Automated classification of schools of the silver cyprinid Rastrineobola argentea in Lake Victoria acoustic survey data using random forests

Abstract

Biomass of the schooling fish Rastrineobola argentea (dagaa) is presently estimated in Lake Victoria by acoustic survey following the simple “rule” that dagaa is the source of most echo energy returned from the top third of the water column. Dagaa have, however, been caught in the bottom two-thirds, and other species occur towards the surface: a more robust discrimination technique is required. We explored the utility of a school-based random forest (RF) classifier applied to 120 kHz data from a lake-wide survey. Dagaa schools were first identified manually using expert opinion informed by fishing. These schools contained a lake-wide biomass of 0.68 million tonnes (MT). Only 43.4% of identified dagaa schools occurred in the top third of the water column, and 37.3% of all schools in the bottom two-thirds were classified as dagaa. School metrics (e.g. length, echo energy) for 49 081 manually classified dagaa and non-dagaa schools were used to build an RF school classifier. The best RF model had a classification test accuracy of 85.4%, driven largely by school length, and yielded a biomass of 0.71 MT, only c. 4% different from the manual estimate. The RF classifier offers an efficient method to generate a consistent dagaa biomass time series.

Discrimination between schools and submerged trees in reservoirs: a preliminary approach using narrowband and broadband acoustics

“…(repetition of regression/classification tree method), as Random Forest or boosted trees [Breiman, 2001], obtains high classification rates on fisheries acoustics data and are increasingly used [Fernandes, 2009, D'Elia et al, 2014, Fallon et al, 2016, Proud et al, 2020, notably because this method is recognized for its ability to deal with small sample sizes and high-dimensional feature spaces [Cutler et al, 2007, Biau andScornet, 2016].…”

“…These tuning parameters were set independently for each RF model by cross-validation, using the accuracy metric (% of good classification) [Breiman, 2001]. A multiple (three times) tenfold cross-validation [Stone, 1974, Breiman, 2001, Proud et al, 2020 has been performed after selection of the tuning parameters to assess the models training accuracy and kappa statistic (proportion of the classification results beyond the results expected to obtain by chance alone, Biau and Scornet, 2016). The contribution of each parameter to the models' powers was evaluated by the "Mean Decrease Impurity" value [Biau and Scornet, 2016].…”

Uncrewed surface vehicle (USV) survey of walleye pollock, Gadus chalcogrammus, in response to the cancellation of ship-based surveys

“…Thus, the development of reliable methods to convert acoustic measurements to species and size without trawling has the potential to transform the use of echosounderequipped USVs. More widespread use of frequency response (Korneliussen, 2018), target-strength measurements (Levine et al, 2021), broadband measurements (Bassett et al, 2017), resonance scattering (Stanton et al, 2012), and application of machine learning (Brautaset et al, 2020;Proud et al, 2020), may ultimately result in better characterization of acoustic targets. Sampling methods that can be adapted to USVs such as small cameras (Fernandes et al, 2016), and environmental DNA samplers (Yamahara et al, 2019;Berger et al, 2020) may prove valuable.…”

“…Survey situations differ, and improvements are likely to be incremental and situation-dependent. A pragmatic approach to select among the many possible approaches is to use simulation (Holmin, 2020) or re-analysis of existing data (Korneliussen, 2016;Proud et al, 2020) to quantify potential effectiveness.…”

Comparison of Decision Tree Classification Methods and Gradient Boosted Trees

“…Data mining is the process of extracting important information from data implicit and previously unknown [1], [2], [3], [4], [5]. Some of the data extraction roles can be played by estimating, predicting, classifying, clustering, and assembly [6,7], [8], [9], [10]. There are several well-known data mining algorithms, including C4.5, Random Forest and others [11], [12].…”