The taxonomy of the endemic tribe Helictopleurini (Coprinae, Scarabaeidae) is well known (Lebis 1960
; Paulian 1986
and other papers; Montreuil 2005
, in press
and other papers). We recorded the sampling information for all specimens in the main collections of the Paris National Museum of Natural History. The collections include 51 species for which locality information is available, sampled from 126 distinct localities (a
) during 1875–1990 (3341 specimens). Half of the specimens have been collected prior to 1926 but almost none since the mid-1970s. The remaining nine species either lack sampling information (three species) or there are no specimens in Paris. We suspect that several of these latter species should be synonymized with the better known species.
Figure 1 Maps showing (a) the historical sampling localities during 1875–1990, (b) our sampling localities during 2002–2006, (c) the sampling localities of 21 apparently extinct species apart from H. undatus and (d) the localities for H. undatus (more ...)
During the years 2002–2006, we sampled Helictopleurini using dung and carrion-baited traps. Our sample of 4880 specimens was collected from 61 localities (b
), including larger samples from Ranomafana National Park (NP), Masoala NP, Makira Reserve, Andasibe NP, Ambila-Lemaintso, Manombo reserve, Isalo NP, Zombitse-Vohibasia NP and Andahohelo NP. Smaller samples were collected by the personnel of forest reserves in 52 localities across Madagascar using trapping kits provided by us. Our samples include four new species (Montreuil 2005
, in press
A single map of forest cover change between the years 1970, 1990 and 2000 was provided by Conservation International (CI) at approximately 30
m resolution (Harper et al
. 2005, unpublished data). This map was reclassified to single out forest cover for the year 2000. The main sources of data were the Inventaire Ecologique et Forestaire National (IEFN) classification of Landsat Thematic Mapper 5 data for the year 1993 and estimates of forest cover for the year 1999 carried out by the Joint Research Centre (JRC)–Space Application Institute, Ispra, Italy, using SPOT-4 data. Where possible, CI used additional SPOT images to add details in the regions covered by dense cloud in the original IEFN and JRC images.
CI also provided a digitized version of the 1953 forest cover map produced by Humbert et al. (1965)
, rasterized at the same resolution as the 2000 map. The original map was produced using aerial photographs and ground truthing. The 1953 study appears to have focused on mapping major forest blocks, as the map does not contain small fragments in remote areas that were present in the satellite images. We assume that the additional small fragments present in 1970 had not grown in the intervening years, and hence any forest cover present in 1970 but absent in the 1953 map was added to the latter.
The maps for 1953 and 2000 were summarized as percentage of forest cover in grid cells of 0.1° resolution (11.2
km at the equator). Using these maps, we calculated the extent of forest cover within the range of species x
at time t
(1953 or 2000) as
is the percentage of forest cover in cell j
is the distance between the sampling locality i
and cell j
in degrees, and the second summation is over the n
sampling localities for species x
. Thus, Fx,t
measures the average amount of forest in the surroundings of the historical sampling localities for species x
, giving decreasing weight to cells with increasing distance from the sampling localities. We assumed α
=10, which gives substantial (greater than 0.05) weight to distances up to 33.6
km. The absolute and relative forest losses were calculated as Fx,1953
As our sampling localities do not evenly cover all of Madagascar, it is possible that we have failed to sample a species because only a few or even none of our sampling localities were within its range. To account for this, we calculated the average distance of the n
historical sampling localities for species x
to our sampling localities as
is the distance (in degrees) from the i
th historical sampling locality to the j
th locality in our sampling. Since we used the value α
=1 in this calculation, 1° (112
km) distance has the weight 0.37.
Other explanatory variables include body size, the last year when the species was sampled prior to our sampling, the (log) number of historical sampling localities, the (log) number of individuals in the Paris collections and the range of the species, defined as the distance between the two most distant historical sampling localities. As the latter three variables are all strongly correlated, we calculated the first principal component as a general measure of past commonness (PC1). PC1 accounted for 90% of variation in the three original variables.