^^     Biodiversity and Conservation 13: 709?732, 2004. 
? 2004 Kluwer Academic Publishers. Printed in the Netherlands. 
Discriminatory power of different arthropod data sets 
for the biological monitoring of anthropogenic 
disturbance in tropical forests 
YVES BASSET'*, JACQUES F. MAVOUNGOU^ JEAN BRUNO MIKISSA', 
OLIVIER MISSA\ SCOTT E. MILLER', ROGER L. KITCHING' and 
ALFONSO ALONSO' 
Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Ancon, Panama City, Republic of 
Panama; 'Institut de Recherche en Ecologie Tropicale I.R.E.TICEN AREST, B.P. 13354, Libreville, 
Gabon; Ecole Nationale des Eaux et For?ts, B.P. 3960, Libreville, Gabon; Australian School of 
Environmental Studies, Griffith University, Brisbane, QLD 4111, Australia; Department of Systematic 
Biology, National Museum of Natural Histoiy, Smithsonian Institution, Washington, DC 20560-0105, 
USA; Cooperative Research Centre for Tropical Rainforest Ecology and Management, Australian 
School of Environmental Studies, Griffith University, Brisbane, QLD 4111, Australia; Smithsonian 
Institution!Monitoring and Assessment of Biodiversity Program, 1100 Jefferson Drive, SW. Suite 3123, 
Washington, DC 20560-0705, USA; ' Author for correspondence (e-mail: bassety@tivoli.si.edu; fax: 
-I-507-212-. 
Received 6 June 2002; accepted in revised form 24 February 2003 
Key words: Guilds, Parataxonomists, Predictor sets, Rarity, Taxonomic resolution 
Abstract. Arthropods were monitored by local parataxonomists at 12 sites of increasing anthropogenic 
disturbance (old and young secondary forests, savanna and cultivated gardens) at Gamba, Gabon. We 
report on the discriminatory power of different data sets with regard to the classification of sites along the 
disturbance gradient, using preliminary data accounting for 13 surveys and 142425 arthropods collected 
by Malaise, pitfall and yellow-pan traps. We compared the performance of different data sets. These were 
based upon ordinal, familial and guild composition, or upon 22 target taxa sorted to morphospecies and 
either considered in toto or grouped within different functional guilds. Finally we evaluated 'predictor 
sets' made up of a few families or other target taxa, selected on the basis of their indicator value index. 
Although the discriminatory power of data sets based on ordinal categories and guilds was low, that of 
target taxa belonging to chewers, parasitoids and predators was much higher. The data sets that best 
discriminated among sites of differing degrees of disturbance were the restricted sets of indicator families 
and target taxa. This validates the concept of predictor sets for species-rich tropical systems. Including or 
excluding rare taxa in the analyses did not alter these conclusions. We conclude that calibration studies 
similar to ours are needed elsewhere in the tropics and that this strategy will allow to devise a 
representative and efficient biotic index for the biological monitoring of terrestrial arthropod assemblages 
in the tropics. 
Introduction 
Since water pollution is often transient and unpredictable, biological monitoring 
may be more appropriate than traditional chemical evaluation of water quality to 
assess contamination of aquatic ecosystems (Gu?rold 2000). Plants, which are 
traditionally used as measures of habitats as well as disturbance in terrestrial 
710 
systems (e.g., Watt 1998), may be less useful than invertebrates for the biological 
monitoring of aquatic systems. This pragmatic reason has driven the development of 
community-level analyses of the effects of anthropogenic disturbance on inverte- 
brates in aquatic ecosystems (e.g., Clarke 1993; Rossaro and Pietrangelo 1993; 
Crowns et al. 1997; Thorne and Wilhams 1997; Cu?rold 2000). The effects of 
pollution on freshwater invertebrate communities can be calculated routinely as an 
'index of biological integrity' (Karr 1991), and different statistical methods are 
available for the study of environmental impact on marine communities (Warwick 
1993). In contrast, such recipes and consensus are almost non-existent for terrestrial 
arthropods, particularly in the tropics (see Urzelai et al. (2000) for nematodes and 
O'Connell et al. (1998) for birds). 
It is important to stress that the ultimate goal of deriving a biotic index based on 
terrestrial arthropods is to monitor the effects of anthropogenic disturbance per se 
on arthropods (Basset et al. 1998). Arthropods represent a substantial proportion of 
all terrestrial biodiversity and, accordingly, their responses to disturbance are 
important. Documenting which species reacts, and how, to varying disturbance 
levels is more significant than assessing and monitoring habitat quality, which is 
rather trivial and can easily be assessed by vegetation censuses alone (Watt 1998). 
The sheer number of arthropods, however, leads to major challenges in fulfilling this 
goal. Specifically, difficulties attendant upon the study of terrestrial arthropods 
include (a) the high diversity of terrestrial assemblages; (b) the high diversity and 
complexity of terrestrial habitats (McGeogh 1998), with concomitant logistical 
problems in sampling (Kitching et al. 2001); and (c) the taxonomic impediment 
(Kitching 1993a). By way of example, compare the 56 families of macrobenthos 
collected by Thorne and Williams (1997) at various locations in the tropics, with the 
222 insect families encountered by Stork (1991) in a single event using insecticide 
fogging in the crowns of 10 Bornean trees. 
It is not surprising that most studies of anthropogenic disturbance on terrestrial 
invertebrates in the tropics focus on well-known, less speciose, taxa, usually 
restricted to the family level and to a specific feeding guild (e.g., Belshaw and 
Bolton 1993; Eggleton et al. 1996; Hill 1999; Intachat et al. 1999; Vasconcelos 1999; 
Davis et al. 2001; McCeogh 1998). This approach may be overly restrictive given 
the absence of any consensus on the appropriate choice of 'indicator' or 'umbrella' 
species, especially in the tropics (e.g.. Landres et al. 1988; Prendergast et al. 1993; 
Hammond 1994; Lawton et al. 1998; McGeogh 1998; Kotze and Samways 1999; 
Basset et al. 2001a; Moritz et al. 2001). Kitching (1993b) and Didham et al. (1996) 
have advocated the use of 'predictor sets' which comprise taxa representative of 
different functional groups ('guilds') as an alternative to the use of taxa selected on 
the basis of taxonomic tractability or familiarity (see also Collins and Thomas 
(1991) and Kremen et al. (1993) for similar arguments). Such 'predictor sets' are 
properly selected only following statistical analysis of a larger, more or less 
complete, data set including all taxa and the catches from several complementary 
sampling methods. Some studies have indeed widened their taxonomic focus to a 
whole order or a few families of different orders, thereby, often unintentionally, 
including representatives of different guilds (e.g.. Kremen 1992; Didham et al. 1998; 
711 
Kotze and Samways 1999; Chung et al. 2000; Kitching et al. 2000). Lawton et al. 
(1998) went further and included both invertebrate and vertebrate orders in their 
study of forest disturbance in Cameroon. They noted the huge effort necessary to 
implement such approaches ef?ciently. 
Recently, a novel approach that relies on the training and input of local parax- 
onomists has allowed the considerable widening of entomological investigations in 
the tropics (e.g., Janzen and Gauld 1997; Basset et al. 2000; Novotny et al. 2002). 
Properly used, this strategy yields higher numbers of statistical replicates that are, 
accordingly, more representative of the system studied (Basset et al. 2000). 
Adequate statistical replication represents a significant obstacle in conservation 
studies of highly complex environments, such as tropical rainforests. The paratax- 
onomist strategy employed in Guyana enabled us to achieve one of the first 
Before-After/Control-lmpact experiments, demonstrating unequivocally the in- 
fluence of selective logging on rainforest insects, despite the excessively low insect 
densities in the study system (Basset et al. 2001a). 
Presently, about 100000 valid species of insects are known from the Afrotropical 
region, but even conservative estimates, such as the scenario of Gaston and Hudson 
(1994), may see this number increase to about 600000 species (Miller and Rogo 
2002). Basic ecological information on described species of Afrotropical arthropods 
is fragmentary and often relates to a few localities only. The level of availability of 
this information also varies greatly from one taxon to another. Gaps in knowledge 
are evident, even for well-studied taxa (Miller and Rogo 2002). 
Entomological studies in Gabon have been few and follow these trends. The few 
recent checklists available for higher taxa are restricted to groups that are not 
particularly speciose, such as Mantodea (Roy 1973); Haliplidae and Dytiscidae 
(Bilardo and Rocchi 2000); Lucanidae (Maes and Pauly 1998); Brentidae (Bar- 
tolozzi and Sforzi 1997); and Apoidea (Pauly 1998). There are no recent reviews of 
agricultural and timber pests in Gabon, although Coccoidea and, in particular, the 
cassava mealybug and its parasitoids, have been well studied (e.g., Boussienguet et 
al. 1991). Ecological studies are likewise infrequent and targeted at a few groups 
such as cockroaches, dung beetles, fig wasps, bees or ants (Walter 1987; Anstett et 
al. 1995; Grandcolas 1997; Roubik 1999; Wetterer et al. 1999). The need for 
baseline information on Gabonese arthropods is obvious. 
The work presented in this paper is part of a wider project aimed ultimately at 
providing baseline entomological data and the assessment of anthropogenic dis- 
turbance on local arthropod faunas within the Gamba Complex, Gabon. Arthropod 
activity is being monitored by trained and supervised local parataxonomists in four 
distinct habitats of increasing degrees of anthropogenic disturbance. The taxonomic 
scope of the material collected is large and includes representatives of several 
feeding guilds. The structure and scope of this project have few equivalents to date 
in tropical Africa. 
In the present report, we consider a preliminary (but by no means small) data set 
which we use to analyse the discriminatory power of different subsets of data 
(including orders, families and morphospecies representative of different feeding 
guilds) against the disturbance gradient. We explore the question of whether a few 
712 
ecologically selected taxa ('predictor sets') may be suitable for the biological 
monitoring of anthropogenic disturbance at Gamba, as surrogates for the entire local 
fauna. In doing so, we also evaluate the need to sort specimens to a specific level (as 
opposed to sorting the material to familial or ordinal levels) and examine the 
contribution of rare species to the discriminatory power of different data sets. 
Methods 
Study area and sites 
The study area was in the Shell-Gabon oil concession of Gamba, within the Gamba 
Complex of Protected Areas in southeastern Gabon (approximately 2?43' S, 10?r E, 
25 m a.s.l.; see Thibault and Blaney (2001) and Doumenge et al. (2001) for 
background information about the area). The Gamba oil field includes a mosaic of 
old growth secondary rainforests, younger secondary rainforests and savanna areas. 
The latter result mainly from anthropogenic action. Primary rainforests are absent 
from the Gamba oil field, following the selective logging of Okoum? (Aucoumea 
klaineana Pierre), mostly over the past 20 years, but these forests are found 
elsewhere in the Gamba Complex. Botanical information about the area is available 
in Prins and Reitsma (1989). The mean annual temperature in the area is 26 ?C and 
annual rainfall ranges between 2000 and 2400 mm per year, with the major wet 
season from September to December (Prins and Reitsma 1989). The Gamba oil field 
has been active since 1967 and Gamba has grown from a small village in 1960 to a 
town of 8000 inhabitants (Bourgeais 2001). The earliest cultivated crop gardens of 
relative size were established near the town as recently as 1998 (see Bourgeais 
(2001) for a summary of environmental concerns in the Gamba Complex). 
We considered four distinct habitats of increasing anthropogenic disturbance (i.e., 
increasing forest clearing and introduction of exotic vegetation) and selected three 
sites (replicates) within each habitat. The four habitat types were: (a) the interior of 
old secondary rainforests, 'old forest'; (b) the edge of young secondary rainforests, 
'young forest'; (c) area of rainforest cleared to install oil rigs and subsequently 
invaded by savanna, 'savanna'; and (d) cultivated crop gardens, 'gardens'. Differ- 
ences between these habitats were obvious and readily noticeable to a non-biologist. 
The main characteristics of the study sites (coded A-L) are indicated in Table 1. 
Sites G-H and 1 were abandoned and active oil wells, respectively. They were 
established in 1980, 1980 and 1968, respectively. There were no old or recent oil 
spills at these three sites and none was burned during the study period. As far as 
possible, sites were spread through the concession, within an area of approximately 
13 X 11 km. The northwestern part of the concession, however, is more forested 
and, accordingly, most forest sites occurred there. The shortest distance between 
sites was ca. 600 m (sites B and G), whereas the longest distance between sites was 
ca. 15 km (sites A and F). 
713 
-o 
t2 
3Q 
?1 R,      'S-, 
a* 
<  -5- 
'I ? 
?. m <-> 
? =3  "I "S 
|3 I i 
% -?   s % Q Q 
i S i a 
i ? 
o   o   i?   o 
a a 
a a I -a" 
s a. S ? 
lu 
o    s    i=^    e "    aj 
5J    ^ 
?g W 
a -S- ?-" 
S -S ^ ^? 
a^ <? ^ S:? 
? -a 
^ a 
^  8. 
j3   c3   y   c3   ig 
a .3 ?? .a a 
I 
B 
g 
o   ^ 
O 
I 
?s 
's 
o   ^ 
aj   o 
^1 
aj 
'3 
fc 
a 
714 
Arthropod collecting and processing 
Each site was equipped with a similar array of traps targeting the flying and epigeic 
arthropods of the understorey and litter. At each site, these traps included: one 
ground Malaise trap, four yellow-pan traps set up on the ground and five pitfall traps 
buried in the ground. In addition, four flight-intercept traps were also set up in 
forested sites, but for ease of comparison among all sites, these data are not 
discussed further. The collecting methods were selected because they are simple, 
inexpensive, and behave more or less consistently across sites with respect to the 
profile of arthropods collected (Kitching et al. 2001). 
The collecting surface of one Malaise trap was 2.7 m" (model similar to Townes 
(1972); Sant? Traps, Lexington, Kentucky). Collecting fluid was 70% ethanol. 
Yellow-pan traps were 27 cm in diameter by 8 cm deep and filled with a mixture of 
water (ca. 80%), 70% ethanol (ca. 20%) and a few drops of liquid detergent to break 
the surface tension of the water. Pitfall traps were small 0.5-1 plastic cups and filled 
with the same water, ethanol and detergent mixture. At each site, a Malaise trap 
occupied the center of the set of traps, with four pitfall traps established to the north, 
south, east and west, 10 m distant from the Malaise trap. Four yellow-pan traps were 
set up at equal distances between the pitfall traps, again 10 m distant from the 
Malaise trap. The fifth pitfall trap was established 30 m north of the Malaise trap. 
The 120 traps were operated for 3 days and surveyed weekly (= one survey), by a 
team of six local parataxonomists, who were trained specifically for this project (see 
Basset et al. (2000) for similar training in Papua New Guinea and Guyana). 
Although the project is ongoing, we consider here preliminary data from the first 13 
surveys obtained from July to November 2001. 
The material collected was first sorted to families or higher taxa (see exceptions 
below) by the parataxonomists in a laboratory equipped with four microscopes and 
two computers. The material, belonging to 22 target taxa (Table 2), was isolated and 
pinned (or otherwise dry-mounted), with each individual identified by a unique 
specimen number. The target taxa were eventually sorted by morphospecies (i.e., 
unnamed species diagnosed using standard taxonomic techniques, sensu Cranston 
1990) by the parataxonomists. Formal taxonomic study of this material is ongoing. 
The 22 target taxa were selected using three main criteria: (a) they were well 
represented in the samples; (b) they were workable taxonomically and sought after 
by taxonomists who had expressed interest in the material; and (c) they represented a 
variety of functional guilds and orders (Table 2). All target taxa of beetles further 
represented 'focal groups' sensu Hammond (1994), that is, they were abundant and 
easily sorted. Taxa were assigned to the following feeding guilds, following Moran 
and Southwood (1982) and Stork (1987): chewers, sap-suckers, pollinators, epi- 
phyte grazers, fungal-feeders, insect predators, other predators, parasitoids, wood- 
eaters, scavengers, ants, 'tourists' (i.e. non-feeding residents), and 'unknown' (i.e. 
taxa for which the sorting resolution did not allow assignment of the material to a 
particular guild; these were not retained for analyses). All of the resulting in- 
formation was data-based using the program BIOTA (Colwell 1997). Voucher 
specimens will be deposited at the Smithsonian Institution (Washington, DC) and in 
an institution in Gabon to be determined. 
715 
Table 2. The 22 target taxa selected, with the total number of individuals collected and morphospecies 
sorted for the first 13 surveys. 
Order Target taxa Guild Individuals Morphospecies 
Mantodea Mantodea" Predators 9 3 
Orthoptera Acridoidea" Chewers 205 13 
Hemiptera Fulgoroidea Sap-suckers 1401 94 
Hemiptera Membracidae Sap-suckers 13 7 
Cole?ptera Cerambycidae Wood-eaters 124 25 
Cole?ptera Chrysomelidae Chewers 536 85 
Cole?ptera Buprestidae Wood-eaters 16 5 
Cole?ptera Scarabaeidae Chewers + scavengers 421 + 449 32 -1- 28 
Cole?ptera Coccinellidae Predators 540 18 
Cole?ptera Histeridae Predators 287 10 
Cole?ptera Cleridae Predators 8 4 
Cole?ptera Tenebrionidae Scavengers 389 21 
D?ptera Asilidae Predators 145 22 
D?ptera Dolichopodidae Predators 3263 38 
D?ptera Syrphidae Predators 110 17 
D?ptera Tephritidae Chewers 72 9 
D?ptera Pipinculidae Parasitoids 6 3 
Neuroptera Neuroptera" Predators 54 9 
Hymenoptera Ichneumonidae Parasitoids 995 80 
Hymenoptera Chalcidoidea'' Parasitoids 1453 73 
Hymenoptera Formicidae Ants 37202 Not yet sorted 
Hymenoptera Apoidea? Pollinators 273 25 
Totals 
- - 
47971 621 
"One family; '11 families; 'six families;    12 families. 
There were a few exceptions to the sorting and mounting pattern. Non-insect 
material was mostly sorted to order. Adults of Lepidoptera were not sorted to 
families, since this wet material was useless. Some individuals, notably within the 
Diptera (both acalypterate and calypterate) and the Cucujoidea, were also not 
assigned to families. Chalcidoidea smaller than 2 mm were sorted, but not mounted 
or morphotyped. Formicidae is the only target taxon that has yet to be sorted to 
morphospecies. 
Statistical methods 
As sampling effort was identical at each site, we have pooled the results of all traps 
at each site for the analyses presented here (i.e. we have summed the occurrence of 
each taxon at each site). We examined the discriminatory power of 11 data sets (the 
matrix of taxa X sites with number of individuals as the elements in each case) 
against the classi?cation of the sites along the disturbance gradient. First, we 
considered higher taxa (orders and families) and guilds. Second, we considered all 
morphospecies of target taxa that belonged to a particular guild. Data sets which 
included enough morphospecies for the analyses included: chewers, parasitoids, 
predators, pollinators, sap-suckers, scavengers and wood-eaters. Subsequently, we 
considered a data set incorporating all morphospecies from all guilds. Sap-suckers 
716 
included only phloem-feeders and predators included only insect predators (see 
Table 2). 
Rare species represent a statistical challenge when comparing different samples, 
as the information provided by singletons is very low (Novotny and Basset 2000). 
'Rare' species may be subsisting at genuinely low population levels or they may be 
apparently rare because (a) they are poorly collected by the chosen sampling 
methods; (b) sampling effort has been insufficient; (c) they may be seasonal or have 
restricted timing of activity; or (d) they may have been collected from marginal 
habitats ('vagrant species', see Novotny and Basset 2000). We analyzed each of the 
11 data sets both with and without the rare species. In the latter case we considered 
only taxa that were represented by 12 or more individuals (i.e. an average of one 
individual collected at each site). We present here detailed results for the data sets 
that exclude rare taxa (see Discussion). There were insufficient data to justify an 
analysis of the wood-eating guild after the exclusion of rare morphospecies. 
We considered four statistics that describe the discriminatory power of the 
different data sets, and calculated them for data sets both with and without rare taxa: 
(a) Standard deviation of multivariate scores: Non-metric multi-dimensional 
scaling has been advocated as a powerful method to study the similarity of samples 
( ';S-diversity' ) in species X samples matrices, as the similarity among samples can 
be computed using a variety of distances or similarity indices (Clarke 1993). 
However, the arbitrariness of orientation of the axes extracted is a serious drawback 
when one is primarily interested in comparing ordinations (Clarke 1993), as is the 
case here. Accordingly, we have chosen to use a more straightforward approach by 
quantifying /3-diversity using detrended correspondence analysis (DCA) with Hill's 
scaling, using untransformed data (Ter Braak and Smilauer 1998). The differences 
between the scores of any two sites on the first axis of the DCA represent a measure 
of species turnover between these two sites. The standard deviation of the scores on 
Axis 1 also represents a measure of the spread of the value within the sites. These 
analyses were computed using the programme CANOCO (Ter Braak and Smilauer 
1998). 
(b) Modified beta diversity of Whittaker: We calculated /3-diversity (?-2) (sensu 
Whittaker 1960) for each data set, using the modified formula of Harrison et al. 
(1992): 
?-2={[(5/a_)-l]/(A^-l)}Xl00 
where S is the total number of species collected, a^^^ is the maximum within-taxon 
richness per sample and A^ is the number of samples per taxon (12 sites in this case). 
?-2 is insensitive to species richness and ranges from 0 (no turnover) to 100 (every 
sample has a unique set of species). 
(c) ANOSIM: We used a non-parametric analysis of similarity (ANOSIM) to test 
for differences in the rank similarities of sites grouped by habitats. This is analogous 
to a one-way analysis of variance (Clarke 1993). ANOSlMs were calculated based 
upon Bray-Curtis distances and their significance was tested with 5000 random 
permutations using the programme BioDiversity Pro (McAleece et al. 1997). 
717 
(d) Non-parametric analysis of variance on multivariate scores: We also per- 
formed non-parametric analyses of variance (Kruskal-Wallis tests) on the scores of 
Axis 1 of the DCA grouped by habitats. 
Statistics (a) and (b) reflect the spread of the sites (the higher the better, the 
decrease of ?S-diversity being a measure of the loss of information), whereas 
statistics (c) and (d) evaluate the classi?cation of sites along the disturbance gradient 
(the more significant the better, the decrease in significance being also a measure of 
the loss of information). We compared the first two statistics for each data set both 
with and without rare species by Wilcoxon signed-rank tests. Since the inclusion of 
additional taxa in the data sets could increase their discriminatory power, we tested 
the independence between the number of taxa included in the analyses and the first 
two statistics by regression, for all data sets. 
The final set of analyses evaluated the usefulness of predictor sets (Kitching 
1993b) for our data. To identify the taxa to be included in the predictor sets, 
Kitching et al. (2000) used principal component analysis (PCA) to obtain a linear 
combination of taxa scores that could be used as an indicator of habitat quality. In 
the present account we are less concerned about deriving an index of habitat quality 
than we are about exploring which taxa should be considered in calculating a 
putative index and, indeed, whether this is a practical strategy at all. The two-way 
indicator species analysis (TWINSPAN) further represents an alternative method to 
PCA, free of the assumptions of linearity. Another method, the indicator value index 
(INDVAL), has also been proposed as an alternative which overcomes several 
limitations of TWINSPAN (DufrSne and Legendre 1997; see McGeogh et al. (2002) 
for an application). In short, this indicator index for a particular species is in- 
dependent of the relative abundance of other species (comparisons between different 
species may be impeded by different catchability, activity patterns, behaviors, etc.), 
and there is no need to use 'pseudospecies'. The significance of each species 
indicator value is tested by a site randomisation procedure (DufrSne and Legendre 
1997). INDVAL was computed for the two best data sets (see Results: families for 
higher taxa and all target taxa for morphospecies), with and without rare species. We 
used PC-ORD, with Monte Carlo permutation tests (1000 permutations; McCune 
and Medford 1999) for this analysis. 
Finally, we combined taxa with significant indicator value indices {P < 0.05) in 
new matrices for families and target taxa, including and excluding rare species, and 
re-calculated the four statistics (a)-(d) above, for these derived data sets. We have 
termed these 'indicator families' and 'indicator target taxa', respectively. 
Results 
Overall results 
In total, 142425 arthropods were collected during the 13 surveys, which included 29 
orders and at least 175 families. Different trap types yielded a similar number of 
individual arthropods (Malaise traps: 55 108, pitfalls: 45571, yellow pans: 41746). 
718 
Table 3. Distribution of guilds (sum of individuals collected) among the different habitats studied, for the 
first 13 surveys. 
Variable Old forest Young forest Savanna Gardens 
Ants 15751 7858 3246 10347 
Chewers 369 445 674 972 
Epiphyte grazers 327 358 446 671 
Fungal-feeders 979 1457 822 2288 
Insect predators 571 907 1294 4046 
Other predators 662 642 436 329 
Parasitoids 1112 961 1000 2473 
Pollinators 1 8 70 194 
Sap-suckers 1263 1740 701 4380 
Scavengers 6061 8109 3131 10866 
Tourists 8867 10588 5152 13580 
Wood-eaters 261 2518 229 789 
Unknown 412 1079 423 560 
Total individuals 36625 36672 17624 51504 
Total morphospecies 203 233 163 317 
Since ants, the major catch in pitfall traps, have not yet been sorted to morpho- 
species, the total number of individuals represented by morphospecies was lower in 
the pitfall than in other types of traps (Malaise traps: 2768, pitfalls: 626, yellow 
pans: 3620). The material assigned to morphospecies resulted in the identification of 
621 units and represented 7.6% of the total material collected (Table 2). The 
taxonomic composition of the material collected will be presented elsewhere, but we 
noted the collection of two specimens of Dilaridae (Neuroptera), representing the 
first record of this family in Gabon and, indeed, in tropical Africa (one species is 
known from South Africa: Oswald 1998). 
During the study period, gardens yielded more individuals and morphospecies 
than, notably, savanna habitats (Table 3). The yields from old and young forests 
were similar, but with apparent differences in the guild structure between these two 
habitats (Table 3). Differences among the various types of habitat studied in terms 
of productivity, species richness and guild structure are indisputable, but will be 
discussed elsewhere, with more appropriate data. 
Data sets including higher taxa 
Among higher taxa (orders, families and guilds), the data set with the most 
discriminant power was that at the family level, with or without rare taxa (Figure 1, 
Table 4; DCA plots of data sets including rare taxa are not shown). The standard 
deviation of the DCA scores on Axis 1 (35.1% of the total inertia) was relatively 
high for this data set and its Kruskal-Wallis test was significant. However, this data 
set, with or without rare families, did not effectively classify the sites along the 
disturbance gradient (Figure 1), its ?-diversity was low and its ANOSIM not 
significant (Table 4). The data set based on guilds was the worst of the three data 
sets and, in general, the inclusion or exclusion of rare taxa identified few differences 
within these data sets (Table 4). 
719 
ORDERS 
E 
xL 
J 
't 
A 
? 
G 
K 
H 
1 
1) 
?      F 
? C 
?                                                                                                         ] 
h D 
L     cr 
Bi A 
G K 
H          ^ 
GUTT.DS 
-0,5 -1,5 
Figure 1. Plot of study sites within the plane formed by the first two axes of DCA for higher taxa and 
groups (orders, families and guilds), excluding rare taxa. Sites are identified by their codes (Table 1) and 
corresponding habitats are indicated by the following symbols: ? old secondary forest; O young 
secondary forest; + savanna; and X gardens. 
Table 4. Discriminatory power of the different data sets, with and without rare taxa: number of taxa 
included in the analyses (n), standard deviation of the site scores on Axis 1 of the DCA (SD), results of 
non-parametric tests for score differences between habitats (Kruskal-Wallis tests: Wand P), yS-diversity 
(?-2) and results of ANOSIM (R and P). 
Data set Excluding rare taxa Including rare taxa 
n SD W P ?-2 R P n SD W P ?-2 R P 
Orders 20 0.350 2.69 0.442 0.48 0.40 2.680 29 0.350 2.69 0.442 2.37 0.40 2.320 
Families 108 0.488 8.64 0.034 0.83 0.49 1.340 175 0.489 8.64 0.034 3.64 0.49 1.220 
Guilds" 14 0.383 1.26 0.740 0.00 0.46 1.079 - - - - - - - 
Target taxa 95 1.834 9.97 0.019 5.06 0.79 0.039 621 2.074 9.97 0.019 16.92 0.81 0.039 
Chewers 14 1.413 9.46 0.024 4.55 0.74 0.039 140 2.226 9.97 0.019 16.36 0.83 0.019 
Parasitoids 18 1.252 9.15 0.027 5.30 0.73 0.019 155 1.464 9.67 0.022 15.63 0.79 0.019 
Pollinators 6 0.742 6.33 0.096 1.52 0.50 2.220 25 0.841 6.33 0.096 7.14 0.50 1.739 
Predators 33 1.414 9.46 0.024 1.62 0.80 0.039 121 1.508 9.46 0.024 11.66 0.81 0.039 
Sap-suckers 15 1.220 9.15 0.027 6.06 0.60 0.059 101 1.429 9.05 0.029 18.73 0.64 0.079 
Scavengers 9 2.572 8.95 0.030 4.55 0.58 0.400 49 2.643 9.05 0.029 15.66 0.59 0.299 
Wood-eaters - - - - - - - 30 1.186 11.00 0.012 21.21 0.41 1.739 
Indicator families 26 0.766 10.39 0.016 0.76 0.78 0.019 27 0.761 10.39 0.016 1.14 0.78 0.019 
Indicator target taxa 32 1.254 10.39 0.015 1.68 0.90 0.019 44 1.458 10.39 0.016 1.72 0.92 0.019 
"Data set identical with or without rare 
wood-eating morphospecies. 
guilds.   Insufficient data for the analyses excluding rare 
Data sets including morphospecies 
Data sets that included morphospecies belonging to different guilds had different 
discriminatory power. They ranked as follows (from highest to lowest): chewers, 
parasitoids, predators, sap-suckers, scavengers, wood-eaters and pollinators, again 
regardless of whether or not rare morphospecies were included (Figure 2, Table 4). 
720 
In spite of the relatively low number of morphospecies (n = 14), the data set on 
chewers classified the sites along the disturbance gradient effectively (Figure 2). 
The Spearman rank correlation coefficient between the scores of Axis 1 of the DCA 
(without rare species, 29.9% of total inertia) and habitats, coded as a categorical 
variable, was 0.907 (P < 0.001) and its ANOSIM was significant (Table 4). The 
data set for parasitoids and predators also performed well, with slightly more 
inconsistencies in the ordering of the different sites. Scavengers isolated forested 
from non-forested sites remarkably well, with a concomitant high standard devia- 
tion, but fared less well in classifying sites at a finer scale (Figure 2, Table 4). This 
outcome was similar to that for wood-eaters when rare morphospecies were 
included (Table 4; DCA plot not shown). No bees at all were collected at several 
forested sites and this greatly diminished the discriminatory power of the pollinator 
data set. In general, the discriminatory power of the data sets was affected little by 
the inclusion or exclusion of rare morphospecies (Table 4). 
Indicator value index and predictor sets 
Not surprisingly, the discriminatory power of the data set which included all target 
taxa was the greatest of all data sets examined, with or without rare morphospecies, 
and its ANOSIM was significant (Figure 3, Table 4; Spearman correlation between 
the scores of Axis 1 of the DCA (excluding rare species, 22.5% of total inertia) and 
habitats, r^ = 0.950, P < 0.001). 
Within the two categories of data sets (based on higher and lower taxa, respec- 
tively) the two data sets with the greatest discriminating power were those based on 
'families' and 'all target taxa'. Accordingly, we calculated indicator value indices 
separately for each of these two data sets (Table 5). Again, data sets excluding or 
including rare taxa showed few differences. Most indicator taxa (families or 
morphospecies) had their optima in the garden habitat, whereas the savanna habitat 
was more difficult to characterise. This habitat could be distinguished by the 
occurrence of the family Geometridae (geometrid larvae were sorted but adults were 
not, cf. Methods) and of one morphospecies of Dolichopodidae (predator). Seven 
families were shared by our initial list of target taxa and by indicator lists (compare 
Tables 2 and 5). Many indicator morphospecies belonged to the predator or 
parasitoid guilds, as those guilds included the highest number of morphospecies 
sorted from our samples (Table 5, including rare species). However, no particular 
guild had indicator morphospecies across all habitats studied (Table 5). We note that 
several popular indicator taxa are conspicuously missing from Table 5, either 
because (a) the sampling methods did not target them (i.e., adult butterflies, moths 
and dragonflies); (b) they were not included in the sorting process (Araneae, with the 
exception of Salticidae); or (c) the family-level information that they provided was 
not very informative for the scale of disturbance studied (Formicidae, dung-feeding 
Scarabaeidae). Table 5 nevertheless included other popular indicator taxa, such as 
Collembola, Carabidae, Mycetophilidae and Apoidea (see review of indicator taxa 
in, e.g., Collins and Thomas 1991). 
The newly combined data sets of indicator families and target taxa proved to have 
721 
i . /' E      D 
PARASTTOinS 
POLLmATORS 
SAP-SIJCKERS 
SCAVENGERS 
Figure 2. Plot of study sites within tlie plane formed by the first two axes of DCA for different guilds 
(chewers, parasitoids, pollinators, predators, sap-suckers and scavengers), excluding rare morphospecies. 
ftesentation follows Figure 1. No pollinators were collected in sites A, B, D and F. 
as much discriminating power as those which included all target taxa, and their 
ANOSlMs were also significant (Figure 3, Table 4). Overall, we regard the 'best' 
data set as being the indicator target taxa (Spearman correlation between the scores 
of Axis 1 of the DCA (excluding rare species, 33.6% of total inertia) and habitats, r^ 
= 0.972, P < 0.001). However, the discriminatory power of the data set based on 
indicator families was also surprisingly good (excluding rare families: Axis 1 = 
50.6% of total inertia, correlation between scores and habitats, r^ = 0.777, P < 
0.01). 
722 
ALL TARGET TAXA 
C    D   F     E 
?     o o    o 
H + 
L 
XX K 
INDICATOR FAMILIES 
E| 
9 F 
o H 
D I K                    J 
0 
C 
A ? 
? 
B 
+ L    X 
X 
G 
+ 
J 
X 
^l     B 
1       ? E F 
1                          C 
? 
o 
D 
o 
o L 
X 
H 
+ 
K 
I    X 
+ G 
1 
1    INDICATOR TARGET TAXA 
Figure 3. Plot of study sites within the plane formed by the first two axes of DCA for all target taxa, 
indicator families and indicator target taxa, excluding rare taxa. Presentation follows Figure 1. 
723 
V 
o. 
?a 2 
"w ?? 
>. 
o 
o T} 
PH QJ 
S   is 
1) 
aj 
i^ o 
?a S 
^ 
?n ? 
s " 
s F^, 
< U u c/5 
3 1 'Q-, s p- < O -J 
B ?3 o 
c < O 
"o a ?? 
Sr 
aj "u < oa ? o 
O     O     O     ?I 
o   o   ^ 
w  s  IX ^ ?I o   o   o   ^ 
m Q -J ^ 
?^ 
Qj "^ & 
?g- Q S c/5 
? y >. _rt 
o 
?< O ?J M 
u  s 
o   ^   o   o 
^ ^    m    o    o    o    o 
^"?ZZZZZZZ^ 
?3 s ?3   i=l 
-O   ? 
:3 < u 
I 
o   ?   ^ 
o  o a. 
U   c M i3 
S 1 ?^ QJ 
T3 0 IH ?s 3 T3 a 
1= t? QJ a 
0 3 
1 1 S 4 e 
0 1= 
1 
QJ 
O 
t? 
QJ 
'o 
QJ t? S S 
c3 3 
?a 1 (LT 
1= t? 
1 ? 
QJ 
t? 
? 
1 rt b ^ 
-o o ? ^ S 3 TJ u 'a 1 'DH 
? 
3 
s t? 
1 
o 
s 
O 
H 
1 
">. 
^ 'a. 5? Cu 
QJ S t? Td s ffi t? 'QJ H 1 K) (U bo QJ" ^ rt >, o 
QJ -o J ?a ffi 1 1 ?& Cu rt x S 0 M 'sH 
'o 
-a 
.y ^ 
1 
"o 
Q 
1 
S? cd 
?3 
1 
.a 
a "o .2 'QJ !? 
< Q J t? Td 
1 
'i S? 
-o 
^o ">. 
Td 
'QJ 1 
?3 '5 J Td 
IS rt ?c 
Cu 
< 
-d 
Cu 
1J Td 1 1 Q 'a. Td 
"rt ^ 1 'C & _o H 
< Cu 
o 
QJ" 
4 
?a (LT 1 ?s 
'o 
-o ?>! 
f? 't? O ?n 
< _0 (LT 1 
1 3 
T3 & 
< 
3 
U 'B QJ 
1 e QJ 
3 
o rt 
QJ 
O 
Td 
1 'o ?3 u 1 (U 
QJ 1 QJ 
o 
1 
CL 
J3 
< ^ O .S3 & ^ o 0 \ 
QJ 
_0 s oo S 
"^^ 'S b" bo 
< 0 U 1^ rt Td t? 
aj l? o 'o QJ 1 
0 
e 
o 
U W 1 ? 
724 
3 -I (a) r = 0.29, p = 0.172 22 
?I 
(b) r = 0.45, p = 0.028 
X^ 
350 
Number of taxa 
700 350 700 
Number of taxa 
Figure 4. Relationships between the number of taxa included in data sets and (a) the standard deviation of 
scores on Axis 1 of the DCA; and (b) /3-diversity. 
Although the standard deviations and ??-diversities of all data sets were sig- 
nificantly higher when including rare taxa than when excluding them (Wilcoxon 
signed-rank test, Z = 2.667, P = 0.008 and Z = 2.934, P = 0.003, respectively), 
this did not translate into a notably better classification of the sites along the 
disturbance gradient (comparison between Spearman's coefficients relating scores 
of Axis 1 of the DCA to habitats for data sets including and excluding rare taxa: Z = 
0.921, P = 0.357). 
For the different data sets examined, the relationships between the number of taxa 
included in the analysis and the standard deviation of the DCA scores on Axis 1 or 
the calculated /3-diversity were not obvious and weak, respectively (Figure 4a and 
b; for the latter, the correlation was still weakly significant after removing the 
outlier, r = 0.42, P = 0.047). This suggests that the discriminatory power of the 
data sets was only marginally influenced by the number of taxa included in the data 
sets. 
Discussion 
Sampling protocol and methodological remarks 
Our protocol (as all others) did not enable us to collect all arthropod species present 
at the study sites. The canopy habitat of forested sites was not sampled and other 
sampling methods, such as light trapping, might have yielded a different fauna. 
Some target taxa, in particular, may well have been better sampled. In comparing 
the performance of smaller data sets we used the data matrix which included all 
target taxa as a surrogate for the entire arthropod fauna present at the study sites. 
This approach is necessarily simplistic. However, the taxonomic and functional 
scopes of these preliminary data were much broader than the taxon-based approach 
of most studies addressing the conservation of terrestrial invertebrates in the tropics 
725 
(cf. Introduction) and, as such, offer a stimulating viewpoint for commenting on the 
discriminatory power of different data sets. 
The present project also indicates that training of, and working with local 
parataxonomists (e.g.. Basset et al. 2000) is a promising strategy in the monitoring 
of invertebrates in tropical systems. Lawton et al. (1998) commented on the high 
costs involved with biodiversity surveys in tropical systems, but did not consider the 
advantages of training of local parataxonomists in their protocols. 
All of the traps used in this study have specific advantages and limitations, as 
discussed, for example, in Adis (1979) and Basset et al. (1997). In short, they 
measure the 'density activity' of relatively active arthropods. Sedentary arthropods 
are less likely to be collected and, thus, similarity measurements derived from our 
data are likely to be higher than if sedentary arthropods had been targeted. 
Several important guilds, such as ants, fungal-feeders, other predators and 
tourists, have not been analyzed in the present account. However, we note that, with 
the exception of fungal-feeders (Endomychidae, Corylophidae, Clambidae, etc.), 
few of these guilds are represented by the indicator families identified in Table 5. A 
potentially more severe problem with any guild analysis is the sensitivity of the 
guild assignment (Stork 1987). The taxonomic study of the material collected is 
ongoing, so that changes in the assignment of morphospecies may be expected. 
Factors such as better taxonomy, a wider spectrum of target taxa (including 
sedentary species) and additional collecting methods will contribute to a better 
differentiation among sites. Our measurement of similarity among sites should be 
regarded as conservative. 
Two observations are of particular note. First, many indicator taxa were added as 
a result of their occurrence in the garden habitat. This may reflect the higher catches 
in the gardens (i.e. the information on the garden fauna was higher than for other 
habitats) and also the very distinctive (i.e. weedy or invasive) fauna of gardens. 
Second, the species richness of gardens also appears superficially to be higher than 
in other habitats. This in turn may be explained by (a) the higher catches in gardens 
(i.e. rarefaction techniques will be needed to compare habitats); (b) the canopy 
habitat of forested sites was not surveyed and it may well account for a significant 
part of diversity, since faunal turnover between the understorey and the canopy is 
usually considerable (e.g.. Basset et al. 2001b); and (c) insects may be more seasonal 
in forests than in gardens and hence temporal turnover may account for a substantial 
amount of diversity in forested habitats. 
Taxonomic resolution of the data sets 
There has been much debate, especially related to aquatic systems, as to what 
taxonomic level (either family- or species-level) is most suited for biological 
monitoring (e.g., Gu?rold 2000; Lenat and Resh 2001). The consensus is that, 
whenever possible, sorting to species is better. However, in some conditions, sorting 
to families may be acceptable (Bailey et al. 2001; Lenat and Resh 2001). Not 
surprisingly, analyses using higher taxa appear to be better suited to studies at 
broader geographic scales (Hewlett 2000). The present report suggests that for 
726 
studies of terrestrial arthropods distributed along a disturbance gradient confined to 
a small geographical area, such as the Gamba oil concession, data sets based on 
orders or guilds achieve only a poor discrimination of sites. Ordinal signatures have 
been found useful for terrestrial invertebrates only when encompassing broad 
geographic areas, such as latitudinal transects (e.g., Kitching et al. 1993). 
Our family data sets, either including or excluding rare families, performed better 
than orders or guilds, but could not, for example, distinguish between sites situated 
in old and young secondary forests (Figure 1). However, filtering and retaining only 
indicator families resulted in a better discrimination of sites (Figure 3). This result is 
reassuring and suggests that counting the individuals within a set of a few arthropod 
families (n = 26 in our case) may be a possible strategy for biological monitoring of 
terrestrial systems in the tropics (cf. Kitching et al. 2000). This task can be 
performed easily by local parataxonomists trained beforehand, as in this study. The 
choice of indicator families may be guided by Table 5, for study systems similar to 
the present one. However, one must bear in mind that the indicator value of the 
families listed in Table 5 depended on collecting methods and on the regional 
species pool available. Further studies of similar scope evaluating other ecosystems 
in the tropics may eventually allow a consensus 'predictor set' to be reached. 
Nevertheless, the data sets based on morphospecies performed better than those 
based on any higher taxa. Our two most discriminating data sets were those which 
included all morphospecies sorted from target taxa, and from the concomitant 
restricted set of indicator target taxa. Sorting only the indicator families identified 
above to morphospecies represents an additional strategy for biological monitoring. 
However, the current taxonomy of some families listed in Table 5 is difficult or 
requires particular techniques (e.g., Aleyrodidae, Entomobryidae, Sciaridae, etc.). It 
is doubtful whether a good correspondence could ever be achieved between a set of 
indicator families and the availability of taxonomic expertise. Choice will ultimately 
always be influenced by taxonomic expertise. However, as the present analyses 
show, considering morphospecies belonging to different guilds (i.e. analyses with 
all target taxa) was a better strategy than considering morphospecies belonging to a 
particular guild, especially since no guild yielded indicator taxa for all habitats 
studied (Table 5). We suggest, therefore, that the choice of target taxa should be 
guided by (1) available taxonomic expertise, (2) the inclusion of representatives 
from different guilds, and (3) considering as priorities taxa belonging to the 
following guilds (Tables 4 and 5): chewers, parasitoids, predators and sap-suckers. 
We also note that including more taxa in the data sets does not ensure that these data 
sets will have greater discriminatory power (Figure 4). The quality and choice of the 
information included in the data sets are more important in this regard. 
The importance of rare species in the analyses 
This debate may be considered from either a statistical or biological viewpoint. We 
focus on the latter and distinguish between 'true' rarity (species with genuinely low 
population levels, occurring at the limit of their geographical distribution, etc.) and 
'apparent' rarity (sampling methods not appropriate, sampling of marginal habitats, 
seasonality, or timing of activity, etc.). Rare species are perceived as important in 
727 
aquatic systems and are usually retained for the analyses, since their preservation is 
often the ultimate aim of biological monitoring (Lenat and Resh 2001). The 
situation is not as evident in terrestrial systems, where the higher and more complex 
habitat diversity means that a greater proportion of 'apparent' rare species is also to 
be expected (i.e. vagrant or poorly sampled species are probably much more 
common in terrestrial than in aquatic systems). Work with local parataxonomists, by 
increasing the scope of the protocol to include other sampling methods and 
obtaining more spatial and temporal replicates, may reduce the proportion of 
'apparent' rare species, but not beyond a certain threshold, which relates to the 
sampling of marginal habitats (Novotny and Basset 2000). Since, in the tropics, the 
proportion of 'apparent' rare species may be relatively high in terrestrial data sets 
obtained with low sampling effort, common sense dictates the exclusion of all rare 
species from these data sets. In contrast, data sets obtained with intensive sampling 
effort should be analyzed with their rare species, since these data are more likely to 
include 'true' rare species, as defined above. 
The present results suggest that inclusion or exclusion of rare species per se in the 
analyses is not as important as the primary information included in the data matrices 
(i.e. which taxa are included). Analyses including rare taxa had slightly more 
discriminating power than those excluding rare species, but the fine discrimination 
of study sites was not improved notably. This conclusion was confirmed when we 
repeated the analyses (not presented here) of DCAs including rare taxa but with the 
raw data log(x + 1 ) transformed, to downscale the effect of abundant species. This 
observation is reassuring and confirms the prospect of reaching a consensus and 
establishing a 'recipe' for biological monitoring of species-rich terrestrial eco- 
systems in the tropics. 
Predictor sets and indicator taxa 
No single taxon was useful in accurately classifying our study sites along the 
disturbance gradient, but a restricted set of families and morphospecies was. 
Accordingly, this study validates the concept of predictor sets for biological 
monitoring of terrestrial tropical systems and provides a starting point for identify- 
ing taxa that may be included in such predictor sets (see also Kotze and Samways 
(1999), for a similar argument). It is germane to ask what would have been the 
results of our study if we had considered only a popular indicator taxon such as, say, 
dung beetles (e.g., Davis et al. 2001; McGeogh et al. 2002). Certainly re-directing 
our whole protocol and sampling effort to increase spatial replicates and using 
different traps (including baits) would have resulted in better representation of dung 
beetles within our data sets, and those may have been able to discriminate the study 
sites more accurately. That is not our argument. Our protocol allowed comparison of 
the relative discriminatory power of similar data sets based either on chewers 
(including 536 individuals and 85 morphospecies of Chrysomelidae; Table 2), 
parasitoids (including 200 individuals and 22 morphospecies of Chalcididae) or 
scavengers (including 449 individuals and 28 morphospecies of dung beetles; Table 
2) obtained with a similar sampling effort. The data sets for chewers and parasitoids 
proved to be more discriminating than the scavenger data set, which could differen- 
728 
tiate only between forested and non-forested sites. These results were similar when 
restricting the data set to dung beetles (analyses not presented here). We do not 
suggest that dung beetles should be excluded from biological monitoring, rather that 
they should be analyzed conjointly with other taxa, as predictor sets. Dung beetles 
are undoubtedly useful indicators of structural differences between ecosystems, in 
contrast to insect herbivores, such as chrysomelids, that reflect plant-feeding 
specialisations (Davis et al. 2001). 
In choosing taxa for inclusion in predictor sets, one must consider that the 
occurrence of some taxa may be related, and hence redundant. For example, 
Coccinellidae, Dolichopodidae and Syrphidae are likely to be predators of 
Aphididae in gardens (Table 5). Accordingly it will be more informative to consider 
but one of these families, redirecting effort to other families that may be of greater 
indicator value for old forests (such as Mycetophilidae; Table 5, and see the study of 
0kland (1994) in Norway). Although Table 5 provides baseline information to 
identify predictor sets for forest-savanna ecosystems in Africa, we remain reluctant 
to propose specific families and morphospecies for predictor sets before studies 
similar to ours are carried out elsewhere in the tropics to build up a minimum level 
of information. 
Conclusions 
This study validated the usefulness of arthropod predictor sets in biological 
monitoring at a taxonomic scope not investigated before in the tropics. Kitching et 
al. (2000) identified 17 subfamilies of moths from light-trapping data that may be 
used as predictor sets of the quality of rainforest remnants in Australia. Work should 
now proceed to gather data similar to those presented here but originating from 
different tropical systems, and seeking a consensus as to which taxa should be 
included in predictor sets of wider taxonomic scope. Such a consensus will allow the 
derivation of a biotic index that will be representative, efficient and applicable to 
terrestrial invertebrate assemblages in the tropics, unlike that based on a few 
putative indicator taxa, such as butternies or dung beetles. Reaching such a 
consensus will not be easy. Proposed indicators are legion. They can indicate 
different effects. Their validation requires studies at different disturbance and 
geographical scales, and they should also be representative of major important 
microhabitats (McGeogh 1998; Andersen 1999; Taylor and Doran 2001). However, 
the calibration studies (with a statistically based, relatively objective, analysis of the 
kind presented here) required to achieve this ambitious task could be reasonably 
quickly performed (in less than half a year, as in the present study) with the help of 
local parataxonomists, with adequate eutaxonomic support. 
Acknowledgements 
Francisco Dallmeier, James Comiskey and Michelle Lee helped to implement the 
729 
project. Parataxonomists Serge Mboumba Ditona, Gauthier Moussavou, Patricia 
Ngoma, Judicael Syssou, Landry Tchignoumba and Elie Tobi collected, processed, 
sorted and data-based most of the insect material with great competence. Jeffrey 
Raymakers and Serge Mboumba Ditona provided the vegetational data. Jan Leps 
commented helpfully on statistical analyses. The manuscript benefited from com- 
ments by Vojtech Novotny and Neil D. Springate. The project was funded by the 
Smithsonian Institution, National Zoological Park, Conservation and Research 
Center/MAB Program through a grant from the Shell Foundation and Shell Gabon. 
References 
Adis J. 1979. Problems of interpreting arthropod sampling with pitfall traps. Zoologischer Anzeiger, Jena 
202: 177-184. 
Andersen A.N. 1999. My bioindicator or yours? Making the selection. Journal of Insect Conservation 3: 
61-64. 
Anstett M.-C, Michaloud G. and Kjellberg F. 1995. Critical population size for fig/wasp mutualism in a 
seasonal environment: effect and evolution of the duration of female receptivity. Oecologia 103: 
453-461. 
Bailey R.C., Norris R.H. and Reynoldson T.B. 2001. Taxonomic resolution of benthic macroinvertebrate 
communities in bioassessments. Journal of the North American Benthological Society 20: 280-286. 
Bartolozzi L. and Sforzi A.   1997. Contribution to the knowledge of the Brentidae from Gabon 
(Cole?ptera Brentidae). BoUettino della Societa Entomol?gica Italiana 29: 79-86. 
Basset Y., Charles E.L., Hammond D.S. and Brown V.K. 2001a. Short-term effects of canopy openness on 
insect herbivores in a rain forest in Guyana. Journal of Applied Ecology 38: 1045-1058. 
Basset Y., Aberlenc H.-R, Barrios H., Curletti G., B?ranger J.-M.,Vesco J.-P. et al. 2001b. Stratification 
and diel activity of arthropods in a lowland rainforest in Gabon. Biological Journal of the Linnean 
Society 72: 585-607. 
Basset Y, Springate N.D., Aberlenc H.-P. and Delvare G. 1997. A review of methods for collecting 
arthropods in tree canopies. In: Stork N.E., Adis J.A. and Didham R.K. (eds), Canopy Arthropods. 
Chapman & Hall, London, pp. 27-52. 
Basset Y, Novotny V, Miller S.E. and Springate N.D. 1998. Assessing the impact of forest disturbance on 
tropical invertebrates: some comments. Journal of Applied Ecology 35: 461-466. 
Basset Y,  Novotny V, Miller S.E.  and Pyle R.  2000.  Quantifying biodiversity:  experience with 
parataxonomists and digital photography in New Guinea and Guyana. BioScience 50: 899-908. 
Belshaw R. and Bolton B. 1993. The effect of forest disturbance on the leaf litter ant fauna in Ghana. 
Biodiversity and Conservation 2: 656-666. 
Bilardo A. and Rocchi S. 2000. Haliplidae e Dytiscidae (Cole?ptera) del Gabon (Parte terza). Atti della 
Societa Italiana di Scienze Naturali e del Museo Civico di Storia Naturale di Milano 140: 215-236. 
Bourgeais J. 2001. Impact de l'exploitation p?troli?re dans le complexe de Gamba: le pire est ? venir. 
Canop?e 21: 5-7. 
Boussienguet J., Neuenschwander P. and Herren H.R.   1991. Essais de lutte biologique contre la 
cochenille du manioc au Gabon: I. - Etablissement, dispersion du parasite exotique Epidinocarsis 
lopezi [Hym.: Encyrtidae] et d?placement comp?titif des parasites indig?nes. Entomophaga 36: 
455-469. 
Chung A.Y.C., Eggleton P., Speight M.R., Hammond RM. and Chey V.K. 2000. The diversity of beetle 
assemblages in different habitat types in Sabah, Malaysia. Bulletin of Entomological Research 90: 
475-496. 
Clarke K.R. 1993. Non-parametric multivariate analyses of changes in community structure. Australian 
Journal of Ecology 18: 117-143. 
Collins N.M. and Thomas J.A. 1991. The Conservation of Insects and Their Habitats. Academic Press, 
New York. 
730 
Colwell R.K.   1997.  Biota:  The Biodiversity  Database Manager.  Sinauer Associates,  Sunderland, 
Massacliusetts. 
Cranston P.S. 1990. Biomonitoring and invertebrate taxonomy. Environmental Monitoring and Assess- 
ment 14: 265-273. 
Davis A.J., HoUoway J.D., Huijbregts H., Kritten J., Kirk-Spriggs A.H. and Sutton S.L. 2001. Dung 
beetles as indicators of change in the forests of northern Borneo. Journal of Applied Ecology 38: 
593-616. 
Didham R.K., Ghazoul J., Stork N.E. and Davis A.J. 1996. Insects in fragmented forests: a functional 
approach. Trends in Ecology and Evolution 11: 255-260. 
Didham R.K., Hammond RM., Lawton J.H., Eggleton R and Stork N.E. 1998. Beetle species richness 
responses to tropical forest fragmentation. Ecological Monographs 68: 295-323. 
Doumenge C, Garcia Yuste J.-E., Gartlan S., Langrand O. and Ndinga A. 2001. Conservation de la 
biodiversit? foresti?re en Afrique Centrale Atlantique: le r?seau d'aires prot?g?es est-il ad?quat? Bois 
et For?ts des Tropiques 268: 5-27. 
Dufr?ne M. and Legendre P. 1997. Species assemblages and indicator species: the need for a flexible 
asymmetrical approach. Ecological Monographs 67: 345-366. 
Eggleton R, Bignell D.E., Sands W.A., Mawdsley N.A., Lawton J.H., Wood T.G. et al. 1996. The 
diversity, abundance and biomass of termites (Isoptera) under differing levels of forest disturbance in 
the Mbalmayo Forest Reserve, southern Cameroon. Philosophical Transactions of the Royal Society 
of London, Series B 351: 51-68. 
Gaston K.J. and Hudson E. 1994. Regional patterns of diversity and estimates of global insect species 
richness. Biodiversity and Conservation 3: 493-500. 
Grandcolas P. 1997. Habitat use and population structure of a polyphagine cockroach, Ergaula capensis 
(Saussure 1893) (Blattaria Polyphaginae) in Gabonese rainforest. Tropical Zoology 10: 215-222. 
Growns J.E., Chessman B.C., Jackson J.E. and Ross D.G. 1997. Rapid assessment of Australian rivers 
using macroinvertebrates: cost and ef?ciency of 6 methods of sample processing. Journal of the North 
American Benthological Society 16: 682-693. 
Gu?rold F. 2000. Influence of taxonomic determination level on several community indices. Water 
Research 34: 487-492. 
Hammond P.M. 1994. Practical approaches to the estimation of the extent of biodiversity in speciose 
groups. Philosophical Transactions of the Royal Society of London, Series B 345: 119-136. 
Harrison S., Ross S.J. and Lawton J.H. 1992. Beta diversity on geographic gradients in Britain. Journal of 
Animal Ecology 61: 151-158. 
Hewlett R. 2000. Implications of taxonomic resolution and sample habitat for stream classification at a 
broad geographic scale. Journal of the North American Benthological Society 19: 352-361. 
Hill J.K. 1999. Butterfly spatial distribution and habitat requirements in a tropical forest: impacts of 
selective logging. Journal of Applied Ecology 36: 364-372. 
Intachat J.,  HoUoway J.D.  and Speight M.R.   1999.  The impact of logging on geometroid moth 
populations and their diversity in lowland forests of Peninsular Malaysia. Journal of Tropical Forest 
Science 11: 61-78. 
JanzenD.H. andGauldl.D. 1997. Patterns of use of large moth caterpillars (Lepidoptera: Saturniidae and 
Sphingidae) by ichneumonid parasitoids (Hymenoptera) in Costa Rican dry forest. In: Watt A.D., 
Stork N.E. and Hunter M.D. (eds). Forests and Insects: 18th Symposium of the Royal Entomological 
Society. Chapman & Hall, London, pp. 251-271. 
Karr J.R. 1991. Biological integrity: a long-neglected aspect of water resource management. Ecological 
Applications 1: 66-84. 
Kitching R.L.   1993a. Biodiversity and taxonomy: impediment or opportunity? In: Moritz C. and 
Kikkawa J. (eds). Conservation Biology in Australia and Oceania. Surrey Beatty and Sons, Chipping 
Norton, Australia, pp. 253-268. 
Kitching R.L. 1993b. Towards rapid biodiversity assessment - Lessons following studies of arthropods 
of rainforest canopies Rapid Biodiversity Assessment.: Proceedings of the Biodiversity Workshop, 
Macquarie University - 1993. Research Unit for Biodiversity and Bioresources, Macquarie Universi- 
ty, Sydney, Australia, pp. 26-30. 
Kitching R.L., Bergelson J.M., Lowman M.D., Mclntyre S. and Carruthers G. 1993. The biodiversity of 
731 
arthropods from Australian rainforest canopies: general introduction, methods, sites and ordinal 
results. Australian Journal of Ecology 18: 181-191. 
Kitching R.L.,  Orr A.G.,  Thalib L., Mitchell  H.,  Hopkins  M.S.  and Graham A.W.  2000.  Moth 
assemblages as indicators of environmental quality in remnants of upland Australian rain forest. 
Journal of Applied Ecology 37: 284-297. 
Kitching R.L., Li D.Q. and Stork N.E. 2001. Assessing biodiversity 'sampling packages': how similar are 
arthropod assemblages in different tropical rainforests? Biodiversity and Conservation 10: 793-813. 
Kotze D.J. and Samways M.J. 1999. Support for the multi-taxa approach in biodiversity assessment, as 
shown by epigaeic invertebrates in an Afromontane forest archipelago. Journal of Insect Conservation 
3: 125-143. 
Kremen C. 1992. Assembling the indicator properties of species assemblages for natural areas moni- 
toring. Ecological Applications 2: 203-217. 
Kremen C, Colwell R.K., Erwin T.L., Murphy D.D., Noss R.F. and Sanjayan M.A. 1993. Terrestrial 
arthropod assemblages: their use in conservation planning. Conservation Biology 7: 796-808. 
Landres RB.,Vemer J. and Thomas J.W. 1988. Ecological use of vertebrate indicator species: a critique. 
Conservation Biology 2: 316-328. 
Lawton J.H., Bignell D.E.,  Bolton B., Bloemers G.F., Eggleton R,  Hammond RM.  et al.   1998. 
Biodiversity inventories, indicator taxa and effects of habitat modification in tropical forest. Nature 
391: 72-76. 
Lenat D.R. and Resh V.H. 2001. Taxonomy and stream ecology - The benefits of genus- and species- 
level identification. Journal of the North American Benthological Society 20: 287-298. 
Maes J.-M. and Pauly A. 1998. Lucanidae (Cole?ptera) du Gabon. Bulletin et Annales de la Soci?t? 
Royale Belge d'Entomologie 134: 279-285. 
McAleece N., Lambshead J., Paterson G. and Gage J. 1997. BioDiversity Professional Beta Release 1. 
The Natural History Museum and the Scottish Association for Marine Science, at http://www. 
sams.ac.uk/dml/projects/benthic/bdpro/index.htm. 
McCune B. and Medford M.J.  1999. Multivariate Analysis of Ecological Data Version 4.10. MjM 
Software, Gleneden Beach, Oregon. 
McGeogh M.A.  1998. The selection, testing and application of terrestrial insects as bioindicators. 
Biological Reviews 73: 181-201. 
McGeogh M.A.jVan Rensburg B.J. and Botes A. 2002. The verification and application of bioindicators: 
a case study of dung beetles in a savanna ecosystem. Journal of Applied Ecology 39: 661-672. 
Miller S.E. and Rogo L. 2002. Challenges and opportunities in understanding and utilisation of African 
insect diversity. Cimbebesia 17: 197-218. 
Moran C.V. and Southwood T.R.E. 1982. The guild composition of arthropod communities in trees. 
Journal of Animal Ecology 51: 289-306. 
Moritz C, Richardson K.S., Ferrier S., Monteith G.B., Stanisic J., Williams S.E. et al. 2001. Biogeog- 
raphical concordance and efficiency of taxon indicators for establishing conservation priority in a 
tropical rainforest biota. Proceedings of the Royal Society of London, Series B 268: 1875-1881. 
Novotny V. and Basset Y. 2000. Rare species in communities of tropical insect herbivores: pondering the 
mystery of singletons. Oikos 89: 564-572. 
Novotny v., Basset Y., Miller S.E., Weiblen G.D., Bremer B., Cizek L. et al. 2002. Low host specificity of 
herbivorous insects in a tropical forest. Nature 416: 841-844. 
O'Connell T.J., Jackson L.E. and Brooks R.P. 1998. A bird community index of biotic integrity for the 
mid-Atlantic highlands. Environmental Monitoring and Assessment 51: 145-156. 
0kland B. 1994. A comparison of clearcut, managed and semi-natural spruce forests in southern Norway. 
Biodiversity and Conservation 3: 68-85. 
Oswald J.D. 1998. Annotated catalogue of the Dilaridae (Insecta: Neroptera) of the world. Tijdschrift 
voor Entomologie 141: 115-128. 
Pauly A. 1998. Hymenoptera Apoidea du Gabon. Mus?e Royal de l'Afrique Centrale Tervuren, Belgique 
Annales Sciences Zoologiques 282: 1-121. 
?^endergast J.R., Quinn R.M., Lawton J.H., Eversham B.C. and Gibbons D.W. 1993. Rare species, the 
coincidence of diversity hotspots and conservation strategies. Nature 365: 335-337. 
732 
Mns H.H.T. and Reitsma J.M. 1989. Mammalian biomass in an African equatorial forest. Journal of 
Animal Ecology 58: 851-861. 
Rossaro B. and Pietrangelo A. 1993. Macroinvertebrate distribution in streams: a comparison of CA 
ordination with biotic indices. Hydrobiologia 263: 109-118. 
Roubik D.W. 1999. The foraging and potential outcrossing pollination ranges of African honey bees 
(Apiformes: Apidae; Apini) in Congo Forest. Journal of the Kansas Entomological Society 72: 
394-401. 
Roy R. 1973. Premier inventaire des mantes du Gabon. Biolog?a Gabonica 8: 235-290. 
Stork N.E. 1987. Arthropod faunal similarity of Bornean rain forest trees. Ecological Entomology 12: 
219-226. 
Stork N.E. 1991. The composition of the arthropod fauna of Bornean lowland rain forest trees. Journal of 
Tropical Ecology 7: 161-180. 
Taylor R. J. and Doran N. 2001. Use of terrestrial invertebrates as indicators of the ecological sustainabili- 
ty of forest management under the Montreal Process. Journal of Insect Conservation 5: 221-231. 
Ter Braak C.J.F. and Smilauer P 1998. CANOCO Reference Manual and User's Guide to CANOCO for 
Windows: Software for Canonical Community Ordination (version 4). Microcomputer Power, Ithaca, 
New York. 
Thibault M. and Blaney S. 2001. Sustainable human resources in a protected area in Southwestern 
Gabon. Conservation Biology 15: 591-595. 
Thome R.S.J. and Williams WP  1997. The response of benthic macroinvertebrates to pollution in 
developing countries: a multimetric system of bioassessment. Freshwater Biology 37: 671-686. 
Townes H. 1972. A light-weight Malaise trap. Entomological News 83: 239-247. 
Urzelai A., Hernandez A.J. and Pastor J. 2000. Biotic indices based on soil nematode communities for 
assessing soil quality in terrestrial ecosystems. Science of the Total Environment 247: 253-261. 
Vasconcelos H.L. 1999. Effects of forest disturbance on the structure of ground-foraging ant communities 
in central Amazonia. Biodiversity and Conservation 8: 409-420. 
Walter P.  1987. Contribution ? la connaissance des Scarabeides coprophages du Gabon [Col.]. 5. 
Nidification et morphologie larvaire de Tiniocellus panthera (Boucomont). Annales de la Soci?t? 
Entomologique de France 23: 309-314. 
Warwick R.M.  1993. Environmental impact studies on marine communities. Australian Journal of 
Ecology 18: 63-80. 
Watt A.D. 1998. Measuring disturbance in tropical forests: a critique of the use of species-abundance 
models and indicator measures in general. Journal of Applied Ecology 35: 467-469. 
Wetterer J.K., Walsh P.D. and White L.J.T.  1999. Wasmannia auropunctata (Roger) (Hymenoptera: 
Formicidae), a destructive tramp-ant, in wildlife refuges of Gabon. African Entomology 7: 292-294. 
Whittaker R.M.   1960. Vegetation of the  Siskiyou Mountains,  Oregon and California.  Ecological 
Monographs 30: 279-338.