the abandonment of less productive Želds. In addition, since 1991 there has been an increase in coffee production within this region, which occurs in areas that differ in location from those used for past agricultural processes. Hence, the landscape patterns reect these changes in land-use processes. Three-date Change Coverages and Landscape Analyses A look at the percentages of forest and non-forest in this landscape (Figure 2) would lead us to assume that the landscape was very static, with changes between classes being within the range of classiŽcation errors ( 6 3%). However, if we view the change images and evaluate change over these time periods at a pixel-by-pixel level, we reveal a very dynamic landscape with 15–20% of the land cover changing across each two-date period. Between 1987 and 1991 there was 9.4% deforestation and 8.5% reforestation, but between 1991 and 1996 there was 9.5% reforestation and only 8.0% deforestation (Figure 2). This change in the direction of trends highlights the importance of looking beyond single-date classiŽcations and percentage land-cover change across only one time step. The analysis of change over the three dates allowed an assessment of the landscape metrics associated with the eight change classes (Table 3), in order to better determine changes which are occurring across this very dynamic landscape. These change classes can also be viewed spatially (Figure 3). The classes that represent no change across all three dates (stable forest and stable agriculture) can be seen to be quite different from the classes which represent change (Table 3). SpeciŽcally, these static classes are much fewer in

Fragmentation of a Landscape

263

Table 3. A comparison of landscape pattern for various land-cover-change categories across a three-date period: 1987–1991–1996 Change class (1987–1991–1996) Stable forest (F–F–F)a Older, more permanent forest regrowth (NF–F–F) Old forest clearing with regrowth (F–NF–F) Recent forest clearing (F–F–NF) Recent forest regrowth (NF–NF–F) Forest regrowth with new clearing (NF–F–NF) Older, more permanent forest clearing (F–NF–NF) Stable permanent agriculture (NF–NF–NF) a

F5

% LAND

LPI

NP

MPS (ha)

ED

MSI

IJI

39.07

22.35

5 850

5.96

73.43

1.35

82.01

8.49

0.3

17 923

0.42

66.32

1.27

75.46

3.85

0.04

14 209

0.24

35.59

1.16

81.90

3.11

0.16

11 241

0.25

27.51

1.15

83.07

11.73

0.21

17 389

0.60

83.74

1.35

74.00

2.37

0.03

11 363

0.19

24.07

1.11

87.32

2.71

0.01

11 072

0.22

26.23

1.15

85.78

28.66

11.15

6 408

3.99

76.12

1.33

73.76

forest and NF 5 non-forest cover for the image dates 1987–1991–1996.

number and are much larger in size. This difference in patch size is signiŽcant (p , 0.05), with stable forest and recent forest regrowth being the largest patches when analysed at the patch level, followed by stable agriculture (Table 2). These areas of no change are also more spatially contiguous, as indicated by the LPI and MPS indices. Figure 3 indicates that most landscape change occurs at the periphery of the stable forest, but inside the boundary of the stable agriculture patches. In addition, whereas the areas of stable forests are found adjacent to all other classes in the landscape, the areas of stable agriculture are predominantly found near those patches which represent current clearing and abandonment indicating the swidden agricultural cycle. This is supported by the large difference in IJI values for these two categories. Areas of non-rotational reforestation (older, more permanent forest regrowth and current forest regrowth) have the largest total number of patches of all the classes, and large edge densities; this indicates that these areas are more fragmented areas than stable forest or stable agriculture, but less fragmented in comparison with areas of deforestation (Table 3). When viewed spatially (Figure 3) we can see that much of the reforestation occurs along edges or fringes of forest classes, predominantly near areas of stable forest. The MSI and ED values support this, indicating that the landscape has long, thin patches of reforestation, as compared to the relatively smaller, more compact patches of deforestation (Table 3). The presence of these elongated patches of reforested area along the boundary of large patches of stable forest (Figure 3) explains the expansion we observe in size of forest patches over time, with non-forest patches being larger than forest patches during 1987 and 1991, and forest patches becoming larger than non-forest patches in 1996—when-single date images are analysed. The IJI is also low for the category of reforestation, again supporting the observation classes, but are predominantly adjoining other patches of reforestation and stable forest.

264

J. Southworth et al.

Figure 3. Change image for 1987–1991–1996, for the study area. Colour Žgure available for viewing at k http://www.cipec.org/publications/southworth_nagendra_and_tucker2002.html l .

Areas of rotational reforestation or deforestation (old forest clearing with regrowth and forest regrowth with new clearing) relate to swidden agriculture, and are signiŽcantly smaller in size than patches of stable forest and stable agriculture (Table 2, p , 0.05). Within this, areas of old forest clearing with regrowth are signiŽcantly larger than areas of forest regrowth with new clearing. They are also composed of more patches (higher NP), are more fragmented (higher ED and MPS), and are less well dispersed across the landscape (lower IJI) (Table 3). Indeed, the forest regrowth with new clearing class has the smallest patch size of all the classes (Table 2, p , 0.05). In addition, this class has the highest IJI (Table 3), which indicates that this class occurs spatially interspersed with all other classes in the landscape. Other areas of deforestation (older, more permanent forest clearing and recent forest clearing) are also small in size, as seen by both the MPS and patch-level statistics (Tables 2 and 3). These areas relate to clearing for agriculture in 1991 and a mixture of clearing for agriculture and for coffee production in 1996. These areas of deforestation are also fairly small in patch size, signiŽcantly smaller than patches of stable forest, stable agriculture and recent forest clearing (Table 2, p , 0.05).

Fragmentation of a Landscape

265

Land-cover-change Analyses and Biophysical Analyses When relating the change classes to the biophysical parameters of the environment we can explain the change patterns much more clearly. In general, areas that remained forest across all three dates were located at higher elevations, on steeper slopes, and at greater distances from roads (Figures 3, 4, 5). In contrast, areas that remained non-forest across all three dates were closer to roads, at lower elevations, and on gentler slopes. In the period 1987–1991, areas of reforestation were located on steeper slopes, at higher elevations and at greater distance from roads than areas which were deforested. However, between 1991 and 1996 areas of reforestation occurred on gentler slopes, at lower elevations and closer to roads when compared to areas of deforestation (Figures 3, 4, 5). This relates to a switch in land use or management practices across these dates.

Figure 4. Accessibility surface of the study area. Higher numbers imply greater inaccessibility with accessibility a function of slope, elevation, distance from roads and distance from towns. (Originally published in Southworth & Tucker (2001)).

266

J. Southworth et al.

Figure 5. Land-cover change classes plotted in elevation, slope and distance space.

Since 1991, the landscape has been impacted by coffee production (clearings on steeper slopes and at higher elevations) and land abandonment due to agricultural intensiŽcation (areas of previous agricultural production, namely lower elevations and less steep slopes). In addition, we also pick up the regrowth of previously logged areas as forested cover in our 1996 image—these logged areas were close to roads (for access), on moderately steep slopes and at mid-level elevations. Conclusions There are many complex and interrelated processes driving recent land-cover change. Within the study area, forests remain primarily on steeper slopes, at higher elevations and at a distance from settlements and roads. Regrowth has occurred in previously logged areas and new logging is prohibited at the county level. Agricultural intensiŽcation appears related to abandonment of some marginal lands. However, the expansion of export coffee production affects land cover in previously forested areas. Between 1991 and 1996, areas of deforestation occurred on steeper slopes, at higher elevations and further from roads when compared to areas of reforestation (Figure 5). This trend is the opposite of what had occurred in the past and what we would expect according to theories such as that of Von Thu¨nen (Mertens & Lambin, 2000), and is due to deforestation for coffee production increasingly being located in more highly inaccessible areas.

Fragmentation of a Landscape

267

As the coffee here is mountain-grown coffee, we expect to see coffee Želds on steeper slopes and at higher elevations. This trend is likely to continue as the currently limited areas of coffee production expand. Our analysis reveals patterns across the landscape which link land cover directly to land use. The methods used link satellite imagery with landscape metrics and draw on social data, allowing us to begin to measure and evaluate the changes in land-use patterns within the study region, and improving our ability to interpret the relationships between landscape-level impacts and changes in practices of land use and management. As highlighted in this research, the use of change images, in association with their related metrics, allows for a detailed analysis of dynamic landscapes, which may appear quite static under initial single-date land-cover analysis techniques. Although there is growing interest in the use of remote sensing to study land-cover change (Hessburg et al., 2000; Hietala-Koivu, 1999; Ochoa-Gaona & Gonzalez-Espinosa, 2000; Zaizhi, 2000), few studies link trajectories of land-cover change (change images) and landscape pattern. A major problem with the use of the multiple metrics from packages such as Fragstats has been a perceived lack of interpretability (Haines-Young & Chopping, 1996). In large part, this is because there have been few attempts to analyse associations between land-cover level and processes at a land-use level, or even between land cover and the biophysical structure of the landscape (GrifŽths & Mather, 2000). The techniques used here provide linkages between all three, permitting an analysis that links pattern and process, which is of great importance for studies of landscape ecology and land-cover change (Forman, 1995; Turner, 1989). SpeciŽcally these methods provide: · information on levels of forest change in the study area; · spatial analysis of forest fragmentation at the patch, class and landscape levels; · land-cover-change analysis (change images) through the incorporation of data from multiple time points; · comparison of spatial patterns across trajectories of land-cover change, through the use of change images; · added explanatory power due to association of change images with biophysical information on the landscape, within a GIS. This analysis combines techniques and provides methods pertinent to issues of global change, habitat conservation, biodiversity, and carbon sequestration, issues that are often anthropogenic and local in origin but are also complex (both in causes and impacts) and multi-scalar. They require multi-disciplinary methods of analysis. This research is a step in this direction as it combines researchers from environmental and social science Želds with techniques from remote sensing, GIS and landscape ecology. Further integration of methods and interpretations across disciplines is needed if we hope to understand fully and hence limit the impacts of global change on our environment. Acknowledgements This research was supported by the National Science Foundation (NSF) (SBR-9521918 ) as part of the ongoing research at the Center for the Study of Institutions, Population, and Environmental Change (CIPEC) at Indiana

268

J. Southworth et al.

University. The authors thank Laura Carlson and John Jacob Matthys for help with the GIS analysis and Michael Kohlhaas for his comments on earlier drafts.

References Apan, A.A. & Peterson, J.A. (1998) Probing tropical deforestation: the use of GIS and statistical analysis of georeferenced data, Applied Geography, 18, pp. 137–152. Brown, D.G., Duh, J.-D. & Drzyzga, S.A. (2000) Estimating error in an analysis of forest fragmentation change using North American Landscape Characterization (NALC) data, Remote Sensing of the Environment, 71, pp. 106–117. Cha´vez Borjas, M. (1992) Co´mo Subsisten los Campesinos (Tegucigalpa, Editorial Guaymuras). Chuvieco, E. (1999) Measuring changes in landscape pattern from satellite images: short-term effects of Žre on spatial diversity, International Journal of Remote Sensing, 20(12), pp. 2331–2346. Duncan, B.W., Boyle, S., Breininger, D.R. & Schmalzer, P.A. (1999) Coupling past management practice and historic landscape change on John F. Kennedy Space Center, Florida, Landscape Ecology, 14, pp. 291–309. Fonseca, J.P., Moreno, M.L. & Padgett, G.S. (1999) Estructura orõ´stica, uso de recursos y educacio´n ambiental del Parque Nacional Montan˜a Celaque, Master’s thesis, Biology Department, National Autonomous University of Honduras. Forman, R.T.T. (1995) Land Mosaics: the ecology of landscapes and regions (Cambridge, Cambridge University Press). GrifŽth, J.A., Martinko, E.A., & Price, K.P. (2000) Landscape structure analysis of Kansas at three scales, Landscape and Urban Planning, 52, pp. 45–61. GrifŽths, G.S. & Mather, P.M. (2000) Remote sensing and landscape ecology: landscape patterns and landscape change, International Journal of Remote Sensing, 21, pp. 2537–2539. Haines-Young, R. & Chopping, M. (1996) Quantifying landscape structure: a review of landscape indices and their application to forested landscapes, Progress in Physical Geography, 20, pp. 418–445. Hall, F.G, Botkin, D.B., Strebel, D.E., Woods, K.D. & Goetz, S.J. (1991) Large-scale patterns of forest succession as determined by remote sensing, Ecology, 72, pp. 628–640. Hessburg, P.F., Smith, B.G., Salter, R.B., Ottmar, R.D. & Alvarado, E. (2000) Recent changes (1930s–1990s) in spatial patterns of interior northwest forests, USA, Forest Ecology and Management, 136, pp. 53–83. Hietala-Koivu, R. (1999) Agricultural landscape change: a case study in Ylane, southwest Finland, Landscape and Urban Planning, 46, pp. 103–108. Jansen, K. (1998) Political Ecology, Mountain Agriculture, and Knowledge in Honduras (Amsterdam, Thela). Jensen, J.R. (2000) Remote Sensing of the Environment: an Earth resource perspective (Englewood Cliffs, NJ, Prentice Hall). Kaimowitz, D. & Angelsen, A. (1998) Economic Models of Tropical Deforestation: a review (Bogor, Indonesia, Center for International Forestry Research (CIFOR)). Luque, S.S. (2000) Evaluating temporal changes using Multi-Spectral Scanner and Thematic Mapper data on the landscape of a natural reserve: the New Jersey Pine Barrens, a case study, International Journal of Remote Sensing, 21, pp. 2589–2611. Macleod, R.D. & Congalton, R.G. (1998) A quantitative comparison of change-detection algorithms for monitoring Eelgrass from remotely sensed data, Photogrammetri c Engineering and Remote Sensing, 64, pp. 207–216. McGarigal, K. & Marks, B.J. (1994) Fragstats: spatial pattern analysis program for quantifying landscape structure, v. 2.0. (Corvallis, OR, Oregon Forest Science Lab, Oregon State University). Mertens, B. & Lambin, E.F. (1997) Spatial modelling of deforestation in Southern Cameroon, Applied Geography , 17, pp. 143–162. Mertens, B. & Lambin, E.F. (2000) Land-cover-change trajectories in Southern Cameroon, Annals of the Association of American Geographers, 90(3), pp. 467–495. Ochoa-Gaona, S. & Gonzalez-Espinosa, M. (2000) Land use and deforestation in the highlands of Chiapas, Mexico, Applied Geography, 20, pp. 17–42. Peralta, P. & Mather, P. (2000) An analysis of deforestation patterns in the extractive reserves of Acre, Amazonia from satellite imagery: a landscape ecological approach, International Journal of Remote Sensing, 21, pp. 2555–2570. Pineda Portillo, N. (1984) Geografõ´a de Honduras (2nd edn) (Tegucigalpa, Editorial ESP).

Fragmentation of a Landscape

269

Pino, J., Roda, F., Ribas, J. & Pons, X. (2000) Landscape structure and bird species richness: implications for conservation in rural areas between natural parks, Landscape and Urban Planning, 49, pp. 35–48. Riitters, K.H., O’Neill, R.V.O., Hunsaker, C.T., Wickham, J.D., Yankee, D.H., Timmins, S.P., Jones, K.B. & Jackson, B.L. (1995) A factor analysis of landscape pattern and structure metrics, Landscape Ecology, 10, pp. 23–39. Riitters, K., Wickham, J., O’Neill, R., Jones, B. & Smith, E. (2000) Global-scale patterns of forest fragmentation, Conservation Ecology, 4(2), 3 [online] URL: k http://www.consecol.org/vol4/iss2/ art3l Roughgarden, J., Running, S.W. & Matson, P.A. (1991) What does remote sensing do for ecology?, Ecology, 72, pp. 1918–1922. Sokal, R.R. & Rohlf, F.J. (1981) Introduction to Biostatistics (2nd edn) (Washington, DC, Island Press). Southworth, J. & Tucker, C.M. (2001) Forest cover change in western Honduras: the role of socio-economic and biophysical factors, local institutions, and land tenure, Mountain Research and Development, 21(3), pp. 276–283. Tucker, C.M. (1996) The Political Ecology of a Lenca Indian Community in Honduras: communal forests, state policy, and processes of transformation, Ph.D. dissertation, Department of Anthropology, University of Arizona, Tucson, AZ. Tucker, C.M. (1999a) Private vs. communal forests: forest conditions and tenure in a Honduran community, Human Ecology, 27, pp. 201–230. Tucker, C.M. (1999b) Evaluating a common property institution: design principles and forest management in a Honduran community, Praxis, 15, pp. 47–76. Turner, M.G. (1989) Landscape ecology: the effect of pattern on process, Annual Review of Ecological Systems, 20, pp. 171–197. Turner, M.G. (1990) Landscape changes in nine rural counties in Georgia, Photogrammetric Engineering and Remote Sensing, 56(3), pp. 379–386. Whighman, D.F., Zugasty Towle, P., Cabrera Cano, E. O’Neill, J.O. & Ley, E. (1990) The effect of annual variatio n in precipitation on growth and litter production in a tropical dry forest in the Yucatan of Mexico, Tropical Ecology, 31(2), pp. 23–34. Woods, C.H. & Skole, D. (1998) Linking satellite, census, and survey data to study deforestation in the Brazilian Amazon, in: Liverman, D., Moran, E.F., Rindfuss, R.R. & Stern, P.C. (Eds) People and Pixels: linking remote sensing and social science, pp. 70–93 (Washington, DC, National Academy Press). Zaizhi, Z. (2000) Landscape changes in a rural area in China, Landscape and Urban Planning, 47, pp. 33–38.