Let us return to the question posed in the introduction on how the data collected through the Malawian CEDRS on Lake Chilwa could be utilized to gain information on the status of the fishery, and following the answer to that summarize our conclusions.
5.1 On the data collected
The catch-rate data contain an enormous variability that, to a large extent, we believe to be mostly administratively induced as caused by the method of raising the daily catch and effort data to arrive at the estimates of monthly catch and catch-rate. As a result, for instance, seasonality is hardly detectable. Seasonality is expected to be clearly visible in the data in a system with highly seasonal changes in productivity. Despite the high variability, it is still possible to significantly detect trends in the various catch-rates series within two to three years of monthly aggregated data. The time windows, over which short- and long-term trends in total catch-rates could be detected, are generally between two and three years of data, and occasionally lower than that. Taking into account that the average duration between periods of recession is six years, and that 30 percent of the variation in water levels is accounted for by these cycles, this is rather a long time-frame to evaluate the effects of any management measure that aims at improving catch-rates. The speed of change in fish stocks in Lake Chilwa appears to be much faster than can be detected through the present CEDRS.
5.2 On the effects of fluctuating water levels and increased effort
The tremendous increase in fishing effort is largely an increase in numbers of gear and labor and not of changed technology. With highly contrasting lake levels, this makes an analysis of the combined effects of effort and changing productivity as a result of changing water levels possible.
The effect that changing water levels have on stock levels is large: it can be detected despite the high administratively induced background noise. Although the effect of water levels seems to be immediate, only a small proportion of the annual variation is explained by it. Two reasons can be given for this:
(1) The amount of error in the data collection and subsequent handling obscures this effect. This error could be considered as random noise during all the series, as sources of bias and error are the same.
(2) The general trend of decreasing catch-rates is caused by the tremendous increase in effort. Changes in water levels either obscure this general trend if conditions are favorable as was the case between 1986 and 1991, or effects of lowered levels in concentrating fish disguise the effects of increased effort initially during receding water. Eventually the drops in catch-rates speed up during the continued decrease in water levels.
Changing water levels are reflected in an immediate effect on catch-rates in the various gears employed, which means that the effect is on the stock abundance. The time lag in the correlation between water levels and catch-rates is generally short (0-1 year) and long-term effects caused by strong or weak cohorts (year classes) of fish over several years, are not detected - except possibly with Clarias. Since most of the variation is accounted for within the first year this indicates that the fishing pattern is aimed at small short-lived, or young fish. However, despite the high effort, this fishing pattern does not seem to influence the regenerative capacity of the stock, as this seems to be more a function of water levels. In other words, when the environmental conditions are favorable (strong water influx) the recruitment of new fish could be independent of the fishing pressure, at least within the present range of observations. That the recruitment appears much more dependent on favorable environmental conditions, than on the actual parent stock sizes, is also manifested by the rapid rebuilding of the stocks that is observed after each major lake level recession. The Lake Chilwa fish stocks appear to be adapted to withstand high natural depletions, and are therefore also able to sustain high exploitation rates.
Delayed effects on catch-rates through dilution and concentration of fish as a result of changes in volumes of water and behavioral change in fish movements, result in changing efficiency of gears. Both effects may be typical for the situation in Chilwa and caused by both the small mesh sizes of the gears employed and the areas fished. Much of the fish caught are small sized (0+ or 1 year old), and an important part of the effort is employed along the shore or in reed beds. The maximum size of Barbus paludinosus is only 12 cm while Oreochromis shiranus (maximum size 25 cm) reaches maturity already at 12-15 cm (Furse, 1979). The fishery thus is adapted to catching small sizes. This means that as the fishery maximizes on harvesting the production of juveniles or small species before they are subjected to high natural mortalities, yields will also be highly variable due to changing water levels. The amount of variation explained at the aggregated level of years confirms this.
What does this mean for using the information gathered through the CEDRS? Our analysis has concentrated mainly on total catch-rates by gear, with an occasional excursion to individual species (groups). Long-term trends in total catch-rates for all gears all point in the same downward direction. As variation in aggregated total catch-rates by gear is lower than for the individual species, it will be more difficult to detect both long and short-term trends by species (groups), even on the aggregated level of the whole lake. Using the information gathered at lower levels of aggregation - for example at the level of main strata representing ecological areas, at village/beach level or at the level of individual fishermen or by species(group) - will be non-informative within a small time window but may be informative in a large time window. At present, different gears are generally targeting different species(groups), which means that gear specific trends can serve as an indicator for their respective targets species. As some gears - e.g. fish traps - are also fairly habitat specific, such trends will also provide information on changes in those habitats. In a first approximation, short-term trends by gear could be related to existing knowledge - both local-knowledge and scientific knowledge (e.g. Kalk, McLachlan and Howard-Williams, 1979) - of the effects of changing water levels related to the species and area specificity of the gear.
The immediate effect of changing water levels can be illustrated without resorting to sophisticated statistical methods. Peaks in total catch-rates and catch-rates of O. shiranus, Barbus paludinosus, Clarias gariepinus and Other species in Matemba seines all coincide with the significant increase in water levels in the same year (Figures 5 and 6). This would indicate that relative change in water level is probably a better indicator for changes in catch-rates (» stocks), than absolute lake-levels, also indicated by the fact that maximum water levels usually score better than mean or minimum water levels. Also in Kariba annual change in lake levels, reflecting the amount of new inundated land every year (= new nutrients) scored better than mean annual lake levels (Kolding, 1994; Karenge and Kolding, 1995).
By eliminating much of the perceived administratively induced variance the information contained in the data collected through the Malawian CEDRS could be made much more sensitive over the short-term to changes both in effort and in productivity. Then the present analysis could be easily extended at a lower aggregated level (by area and by species-gear combination). Furthermore, at a higher aggregated level, overall effects of management measures could be detected more quickly, even with the observed high variation in catch-rates caused by changing water levels. This could make the CEDRS a much better instrument to evaluate the biological effects of (co-) management measures in such an adaptive environment.
 That the effect of water
levels on catch-rates can still be seen indicates that the assumption that
"administratively induced error" is a random effect may be correct.|