Habitat loss and fragmentation have been, and continue to be, the primary cause of extinction at all spatial scales, from local populations to the scale of the earth (Millenium Ecosystem Assessment, 2005). Species that are restricted to areas within which habitat is converted to something completely different disappear quickly; as happens, for instance, when forest is cleared. But these are not the only species affected, as habitat loss may significantly reduce the viability of species that occur also outside the converted area but whose long-term persistence depended on the populations that were lost. Such species comprise the extinction debt, the species that are predicted to go extinct, sooner or later, due to loss of a part of their habitat. This may happen at different spatial scales. Soga & Koike (2013) have analyzed the extent of extinction debt in butterflies at the scale of woodlands in a part of the Tokyo metropolitan area, where forest cover declined to one-quarter in 40 years since 1971, while the remaining area became highly fragmented into small forest patches. Soga & Koike (2013) report significant extinction debt especially in small forest fragments that had lost much of their area since 1971.
Extinction debt at the scale of individual habitat fragments is indeed expected to be greatest in the case of fragments that have been reduced to a small size within a short period of time. One reason is that the remaining habitat may not represent all the ecological conditions that were represented in the original area, and hence some species do not find suitable conditions in the remaining area. The second reason is that, very generally, the prospects are not good for long-term persistence of small populations in small habitat patches regardless of habitat quality. Any extinction debt in small patches is typically paid fast because small populations have a high risk of extinction due to demographic and environmental stochasticities (Lande, 1993) and inbreeding depression (Saccheri et al., 1998). One may attempt to reduce the extinction risk of small remnant populations by improving habitat quality, as suggested by Soga & Koike (2013), but this will not solve the generic predicament of small populations. Small isolated populations, even if they would survive the immediate hazards under favorable environmental conditions, will gradually accumulate a high genetic load (Mattila et al., 2012) and will ultimately perish because of reduced fitness.
At the landscape level, species may persist as metapopulations, networks of local populations in which recolonizations compensate for local extinctions. Soga & Koike (2013) found no evidence for either the current or the past landscape connectivity affecting species richness in the forest fragments that they studied. However, a more comprehensive analysis of the entire patch network rather than a set of focal fragments would be a more powerful way to examine possible connectivity effects and extinction debt at the landscape level. Extinction debt can certainly occur at the level of metapopulations, when habitat loss and fragmentation reduce the numbers and areas of, and the connectivities among, the remaining habitat patches, and species thereby drop below their species-specific extinction thresholds. An important additional twist in extinction debt at the patch network level is the speed with which metapopulations approach the new equilibrium, which may be network-wide extinction, following habitat loss and fragmentation. The dynamics slow down when the metapopulation is located close to its extinction threshold (Hanski & Ovaskainen, 2002), which is a general result about dynamic systems at critical transitions (Scheffer et al., 2012). This is an important result, because it means that many rare species living in highly fragmented landscapes may take an especially long time to go extinct, even though they are set to go extinct, and hence these communities have an especially large extinction debt.
Soga & Koike (2013) attempted to map extinction debt in a network of 182 forest fragments by comparing current species richness as predicted by landscape features (in practice by patch area) with the equilibrium species richness predicted by the species-area relationship and estimated from habitat fragments that had remained relatively stable. Although conceptually justified, this is a complex calculation with much uncertainty, and in the case of Soga & Koike (2013), it essentially assumes that the dynamics of local populations are independent in different patches. Looking at the map of the study area (Soga & Koike, 2013: fig. 4), this is unlikely to be the case: many fragments are located close to each other. An alternative approach would be to fit a metapopulation model to the empirical data and use that model to examine the dynamics of the species at the network level (see, e.g. Hanski, Moilanen & Gyllenberg, 1996). Admittedly, this approach would also involve much uncertainty unless much more data would be available to parameterize the model.
Soga & Koike (2013) have presented convincing evidence for extinction debt at the level of individual habitat fragments. It is likely that extinction debt also occurs at the landscape level in their study area, although this would require additional data and modeling to demonstrate. A recent study has examined extinction debt in the Brazilian Amazon (Wearn, Reuman & Ewers, 2012). The larger the area, the longer it takes for extinctions to occur and, which is the positive angle to extinction debt, more time there is for us to pay the extinction debt without allowing the species to go extinct.