Acoustic investigations on bearded goby and jellyfish in the northern Benguela ecosystem

Abstract

Historically the nutrient rich Benguela ecosystem supported large stocks of commercially important fish which sustained the Namibian fishing sector. Recently, non-exploited species i.e. bearded goby (Sufflogobius bibarbatus) and jellyfish (Chrysaora fulgida and Aequorea forskalea) have become more apparent and are described as key-species in this ecosystem. Empirical evidence for understanding the stock abundance and dynamics of S. bibarbatus and jellyfish are still sparse, as research focus has been mainly on commercially important fish. The abundance of these non-exploited species in relation to the environment and commercial species are also not well understood. Lack of methods to effectively assess jellyfish and S. bibarbatus have furthermore limited our knowledge. Acoustics can cover large water volumes and observe many trophic groups and interactions simultaneously hence currently proposed as the most reliable observation tool available to remotely study multiple species that are overlapping and widely distributed in marine ecosystems. For acoustic assessments using echo sounders, the ability to detect, identify and distinguish targets from each other and the echo ability (target strength: TS) of individual targets is pivotal to convert acoustic data from a calibrated system into significant biological measures. The lack of effective acoustic identification (ID) techniques and knowledge about TS of species may limit the application of acoustics. The swimbladder generally contributes more than 90% to the backscattered energy from fish, which makes knowledge of the swimbladder vital for understanding the acoustic properties of a fish. Prior to this study, the presence or absence of a swimbladder within S. bibarbatus has been uncertain. This thesis is an exploratory study addressing 1) the acoustic identification challenge of species in aggregating in mixed assemblages and 2) the acoustic characteristics of the target species. The latter two are of essence to assess the biomass, distributions and ecological interactions of these non-exploited. The multiple frequency data (18, 38, 70, 120 and 200 kHz) and trawl data used in this study were collected on a survey conducted by the RV G.O.Sars during April 2008 in the northern Benguela. Fifteen validated assumed to be ‘single species’ trawl and acoustic datasets were selected and used in the application and developing of ID techniques. Traditional acoustic identification techniques (Sᵥ-differencing and relative frequency response r(f)) were adopted and found ineffective as standalones to discriminate the species under study. The overlaps in the Sᵥ differences of the three species complicated separation. A multivariate statistical approach, Linear Discriminant Analysis (LDA) was applied to predict which of the variables s[subscript(A)], S[subscript(A)], Δs[subscript(A)] and r(f) discriminated the three species groups from each other with a higher accuracy. It was found that by combining backscattering strength S[subscript(A)] and r(f) a correct classification accuracy of up to 95% could be obtained. Limitation is that the LDA technique as any classification method is not applicable in “real time” during surveys. A new technique, here within referred to as the Separator Technique, which incorporates the standard techniques, LDA results, a novel r(f) similarity comparison technique and a threshold s[subscript(A)] response technique was established. The effectiveness of the Separator Technique is in the recognition of similarities and stability in frequency response by simple correlation of the observed frequency response at systematic Sᵥ-threshold levels. Accurate acoustic classification depends on good and valid training datasets and there has so far not been a simple way of acoustically detecting if the selected assumed “pure” datasets is contaminated or not. Only available reliable source are the trawl samples. The r(f) similarity comparison method showed that some of the assumed ‘single species’ trawls were mixed and that presence of <1% of strong scatterers could mask a weaker scatterer. By evaluating the threshold s[subscript(A)] frequency response, the proportion of thresholded backscattering could be quantified. A frequency which is more appropriate for the acoustic assessment of the respective species in mixed aggregations could also be identified. Further improvements of the Separator Technique are required in terms of the precise Sᵥ-cut levels. The presence of S. bibarbatus’ swimbladder was confirmed from two thawed specimens. From further investigations on 26 dissections of sampled S. bibarbatus, the swimbladder was identified as a physoclist (closed swimbladder) with an extensive gas gland, and its morphology was roughly described as prolate spheroid shaped and with about 5ᴼ negatively tilted compared fish vertebra. This means that the strongest echo from a goby will be found when the fish is at about 5ᴼ head down relative to the horizontal. The in situ TS of 8 cm sized S. bibarbatus and the two jellyfish species: C. fulgida [umbrella diameter: 21.7 cm] and A. forskalea [16 cm] at multiple frequencies (18, 38, 70, 120 and 200 kHz) was estimated. At 38 kHz, the TS was -53 dB for S. bibarbatus, -58 dB for A. forskalea and -66 dB for C. fulgida. The single echo detection (SED) approach which is assumed to be a more accurate method for estimating TS than the previously applied methods for jellyfish. The TS results for S. bibarbatus are of similar magnitude to other published TS values of C. fulgida. This suggests that estimates of jellyfish may be overestimated due to inaccuracies in target identification. This thesis established the acoustic characteristics of jellyfish and S. bibarbatus within the northern Benguela which makes it possible to acoustically assess and monitor jellyfish and/or fish. The identification technique though still in early phases of development, can be applied to enhance quality of training datasets (samples) used in classification. This piece of work can reduce variability in biomass estimates that arises from masking or misclassification of echoes.