Fragmentation and habitat loss

"Fragmentation" is one of the most used words in conservation and one of the most ambiguous. It can mean the loss of habitat, the breaking of what remains into pieces, or both. This series measures it the way GuidosToolbox does, around every cell of a map rather than patch by patch, as the US Forest Service and the European Commission's Joint Research Centre have done for national and global forest assessments. This first part separates fragmentation from habitat loss, then measures the simplest and most used of GuidosToolbox's indices, the forest area density (FAD): the share of forest in a window around every forest cell, at one scale and at five. At the end you will have a map of how fragmented a forest is at the scale of a few hundred metres and at the scale of several kilometres, and the numbers to report.

The idea#

Two processes behind one word#

As a process, the fragmentation of a habitat usually does four things at once: there is less habitat, more patches, smaller patches and more isolated patches (Fahrig, 2003). The first is habitat loss; the other three describe the configuration of what remains, and are called fragmentation per se.

Habitat amount and fragmentation — Four landscapes of 150 × 150 cells drawn by Terraflux's neutral landscape models: 60% or 20% habitat (rows), aggregated or fragmented (columns). Down a column the habitat is lost; along a row it is fragmented at a constant amount. Real landscapes mostly move along the diagonal, from the top left to the bottom right.

The distinction matters because the two have different effects. Reviewing more than a hundred studies, Fahrig (2003) found that habitat loss has large, consistently negative effects on biodiversity, while fragmentation per se has weaker effects, as often positive as negative. A later review went further: most significant effects of fragmentation per se are positive (Fahrig, 2017). Others disagree, stressing that large intact areas are irreplaceable for many specialist species and that the positive effects often come from generalists and edge species (Fletcher et al., 2018). And at the scale of the planet, Haddad et al. (2015) found that 70% of the world's forest is within 1 km of an edge, and that experiments of fragmentation consistently report losses of species and of ecosystem functions over time.

You do not need to take sides to draw the practical lesson: always measure the amount of habitat alongside its configuration, and interpret a measure of configuration knowing the amount. In the figure above, the number of patches is 73 and 145 at 60% habitat, 95 and 767 at 20%: it changes with both.

Fragmentation around every cell#

Patch metrics (landscape metrics) give a number per patch, class or map. Riitters et al. (2000) proposed another view for continental assessments: move a window across the map, and describe the habitat in the window around every cell. The result is a map, which shows where the forest is fragmented, not just how much, and which does not depend on how patches are delimited. The simplest such measure is the share of habitat in the window.

Key idea: forest area density

For a forest cell, its forest area density is the share of forest among the cells with data in the square window of \(w \times w\) cells centred on it:

\[ \text{FAD} = 100 \times \frac{\text{forest cells in the window}}{\text{cells with data in the window}} \]

Missing cells (outside the map, or with no data) are left out of the denominator. FAD is given only for forest cells; background cells get GuidosToolbox's code 101, missing ones 102. GuidosToolbox reports it in six classes (Vogt & Riitters, 2017): rare (below 10%), patchy (10 to 40%), transitional (40 to 60%), dominant (60 to 90%), interior (90 to 100%) and intact (100%: no break in the window).

FAD of a toy forest — A forest of 15 × 15 cells run through fragmentation.fad with a window of 7 cells. Left: the window of the cell marked in red holds 34 forest cells of 49, so its FAD is 69% (dominant). Right: the FAD of every forest cell. Cells deep in the large block score in the 80s and 90s; cells of the narrow strip, surrounded by fields, in the 20s and 30s.

The window's side is the observation scale. With cells of 25 m, a window of 27 cells covers 675 × 675 m (46 ha); one of 243 cells, 6 km. A forest cell can be intact within 175 m and patchy within 6 km: that is not a contradiction but two answers to two questions, and GuidosToolbox's multiscale analysis (FAD-MS) gives five of them at once, with windows of 7, 13, 27, 81 and 243 cells, and their average (Vogt & Riitters, 2017).

When to use it#

Mapping where the habitat is fragmented, at the scale that matters to a species or a process: where the forest is intact, where it is dissolving into fields.
Reporting fragmentation of a region or a country: the share of the forest in each class. The FAO's State of the World's Forests 2020 reported the world's forest by FAD's classes at several scales (Vogt et al., 2019), as the US Forest Service does for the forests of the United States (Riitters et al., 2002; Riitters & Wickham, 2012).
Following change between two dates (part 3).
Not for the structural role of each cell (core, edge, bridge): that is MSPA. Not for connectivity for a species with a dispersal distance: that is the habitat networks. FAD measures the amount of habitat around each cell; its arrangement in the window needs the other indices of part 2.

What you need#

A map of the habitat, coded as GuidosToolbox expects: 0 missing, 1 background, 2 foreground (the habitat), and optionally 3 and 4 for special backgrounds (3 a background that fragments, such as roads; 4 one that does not, such as water that a forest species simply does not use). Or any map of classes, with the classes of the habitat given in foreground: no reclassification is needed.

We use GuidosToolbox's sample forest map, input.tif (1,000 × 1,000 cells), the map of the habitat patterns tutorials. In the application, Templates > Patterns (GuidosToolbox) > Fragmentation of a forest starts a model with FAD-MS and FOS on a forest drawn by chance; replace its first node by a raster.read of your map.

Build the model#

The forest map, its FAD in a window of 27 cells, and its FAD at five scales, with a Plot of the classes at 27 cells.

Model file (TOML)

format = "terraflux/1"

[[nodes]]
id = "forest"
op = "raster.read"
params = { path = "input.tif" }
note = "0 missing, 1 background, 2 forest."

[[nodes]]
id = "fad"
op = "fragmentation.fad"
inputs = { map = "forest.raster" }
params = { window = 27 }
note = "The forest's share around every forest cell; its six classes."

[[nodes]]
id = "multiscale"
op = "fragmentation.fad_multiscale"
inputs = { map = "forest.raster" }
note = "Windows of 7, 13, 27, 81 and 243 cells, and their average."

[[plots]]
id = "plot"
title = "FAD, 27 cells"
layers = [{ from = "fad.classes" }]

Read the map (node 1): raster.read, path the file. Its output map is the forest map.
FAD at one scale (node 2). In the library's Fragmentation, click fragmentation.fad, and drag from node 1's map to its map. Set window to the side of the window in cells (odd, so that it has a centre; default 7). With a map of many classes, set foreground to the habitat's class (or list of classes).
FAD at five scales (node 3): fragmentation.fad_multiscale, connected the same way. windows changes the scales (default 7, 13, 27, 81, 243).
Look: click node 2 and show its output classes in the Viewer: the six classes in GuidosToolbox's colours. Plots › Save as a new plot keeps that view with the model (here titled FAD, 27 cells).

The two operators read the same map and do not depend on each other. Each gives:

map: the density (0–100%) on the forest cells, 101 on the background, 102 on missing cells;
classes: the classes 1 (rare) to 6 (intact), with the same codes elsewhere;
stats: per window, the share of the forest in each class and the average density.

FAD-MS also gives scales, a stack of the density at each window, so you can look at any scale on its own.

Read the results#

One scale#

Density — The south of the map. Left of the handle, the density of forest within 27 cells of every forest cell (grey: not forest, as on the right); right, its six classes. The interiors of the large forests are dark green (interior and intact); their edges and the woods of the valley dominant and transitional; the tree lines across the fields patchy and rare.

Classes — The south of the map. Left of the handle, the density of forest within 27 cells of every forest cell (grey: not forest, as on the right); right, its six classes. The interiors of the large forests are dark green (interior and intact); their edges and the woods of the valley dominant and transitional; the tree lines across the fields patchy and rare.

FAD classes of the whole map — FAD's six classes over the whole map, window of 27 cells.

The stats table of node 2:

Class	Intact	Interior	Dominant	Transitional	Patchy	Rare	Average FAD
% of the forest	7.3	32.7	32.2	14.0	12.2	1.6	73.5%

Read it as: within 27 cells, 40% of this forest is interior or intact (90% forest or more around it), a third dominant, and 28% transitional to rare, in surroundings that are mostly not forest. The average density, 73.5%, summarises it in one number: a forest cell's surroundings are, on average, three-quarters forest.

Five scales#

Move the slider from the smallest window to the largest, and then to their average:

Window (cells) 7 — FAD's classes over the whole map at the five windows of FAD-MS and their average. As the window grows, the intact class disappears (no window of 81 cells is all forest), interior shrinks to the cores of the largest forests, and transitional and patchy spread: at the scale of 6 km, this is a forest in a mixed landscape. Numbers: % of the forest in each class.

Window (cells) 13 — FAD's classes over the whole map at the five windows of FAD-MS and their average. As the window grows, the intact class disappears (no window of 81 cells is all forest), interior shrinks to the cores of the largest forests, and transitional and patchy spread: at the scale of 6 km, this is a forest in a mixed landscape. Numbers: % of the forest in each class.

The average density falls steadily with the window, from 85% at 7 cells to 51% at 243: this forest is solid at the scale of a few hundred metres and only transitional at the scale of several kilometres. That is a result, not a nuisance: report the curve, rather than choose one scale. The FAD-MS average combines the five scales into one map of five classes (intact cannot be reached by an average, so interior includes it); a cell is "interior" there only if its surroundings are forest at every scale.

FAD-MS's stats also gives AVCON at every scale, the average density over the whole area with data (not only over the forest): 36.5% at 7 cells, 22.0% at 243. It falls as the forest is lost, even where the forest left is not fragmented, so it combines amount and fragmentation in one indicator.

Report it

Report the map (source, date, cell size, what counted as forest), the window(s) in cells and in metres, the share of the forest in each class at each window, and the average density. For one scale, choose it from the ecology of the species or process (the distance of edge effects, a home range); when in doubt, report FAD-MS.

Settings that change the answer#

The window: above. In metres, it depends on the cell size: 27 cells are 675 m at 25 m, 2.7 km at 100 m.
What is foreground: the classes of the habitat. Forest only, or forest and woodland? A definition that leaves out tree lines and small woods makes the forest look less fragmented.
Special backgrounds (3 and 4): GuidosToolbox's codes for backgrounds that fragment the forest (3, roads for example) and that do not (4, for example water that a forest species simply does not use). Coding them apart says what counts as a break in the forest.
The map's resolution: a finer map sees more small openings and tree lines; Wickham and Riitters (2019) found forest fragmentation markedly different with data of 1 m than of 30 m.

Common pitfalls#

Reading FAD as fragmentation per se. FAD is the amount of forest around a cell. A forest cell with 50% forest around it may be at the edge of a large block or in a scatter of small woods; the indices of part 2 tell them apart.
Comparing FAD at different windows or cell sizes. Compare like with like.
Forgetting the map's border. Near the border, windows are cut: the density is computed on the cells with data only, so it is not biased, but it rests on fewer cells.
A map coded 0/1. GuidosToolbox reads 1 as background and 2 as foreground; a map with forest coded 1 would make the forest the background. Give foreground, or reclassify.
Degrees. On a map in degrees (as this sample), cells are not square in metres and differ by latitude; for scales in metres, project first.

Going further#

Part 2: the arrangement of the forest in the window (connectivity, contagion, entropy), % cover maps and the landscape mosaic.
Part 3: the change of fragmentation between two dates, and GuidosToolbox's reports.
Habitat patterns: MSPA, the structural role of every forest cell.
Neutral landscapes: landscapes of controlled amount and fragmentation.
The operator reference: fragmentation.fad, fragmentation.fad_multiscale.

References#

Fahrig, L. (2003). Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution, and Systematics, 34, 487–515. https://doi.org/10.1146/annurev.ecolsys.34.011802.132419
Fahrig, L. (2017). Ecological responses to habitat fragmentation per se. Annual Review of Ecology, Evolution, and Systematics, 48, 1–23. https://doi.org/10.1146/annurev-ecolsys-110316-022612
Fletcher, R. J., Jr., Didham, R. K., Banks-Leite, C., Barlow, J., Ewers, R. M., Rosindell, J., et al. (2018). Is habitat fragmentation good for biodiversity? Biological Conservation, 226, 9–15. https://doi.org/10.1016/j.biocon.2018.07.022
Haddad, N. M., Brudvig, L. A., Clobert, J., Davies, K. F., Gonzalez, A., Holt, R. D., et al. (2015). Habitat fragmentation and its lasting impact on Earth's ecosystems. Science Advances, 1(2), e1500052. https://doi.org/10.1126/sciadv.1500052
Riitters, K., Wickham, J., O'Neill, R., Jones, B., & Smith, E. (2000). Global-scale patterns of forest fragmentation. Conservation Ecology, 4(2), 3. https://doi.org/10.5751/ES-00209-040203
Riitters, K. H., Wickham, J. D., O'Neill, R. V., Jones, K. B., Smith, E. R., Coulston, J. W., et al. (2002). Fragmentation of continental United States forests. Ecosystems, 5(8), 815–822. https://doi.org/10.1007/s10021-002-0209-2
Riitters, K. H., & Wickham, J. D. (2012). Decline of forest interior conditions in the conterminous United States. Scientific Reports, 2, 653. https://doi.org/10.1038/srep00653
Vogt, P., & Riitters, K. (2017). GuidosToolbox: universal digital image object analysis. European Journal of Remote Sensing, 50(1), 352–361. https://doi.org/10.1080/22797254.2017.1330650
Vogt, P., Riitters, K. H., Caudullo, G., & Eckhardt, B. (2019). FAO – State of the World's Forests: Forest fragmentation (EUR 29972 EN). Publications Office of the European Union. https://doi.org/10.2760/145325
Wickham, J., & Riitters, K. H. (2019). Influence of high-resolution data on the assessment of forest fragmentation. Landscape Ecology, 34(9), 2169–2182. https://doi.org/10.1007/s10980-019-00820-z