Flooding is a key driver of the Tonle Sap dai fishery in Cambodia

As one of the richest sources of fisheries-related data in the lower Mekong basin, the Tonle Sap dai fishery has received considerable attention in the literature in recent years as concerns grow over the impacts of hydropower dams on fisheries, which are important for livelihoods and food security [1][2][3] .Ngor et al. 4 reported a decline since 2000 in the catch of larger species which tend to occupy higher trophic levels; compensatory increases in the catch of smaller species; and declines in the mean body weight (and length) of common species in the Tonle Sap dai fishery, as evidence of the effects of indiscriminate fishing or "fishingdown" of the multi-species fish assemblage in the lower Mekong basin.We provide evidence below that suggest that these apparent recent changes are more likely to reflect changing hydrological conditions than fishing-down effects, possibly caused by climate change and recently also by hydropower development.The dai fishery has been reliably monitored since 1997-98.Without explanation, Ngor et al. excluded the first three seasons (1997-98 to 1999-2000) of monitoring data which include one of the driest fishing seasons on record (1998-99).The authors thereby created a time series beginning with the three wettest seasons (largest floods) since monitoring began (2000-1 to 2002-3) that were followed by 12 seasons of variable, but decreasing flows caused by hydropower dam construction, low rainfalls possibly resulting from climate change, and abstractions for agriculture 5,6 (Fig. 1).Without reference to hydrological variation, Ngor et al. reported a temporal decline in the catch of larger, high-trophic level species; compensatory increases in the catch of smaller species; and declines in the mean body weight (and length) of six common species.The authors reported these findings as evidence of the effects of indiscriminate fishing (fishing-down) in the lower Mekong basin, without considering alternative explanations for the observed patterns.Ngor et al. describe their dataset as being generated from a "standardized biological catch assessment".In reality, this assessment is complex having undergone numerous changes through time, including to sampling effort 2 .Moreover, the fishery itself is not standardised, with net types, mesh sizes and flow rates through the gears varying inter and intra-annually.Catches may therefore not reflect changes in underlying abundance 7 .Despite this, we re-examined the analysis of Ngor et al. at face value.Using the 15-season dataset that accompanied Ngor et al., we re-fitted regression models to the log e -transformed catch time series of larger species (TL > 45 cm) excluding those with zero catch in any year.These 28 species formed approximately 16% of the total catch during the study period.We also found negative regression coefficients for all 28 species, supporting the findings of Ngor et al.However, the combined annual catch of these 28 species did not decline significantly through time (R 2 = 0.22; p-value = 0.07).We did however find that the combined annual catch of these 28 larger species varied significantly with the annual flood index (FI)-a measure of flood extent and duration (R 2 = 0.46; p-value < 0.01) (Fig. 2).The FI is a key driver of fish growth and fisheries productivity in floodplain-river systems 8 .Annual catches for each of the 28 species were also found to vary (correlate positively) with the FI.R 2 values also improved for 23 of the 28 regression models when 'year' was replaced by FI as the independent explanatory variable (Table 1).The five species for which R 2 values did not improve formed less than 1% of the catch reported for the study period.Contrary to Ngor et al. we found no clear evidence of a compensatory response by small species.The annual catch of the three most prolific small species of the genus Henicorhynchus, that formed 42% of the total catch during the study period, exhibited no significant trend through time (Adjusted R 2 = -0.0216,p-value = 0.42).The growth of species of fish caught by the dai fishery is non-linear and seasonal 2 .The time series analysis of mean fish weight illustrated in Figure 4 of Ngor et al. was subject to sampling-related bias because the number of observations of fish weight in each month varied significantly each year.To avoid this bias, and minimise any gear selectivity effects, we examined how the mean body weight of the six common species examined by Ngor et al. varied in both December and January each year corresponding to the end of the flood, when any flood-related effects on fish growth would manifest.We compared regressions of