You are on page 1of 7

Issues in Measuring Landscape Fragmentation Author(s): Carlos Davidson Reviewed work(s): Source: Wildlife Society Bulletin, Vol.

26, No. 1 (Spring, 1998), pp. 32-37 Published by: Allen Press Stable URL: http://www.jstor.org/stable/3783798 . Accessed: 10/05/2012 16:00
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Allen Press is collaborating with JSTOR to digitize, preserve and extend access to Wildlife Society Bulletin.

http://www.jstor.org

32

MEASURING FRAGMENTATION
T * * l ]

ssues m measurlng

lanclscape

fragmentation
CarlosDavidsorz
Landscape patterns are of increasing concern in land management planning, species conservation, and ecological studies. The spatial relationship of habitat has been important in assessing the status of organisms as diverse as the spotted owl (Strix occidentalis; Thomas et al. 1990) and the mountain yellow-legged frog (Rana muscosa; Bradford et al. 1993). Landscape analyses have been used by a number of national forests to reduce forest fragmentation caused by timber harvesting (Hussey et. al 1989, Marsh et al. 1990, Craig and Carlson 1993). However, despite the increased use of spatial analysis, experts have not reached an agreement on how to measure landscape patterns. I examine 2 examples of land management planning that attempted to minimize fragmentation and make general recommendations for measuring fragmentation. between total area and edge length (e.g., perimeter: area ratios, shape indices, fractal dimensions). None of these approaches take isolation into account. They do not differentiate between cutting of an isolated patch and cutting a patch of equal area and shape that has a close neighbor (Fig. 1). There are 2 basic solutions to the lack of an overall index of fragmentation. The first is to select the single aspect of fragmentation that is of most concern to the question of interest. Different species will be affected by different aspects of fragmentation (Lehmkuhl and Ruggiero 1991, van Dorp and Opdam 1987). For a large7highly mobile animal, total habitat area may be more important than interior area or edge length (Lehmkuhl and Ruggiero 1991). For a small, less mobile animal with a small home range, patch isolation may be more important (Lehmkuhl and Ruggiero 1991). Interpreting single-factor measures of fragmentation is tricky. Edge length, patch area, interior area, isolation, and other indicators of fragmentation all interact and may change in contradictory directions as fragmentation proceeds. For example, with "matrix reverse," as cutting (or other disturbance) proceeds in an area, a matrix of forest with patches of clearcuts eventually becomes a matrix of harvested areas with remnant patches of forest. The measure of total edge length between forest and openings, a possible indicator of fragmentation, will initially increase as cutting begins, but, depending on the pattern of cuts, may later decline. The second approach to measuring fragmentation is to use several measures. Use of multiple measures is most appropriate when concern is for the integrity of the entire ecosystem, rather than the impact on a single species with specific needs. Measuring frag-

Quantitative analysis of fragmentation


Fragmentation, the breaking up of continuous habitats, produces many changes in the landscape that can be quantified: reduced area of certain habitats, increased edge, reduced interior area, increased isolation of patches, and possibly increased number of patches and decreased average patch size. Most of these effects can be measured separately; however, there is no single measure that captures all aspects of fragmentation. This poses problems for efforts to quantitatively evaluate changes in fragmentation or alternative landscape designs. Often a single measure is mistakenly used as an overall measure of fragmentation. Common proxies for fragmentation include interior area or various measures of the relationship

Author's addressduring research: this USDAForest Service,PacificSouthwest, Research Station Region5, 630 SansomeStreet, and San
Francisco, CA 941 l l, USA. Author's present address: Section Evolution and Ecology, University of California, Davis, Davis, CA, 95616

Keywords: connectivity, fragmentation, isolation,landscapeindices,spatialanalysis


Wildlife Society Bulletin 1998, 26(1):32-37

Peer edited

Measuringfragmentation* Davidson

33

avoiding the use of 2 highly correlated measures, indices should be chosen that are sensitive to the aspect of landscape pattern of concern.

Iwo examples of landscape analysis


Mount Hood National Forest
Fig.1. PatchesA, B, and C represent equal areasof habitat.Eliminatingany 1 patchleaves a landscapewith equal patcharea, interiorarea,and area-to-edge relationships. Fragmentation indices based on area or edge-to-arearelationships not revealthat a do landscapeof patchesA and B is less isolatedthana landscapeof AandCorBand C.

mentation requires balancing different measurements, often in different metrics (e.g., interior area and isolation distance). There is not an unambiguous way to combine several measures into 1 fragmentation index. For example, an evaluation of which of 2 timber harvest plans creates more fragmentation may depend upon the relative weight given to area versus that given to isolation (Fig. 2). There are many indices to choose from. Riitters et. al (1995) compared 58 metrics to determine which ones were highly correlated and which were sensitive to different aspects of landscape pattern and structure. Their study was aimed at selecting metrics to detect environmental change from remotely sensed images, and, therefore, does not indicate which metrics are best for all applications. While

The Mount Hood model was originally developed in response to a congressional mandate to evaluate the effects of 1990 timber sales on fragmentation of old growth (Hussey et al. -1989). All 15 major drainages on the Mount Hood forest were inventoried and mapped using low-level color aerial photography, field surveys, and geographic information system (GIS) analysis. Areas were characterized as oldgrowth forest based on Pacific Northwest (PNW) Experiment Station (Franklin et al. 1986) and PNW Regional Guide definitions (U.S. Dep. Agric., For. Serv. 1984). The analysis considered 4 types of areas: (1) oldgrowth forest interior (core), (2) old-growth edges surrounding core area, (3) edge without interior area, and (4) other (openings and all other forest types). The depth of edge between old-growth forest and "other"(whether it was clear-cut or less-than-mature forest) was defined as 120 m. Mount Hood National Forest created a fragmentation index by assigning weights to the 4 area types and calculating the weighted proportion of old-growth area. Core area was given a weight of 1.0; stand edge with core, 0.5; edge without core, 0.1; and "other" received zero weight. The fragmentation index was then calculated as:
[(OG X 1.0) + (Edge X 0.5) + (Edge-No-Core X 0.1)]
FI

Total Area

B
Fig.2. Whether landscapeof patchesA and B or A and C is less a fragmented dependsuponthe importance areaversusisolation of in measuringfragmentation. differentspecies, total habitat For areaor habitatisolationmay be moreor less important assessin ing fragmentation.

Where: OG = total area old-growth interior Edge = total area old-gronvthedge Edge-No-Core= total edge area without core. The index measured weighted percent of oldgrowth interior, but did not address important aspects of fragmentation such as isolation or connectivity (Fig. 3). The effects of timber sales on old-growth fragmentation were analyzed using a weighting system that gave more value to core area >1.62 ha [4 acres] from patch perimeter and took into account different types of edges. Using a GIS, the forest map was grided into 0.81-ha [2-acre] cells, and each cell was assigned weights as follows: interior old growth, 20;

34
A

Wildlife Society Bulletin1998, 26(1):32-37


This may be useful for analysis of species sensitive to 1 type of edge but not another. The neighborhood analysis extends the effects of edge beyond the narrow buffer, although using a wider buffer would produce similar results. The neighborhood analysis does create a gradient of values, rather than a simple interior-edge boundary. A majorweakness of the Mt. Hood approach is that the analysis does not put a value on maintaining connections between old-growth patches or ripariancorridors. The fragmentation index measures a weighted old-growth core area, but it does not account for the spatial arrangement of a core area or the degree of landscape connectivity. A measure of core area should be supplemented with some measure of isolation or connectivity. The fragmentation index that was calculated was apparently never used. Instead, proposed timber sales were evaluated using fragmentation cost which does not adequately measure fragmentation. For example, 2 proposed timber sales in core areas of equal size, 1 circular and the other rectangular,would have had the same fragmentation cost. Yet the rectangular cut would have created more edge than the circular cut, and converted more acreage from core to edge. This could easily have been corrected by calculating the fragmentation index before the sale, mapping the proposed sale, changing the grid values to zero, and then recalculating the index. The fragmentation effect of the sale could then have been measured by the change in the index. Individual timber sales could be evaluated in this manner, but the results are not additive. The fragmentation effect of 5 proposed sales (either in different locations or across a number of years) is not simply the sum of the individual sales; the total may be greater than the sum of the parts (e.g., when a series of small sales ends up dividing a continuous area into 2 separate patches, no 1 sale is responsible for the outcome). The total fragmentation effect (cumulative impact) is measured by the total change in the index: total effect = (fragmentation index before all sales) - (fragmentation index after all sales). The timber-sale analysis is done drainage by drainage. Excluding the connections across drainages leads to greater fragmentation than an analysis carried out across multiple drainages (Marsh et al. 1990).

Fig.3. The rectanglein A and the 2 squaresin B and C have the sametotalareaandsametotalperimeter length.Assuming the that areasare largerelativeto a 120-m edge depth,the MountHood fragmentation index would assignthe same value to all 3 landscapes. A perimeter:area would also fail to distinguish ratio between the 3 landscapes.Note: The areas in the figuresare not drawnexactlyto scale.

old-growth edge bounded by forest, 15; old-growth forest bounded by road, 5; old-growth forest bounded by openings, 3; and all other, 0. A neighborhood analysis was used to reweight each cell's value based on the average value of all the surrounding cells, 2 cells in all directions. The neighborhood analysis expanded the depth of edge from the defined 120 m to a maximum of approximately 360 m (the diagonal distance across 2 cells). Analysis of the fragmentation effects of logging was done drainage by drainage. Proposed timber sales were mapped onto the weighted-grid forest map. The "fragmentation cost' of each proposed timber sale was calculated as the sum total value of the cells within the harvest area. The proposed cuts were then ordered and evaluated in terms of their fragmentation cost. Evaluation. The Mount Hood approach has many strengths as well as several serious, but easily correctable problems. Using GIS,the analysis is flexible, allowing for evaluation of how different parameter values (e.g., edge width) affect the results. This method may also present problems. As with any complicated computer application, it is not always clear what is driving the results, and therefore, the results are subject to manip- Willamette National Forest ulation or to undetected errors. The Willamette National Forest in Oregon also deThe Mount Hood approach recognizes that not all veloped a landscape analysis project for timber haredge is alike by assigning different values to edge vest planning to reduce old-growth fragmentation with roads, non-old-growth forest, and openings. (Marsh et al. 1990). The analysis was done for the

Measuringfragmentation* Davidson
Upper Fall Creek Basin, a 8,910-ha (22,016-acre) area on the Lowell Ranger District. The goal of the analysis was to identify oldgrowth stands, rank the stands (based on stand structure, interior area, shape, and connection to other stands), and develop timber harvest schedules that preserved "ecologically significant" stands and their interconnections for as long as possible. The first step was to identify stand characteristics using true-color aerial photography and field surveys and to map stand polygons. Polygons with similar landscape functions were then combined into single map units. This was to avoid assessing adjoining patches of different vegetation as separate units and overlooking that together they formed a larger unit, such as foraging habitat for spotted owls. The combined polygons were connected with riparian corridors and split into blocks of 40 to several hundred acres. The blocks were the basic unit for the rest of the analysis. Blocks were rated on 5 criteria: (1) presence and amount of old-growth characteristics, (2) interior area (total area after subtracting a 400-foot buffer zone), (3) landscape position or isolation, (4) presence of unique features such as very large trees or unusual plant communities, and (5) evidence of human intrusion. A team of judges assigned points on a 1-10 scale for each criteria. The Willamette report is not explicit about how criteria were defined or scored (e.g., how to define and score isolation). Blocks were ranked as high, medium, or low value based on their combined scores for all criteria. Following Noss and Harris (1986), the high-value blocks were used along with existing special management areas (Spotted Owl Habitat, Special Interest, Research Natural, etc.) to design a landscape-level network of interior habitat and connecting corridors. A secondary network was formed using the medium-value blocks with a corresponding set of corridors. Planners for the Willamette National Forest mapped 3 alternative harvest plans for the next 40 years: (1) traditional staggered setting, (2) minimized fragmentation within each small watershed, and (3) minimized fragmentation across the entire planning area. Minimizing fragmentation meant postponing the cutting of large, high-value blocks and connectors for as long as possible. Apparently, harvests were scheduled with the intention of getting each year's harvest out of the lowest valued blocks available by going through all the low-value blocks before cutting medium-value blocks. Of the 3 alternative harvest plans, the landscape-wide plan preserved more high-value blocks for a longer period of time than either the staggered setting or

35

small-drainage, minimum-fragmentation plans. In the long-run (40 yrs), however, nearly all of the high- and medium-value blocks were cut, regardless of which harvest plan was chosen. Without reduced timber harvesting, landscape planning may simply postpone fragmentation. Evaluation. The Willamette approach had advantages over the Mount Hood approach. The Willamette analysis was on a wider geographic scale and over a longer time period. The multipledrainage scale was advantageous. The longer time period was critical (although not sufficient) for reducing fragmentation. The Willamette analysis revealed that timber-harvest decisions made in the first 10 years of a plan could severely limit the ability to maintain old-growth interior in future years. Another advantage of the Willamette approach was the inclusion of corridors in the landscape design. The rating system used for blocks allowed for the inclusion of conservation considerations, such as the presence of unique plant assemblages, not provided for by a simple yes or no old-growth classification system. The rating system also allowed multiple indicators of fragmentation (in this case, both interior area and isolation). The 3 alternative harvest plans were evaluated according to how long and how many blocks of high and medium value were maintained. A drawback of both the Willamette ancl Mount Hood approaches was that the value of forest stands was evaluated only at the beginning of the process. Proposed actions were then evaluated according to the value of the land affected, fragmentation cost of the cells in the Mount Hood model or number of blocks of certain value in the Willamette approach. Landscape fragmentation cannot be adequately measured as a l-time evaluation. Evaluating impacts requires taking measurements before and after a (simulated) change and comparing the 2 values (Fig. 4).

Conclusions
The following recommendations are for research or planning exercises that involve measurements of landscape patterns: 1. The choice of scale (both extent and grain) greatly affects analysis of fragmentation. Fragmentation can take place on many different scales (Lord and Norton 1990). There is no 1 correct scale for analysis;the relevant scale is different for each species. For example, habitat fragmentation for a salamanderand an owl will be different. Efforts to minimize fragmentation at 1 scale may actually increase

36

Wildlife Society Bulletin1998, 26(1):32-37


scale of analysis, number of patch types, or patch definitions. 4. Perimeter:area ratios should not be used as measures of fragmentation. Perimeter:area ratios may change unpredictably (Laurance and Yensen 1991), and the ratios do not capture important aspects of fragmentation, such as isolation (Fig. 3). 5. It is preferable to measure and evaluate different aspects of fragmentation separately (e.g., interior area, extent of edge, and isolation), rather than trying to combine them all into 1 fragmentation index. A good analogy might be to think of the value of separate measures of blood pressure, resting heart rate, and body temperature, rather than a measure combining all 3 into an overall health index. Acknowledgments. Thanks to M. Chapel, K. Gutzwiller, M. S. Fisher, P. Manley, S. Rosenstock, and S. Salmons, who provided helpful comments on an earlier draft of this paper.

Literature cited
BRADFORD, F. TABATABAI, D. M. GRABER. D. F., AND 1993. Isolation of remaining populations of the native frog Rana muscosa, by introduced fish in Sequoia and Kings Canyon National Parks, California. Conservation Biology 7:882-888. CRAIG L., AND D. CARLSON. D. A. 1993. A proposed landscape apFig.4. PatchesA-G have equal areasand are connectedby corproach for managing late-seral-stageforests and riparian habiridors equal resistance. of Patches B, D, G, and Fareall equally A, tats for associated vertebrate species. 1993 Transactions of the isolated(each have equal distancecorridors 2 otherpatches). to Western Section of The Wildlife Societ,v29:25-33. No fragmentation index can assigna l-time value to the patches F., W. C. MAEsER,J. NANAN,J. PoPbasedon areaand isolationso thatcuttingD will be measured as FRANKLIN,J.F. HALL, LAUDENSLAYER, PINO, J. RALPH T. SPIES. C. AND 1986. Interim definition for oldcreatingmorefragmentation cuttingA, B, G, or F. Yetcutting than growth Douglas-fir and mixed conifer forest in the Pacific D dividesthe habitatinto2 unconnectedpieces, whereascutting A, B, G, or Fdoes not. Forthisdistinction be captured to fragmenNorthwest and California. U.S. Department of Agriculture, Fortationshouldbe measured the difference before-cutting as of and est Service, Research Note PNW-477. U.S. Department of Agriafter-cutting values. culture, Forest Service, Portland, Oregon. HUSSEY, B. KOTT, ZALUNARDO, R. McNAIR. 1989. FragT., D. AND mentation Model Mount Hood National Forest. U.S. Departfragmentation at another scale (e.g., smallment of Agriculture Forest Service, Mount Hood National Forgroup-selection logging avoids the large-scale est, Gresham, Oregon. fragmentation created by clearcuts, but by LAURANCE,F., AND YENSEN. W. E. 1991. Predicting the impacts of creating many small openings over a more exedge effects in fragmented habitats. Biological Conservation tensive area, may create fragmentation at a 55:77-92. LEHMKUHL, AND F. RUGGIERO. J. F., L. 1991. Forest fragmentation in finer scale). the Pacific Northwest and its potential effects on wildlife. 2. Results of analysis are affected by number of Pages 34-36 in L. F. Ruggiero, K. B. Aubry, A. B. Carey, and M. patch types and patch definitions. An analyH. Huff, technical coordinators. Wildlife and vegetation of unsis of fragmentation of oak (Quercus spp.) managed Douglas-firforests. GTR-PNW-285.U.S. Department woodlands may reach different conclusions of Agriculture, Forest Service, Pacific Northwest Research Station, Portland, Oregon. than an analysis disaggregating oak woodland J. D. 1990. Scale and the spatial concept into blue oak (Q. douglasii) woodland, coast LORD, M., AND A. NORTON. of fragmentation. Conservation Biology 4:197-202. live oak (Q. agrifolia) woodland7 etc. MARSH, J. BELCHER, M. HUNTER. G., AND 1990. A landscape con3. Sensitivity analysis should be done to see if text for project decisions affecting old growth and mature analysis results are robust to changes in the forests. U.S. Department of Agriculture, Forest Service,

N.-E

Ph.D.
r _

S.

candidate
F .

in

* hAeasuringfragmentation Davidson
Lowell Ranger District, Willamette National Forest, Lowell, Oregon. Noss, R. F., ANDL. D. HARRIS.1986. Nodes, networks, and MUMS: Preserving diversity at all scales. Environmental Management
D

37

Carlos Davidson is a ecology at the University of California, Davis His dissertation

__ _

_
w 8

10:299-309.
D. RIITTERS, H., R. V. ONEILL, C. T. HuNsAKER,J.D. WICKMAN, H. K. ANDB. L. JACKSON. 1995. A facYANKEE, P. TIMMINS, B. JONES S. K. tor analysis of landscape pattern and structure metrics. Landscape Ecology 10:23-39. B. THOMAS, W., E. D. FORSMAN, B. LINT,E. C. MESLOW, R. NOON, J. J. ANDJ. VERNER.1990. A conservation strategy for the Northern Spotted Owl. Report of the Interagency Scientific Committee to address the conservation of the Northern Spotted Owl. U.S. Department of Agriculture, Forest Service, U.S. Department of the Interior, Bureau of Land Management, Fish and Wildlife Service, National Park Service, Portland, Oregon. FOREST SERVICE.1984. Regional U. S. DEPARTMENT AGRICULTURE, OF Guide for the Pacific Northwest Region. U.S. Department of Agriculture Forest Service, Portland, Oregon. VANDORP,D., ANDP. F. M. OPDAM. 1987. Effects of patch size, isolation and regional abundance on forest bird communities. Landscape Ecology 1:59-73.

is on landscape-scale spatialanalysisof amphibiandecline in California. He has 1 _ workedfor the Pacific t _ Southwest Research s Dt _....................................................................... _ Southwest Regional Office of the U.S. For_ est rervice. Is re_"_ _ search interests i nclude studyof the relag_ tionship between i l_ economic growth and environmental quality. He received his M.A. and B.A. in ecoBerkeley. nomicsfromthe University California, of _L.! _^81| _
a

.+ |
|

:aS

<. m.

You might also like