The world's oceans have been routinely exploited for human consumption for at least 1000 years with widely documented impacts (Brander 1981; Tegner 1996; Jackson et al. 2001; Christensen et al. 2003; Dulvy et al. 2003; Sadovy & Cheung 2003; Rosenberg et al. 2005; Devine et al. 2006). In addition to harvesting marine species as a source of bulk protein, many fisheries supply demand-driven commodity markets (Hughes et al. 2003). These luxury wildlife trades are of particular conservation concern due to high prices which, coupled with strong demand, ensure incentives for persistent heavy exploitation. However, lack of long-term data limits monitoring and assessment of the impacts of these trades and hence the scope for effective management and conservation.

The live reef fish trade (LRFT) involves the exploitation of coral reef fishes from across the Indo-Pacific to supply luxury seafood restaurants (Johannes & Riepen 1995; Sadovy et al. 2003; Scales et al. 2006). The unsustainability of the LRFT is demonstrated by repeated boom-and-bust patterns of trade at a country level and a wave of exploitation spreading away from Hong Kong, the main centre of the trade (Scales et al. 2006). No studies have quantitatively assessed localized impacts of the trade on target species—mainly large-bodied groupers (Serranidae), wrasses (Labridae) and snappers (Lutjanidae) (Lee & Sadovy 1998). These species are inherently vulnerable to overexploitation due to a suite of shared life-history characteristics (e.g. slow growth rate, late age at maturity; Jennings et al. 1999; Morris et al. 2000; Reynolds et al. 2001; Sadovy et al. 2004; Sadovy 2005b). The rate of extraction of groupers for the LRFT is thought to exceed potential sustainable yield (Sadovy et al. 2003; Warren-Rhodes et al. 2003), while ecosystem-level impacts of removing these species for the LRFT are largely unknown and likely to be complex (Jackson 2001).

We present the most detailed quantitative analysis to date of LRFT impacts using the data from live reef fish traders operating in northern Borneo where the trade has operated since 1980 and is anecdotally reported as being in decline (Bentley 1999). While there are various subsistence fisheries exploiting multiple species of coral reef fish in northern Borneo, inspections of local markets in Kudat revealed an absence of the particular species favoured by the LRFT (H. Scales & A. Manica 2001–2004, personal observations). Informal interviews with fishers also indicated that any captured specimens of desired species that can be kept alive are sold to the LRFT (H. Scales, unpublished data). Therefore, we believe that the declines observed in the daily catch data reflect the impact of the LRFT, since the species involved are rarely caught by other fisheries. Continuous daily fish catch data provided species-specific information on the mass of all the fishes sold to each trader over intervals of between two and eight years. We quantitatively assessed population-level impacts of the LRFT using changes in summed monthly total catch and catch-per-unit-effort (which indicates relative abundance of exploited populations). The longest dataset revealed dramatic species-specific declines in both total catch and relative abundance, which we believe to reflect the impact of the LRFT on populations of reef fish.

Figure 1 Maps of (a) Southeast Asia and (b) northern Borneo region showing location of three traders visited during the study.

The longest dataset was from one Kudat trader (Kudat 1), with data from January 1995 until January 2003 (excluding January 1998–July 1999, due to a missing record book; n=2442 daily records, because on several days no fishes were delivered). A second Kudat trader (Kudat 2) provided data from November 1999 to June 2003 (n=763 daily records; on several days no fishes were delivered). The Malawali trader provided data from August 2001 to August 2003 (n=3815 records; multiple records were made on most days).

We aggregated the daily data into monthly figures and used linear mixed models to analyse temporal changes in total catch and relative abundance for each species (both response variables were log transformed to linearize relationships), taking year as a fixed effect and month as a random effect, to account for seasonal trends. Year was set as a continuous variable, allowing the significance of linear trends through time to be tested. A fully saturated model (i.e. a model with all factors) was initially fitted and a minimal model (i.e. a model that included only factors that had a significant explanatory power) was subsequently obtained through backwards-stepwise elimination (Venables & Ripley 2002). Significance at α=0.05 was adopted as the criterion for terms to be kept in the minimal model. This approach was preferred over more sophisticated time-series analysis (e.g. Box–Jenkins models) due to its robustness when faced with missing data and short time-series.

Our power to detect declines was determined by dataset length for which our three time-series differed markedly. To investigate the ability to detect trends in the shorter datasets, we repeated analyses on truncated versions of the longest dataset (Kudat 1).

Total monthly catches in the longest dataset (Kudat 1) declined significantly between 1995 and 2003 for all targeted species (changes range from −73 to −99%; table 1 and figure 2). The monthly number of boats selling fishes to the Kudat 1 trader declined significantly between 1995 and 2003 (R²=0.58, n=76, p<0.001). However, this reduction in total fishing effort did not always explain the declines in total monthly catches; relative abundance of three species (Napoleon wrasse −98%, Epinephelus −89% and bluelined groupers −99%) also declined significantly between 1995 and 2003 (table 1 and figure 2).

Table 1 Results of linear mixed models testing the temporal changes in total monthly catches (kg per month) and relative abundance (kg per fishing boat or per fisher per trip)±s.e. of target species in the LRFT from Kudat 1, Kudat 2 and Malawali datasets (more ...)

Figure 2 Mean (±1s.e.) total annual catches (kg) and relative abundance (kg per fisher or boat per trip) for Napoleon wrasse, leopard grouper, Epinephelus species, spotted grouper, saddleback grouper, squaretail grouper and bluelined grouper from (more ...)

Results in shorter datasets were equivocal. Total monthly catches in the Kudat 2 dataset declined significantly between 1999 and 2003 for all species except spotted grouper (Plectropomus maculatus; table 1 and figure 2). The total monthly number of boats in the Kudat 2 dataset also declined between 1999 and 2003 (R²=0.30, n=35, p=0.001). In contrast to Kudat 1, this reduction in fishing effort explained all the declines in total monthly catches in the Kudat 2 dataset, since none of the relative abundance trends declined significantly (table 1 and figure 2).

The shortest dataset (Malawali) revealed very few clear patterns. The only significant temporal change was an increase in the total monthly catch of Napoleon wrasse (table 1 and Fig. A1 in the electronic supplementary material).

Truncating the Kudat 1 dataset confirmed that lack of clear patterns in other time-series was probably due to their short coverage. Restricting the analysis to 1999–2003 data only from Kudat 1 produced a similar pattern to that in Kudat 2; while the total monthly catches of four species declined significantly, all but one of the relative abundance trends were non-significant (the trend for bluelined grouper became positive; table 1). Likewise, taking 2001–2003 data from Kudat 1 produced inconsistent results; total catch and relative abundance declined in one species (and total catch in a second species), but apparently increased in two others (table 1). All other trends were non-significant.

Finally, to test the importance of species-specific data for identifying temporal trends, we lumped species together and examined overall trends in catch and relative abundance. For Kudat 1 and Kudat 2, total catches declined significantly over time, but in neither case was there a significant trend in relative abundance over time (table 1 and figure 3). The Malawali datasets showed no significant trends over time (table 1, figure 3 and fig. A1 in the electronic supplementary material).

Figure 3 Mean (±1s.e.) total annual catches (kg) and relative abundance (kg per fisher or boat per trip) for all species lumped from three fish traders in northern Borneo: Kudat 1 (circles), Kudat 2 (triangles) and Malawali (squares). Filled symbols (more ...)

We found significant exponential declines of several coral reef fish species (in total catch and relative abundance) targeted by the LRFT in northern Borneo, which we interpret as the first quantitative evidence indicating severe population impacts of the trade. These declines took place in under a decade and are especially worrying given the mobile nature of the Kudat fishing fleet, since it is probable that vessels shifted range when nearby populations became depleted, thus maintaining catch rates for longer. The masking of serial depletions by spatial shifts in exploitation range is a major obstacle in assessing impacts of mobile fishing fleets on global marine resources (Berkes et al. 2006). That relative abundance in our datasets has nevertheless declined implies that population-level reductions were even more dramatic and sustained over a wide area. It is improbable that these trends were caused by other factors such as major changes in reef health; Borneo was relatively unaffected by the 1998 global coral bleaching event (Wilkinson 2000) and international trade statistics show that stocks of different LRFT species did not simultaneously collapse but were serially depleted according to market price (Scales et al. 2006).

It was necessary to assume that the daily effort per fisher or boat and allocation of effort between targeted species had not changed over time (this was supported by interviews with 112 fishers and 13 traders in the region; H. Scales, unpublished data). Any unaccounted changes in effort were most probably improvements rather than declines in gear efficiency and available technologies; thus if anything we have underestimated effort and hence overestimated relative abundance in later years.

We emphasize the crucial importance of obtaining datasets that are (i) long term, and (ii) species-specific, in order to maximize chances of detecting temporal trends, including dramatic collapses in multi-species trades. A vital application of such datasets is to monitor naturally rare species, e.g. Napoleon wrasse, that are challenging to directly study in the wild. This is especially important for species listed on the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES), which prohibits their trade without such population information (Sadovy 2005a). Our data, together with global boom-and-bust trends in the trade (Scales et al. 2006), are a stark warning that similar collapses may be expected throughout the LRFT and in other high-value wildlife trades, but without long-term datasets these trends may go undetected.

We thank S. Oakley and all the staff at TRACC, A. Petherick, J. Wee and H. Ung for their help with data collection and logistics in Borneo. We also thank I. Scales, T. Fayle and J. Scharlemann for comments on earlier drafts.