CAPITULO II MARCO CONCEPTUAL
PLANTEAMIENTO OPERACIONAL
3. ESTRATEGIA DE RECOLECCION DE DATOS
3.4 Financiamiento Autofinanciado.
Data were downloaded using Seastar software and exported to Microsoft Excel. Any measurements taken whilst the nets were out of the water, either prior to shooting or on deck during hauling were removed. Net height was then calculated as the difference in depth recorded by each pair of tags deployed on a single net. Tidal data including tide height, current speed and current direction were predicted for periods when nets were deployed using POLRED Offshore computation software.
3.3.5 Analysis
Because measurements collected by the depth loggers at 10 minute intervals are serial data, they will be auto correlated. If this autocorrelation is not accounted for, analysis of these data could result in estimates with smaller confidence intervals than would be expected in the data were independent. To account for this autocorrelation, data were grouped at the most appropriate temporal level (e.g. hourly, six hourly) for analysis.
3.3.5.1 Comparing the active fishing height of float lines
To investigate whether the recorded float line heights of deployed gillnets were significantly different, a generalized linear model (GLM) was constructed using data collected by loggers at 6 hour intervals for the duration of each deployment. Data on the difference of each headline measurement to the rigged net height were used so that no non-negative values were added to the data. The response variable was the difference in measured float line height from the theoretical rigged height. The error structure of the response variable was assumed to be Gamma distributed and was modeled through an inverse link function. The explanatory variables net type, current speed and current direction were offered to the model using forward and backward selection, governed by AIC. Current speed and current direction were predicted using the POLPRED Offshore computation software as empirical data on current speed were not collected during the field trials.
Bootstrapped Kolmogorov-Smirnov tests (R package matchings) were used to investigate whether there was any difference in the distribution of float line heights recorded for individual nets between different deployments. The dataset tested contained float line height measurements recorded at hourly intervals. This methodology was also used to determine if the distribution of differences between pairs of nets were similar for all deployments (e.g. Double v Single, Double v Cigar, Single v Cigar).
The mean float line heights of each net during each deployment were calculated using data collected at ten minute intervals. These data were also used to calculate the proportion of the theoretical net area fished by each net during each deployment. The two dimensional theoretical fishing area of a gillnet can be calculated as the length of the net multiplied by the rigged height of the net. While mean values of recorded float line heights provide an overview of the average fishing profile of a net, they are not informative for ascertaining how the profile of the net changes relative to the theoretical rigged height as the net is fishing. Therefore the two dimensional area that each net fished relative to the theoretical rigged height was calculated for each 10 minute measurement of float line height.
3.3.5.2 Analysis of the effect of reduced net profile on statistical significance and power
The relationship between the actual fishing profile of a gillnet and cetacean bycatch is unclear. However, if a reduction in net profile leads to a reduction in bycatch rates then the statistical significance of observed bycatches may be overstated if a gear modification indirectly leads to a reduction in the fishing profile of an experimental net. In addition, if such effects are not considered during experimental design then a greater number of hauls than assumed may need to be observed to have sufficient power to detect a specified reduction in bycatch rates in experimental fishing gear.
Confidence intervals around the underlying observed bycatch rates of porpoises in an experimental net relative to a control net can be used to determine which combinations of counts of bycatches in each net are significantly different. A simulation was run of experimental trials comprising of 200 hauls of both control and experimental nets, with a background bycatch rate of 0.04 animals per haul. A data set was then constructed that contained the outcomes of all combinations of each net catching between zero and eight animals per trial, resulting in a total of 81 trials. Confidence intervals were obtained using a modified version of the R function
riskcoreci (Mike Lonergan, SMRU). This function calculates confidence intervals
around the ratio of bycatch rates between control and experimental nets given the number of animals caught in each treatment per simulated experiment. The coverage of the confidence interval is 0.9 giving a 5% chance of a Type I error in the 1-tailed test. If the upper confidence level is greater than or equal to one the null hypothesis is rejected. If the experimental net was assumed to have a 25% reduction in fishing profile relative to the control net then any confidence interval with an upper bound of less than 0.75 indicates that a ratio of capture rates in both net types remains significant, even if 25% of the difference in bycatch rates is a result of the difference in net profiles between the two treatments.
The upper bound of the confidence interval of each simulated experiment was then used to determine which combinations of counts of animals in the control and experimental net were significantly different given equal and reduced float line heights in the experimental net relative to the control net (Table 3).
Reduction in float line height - experimental net relative to control net.
Cut off point for significant difference ~ upper limit of confidence interval
0% < 1
25% < 0.75
40% < 0.6
50% < 0.5
60% < 0.4
Table 3: Cut off point of upper limit of confidence interval to detect a significant difference in the number of individual animals bycaught if net profile is reduced in the experimental net.
Power analyses are commonly used to determine the sample size required to detect an effect between two treatments with a specified level of power. In general, field trials of gear modifications to reduce bycatch aim to detect a 50% reduction in bycatch with a power of 0.8 (Dawson et al. 1998). This requirement means that the development and testing of a specific modification will only be continued if it is shown to reduce bycatch substantially. Such power analyses are based on the assumption that, except for the specific gear modification, all other variables relating to fishing behaviour of the modified and unmodified fishing gear are the same.
The number of hauls needed to have a power of 0.8 to detect a 50% reduction in bycatch in an experimental net with a range of fishing areas relative to a control net was calculated using the power.of.sample function (Mike Lonergan, SMRU). This function allows a threshold to be set for magnitude of difference that is required in bycatch rates in experimental hauls to reject the null hypothesis of an insufficient difference in bycatch reduction. For example, specifying a threshold of <=0.5 requires a reduction of 50% or more in the experimental net. Changing this threshold to <=0.4 requires a reduction of 50% or more in the experimental net given the net had a 20% reduction in fishing height relative to the control net.