Application of a deep learning image classifier for identification of Amazonian fishes.

Given the sharp increase in agricultural and infrastructure development and the paucity of widespread data available to support conservation management decisions, a more rapid and accurate tool for identifying fish fauna in the world's largest freshwater ecosystem, the Amazon, is needed. Current strategies for identification of freshwater fishes require high levels of training and taxonomic expertise for morphological identification or genetic testing for species recognition at a molecular level. To overcome these challenges, we built an image masking model (U-Net) and a convolutional neural net (CNN) to classify Amazonian fish in photographs. Fish used to generate training data were collected and photographed in tributaries in seasonally flooded forests of the upper Morona River valley in Loreto, Peru in 2018 and 2019. Species identifications in the training images (n = 3068) were verified by expert ichthyologists. These images were supplemented with photographs taken of additional Amazonian fish specimens housed in the ichthyological collection of the Smithsonian's National Museum of Natural History. We generated a CNN model that identified 33 genera of fishes with a mean accuracy of 97.9%. Wider availability of accurate freshwater fish image recognition tools, such as the one described here, will enable fishermen, local communities, and citizen scientists to more effectively participate in collecting and sharing data from their territories to inform policy and management decisions that impact them directly.

Dado el aumento del desarrollo agrícola e infraestructura y la escasa información disponible para apoyar la toma de decisiones con respecto al manejo y la conservación de la fauna, es necesario contar con una herramienta más rápida y precisa para la identificación de peces en el ecosistema de agua dulce más grande del mundo, la Amazonía. Las estrategias actuales para la identificación de peces de agua dulce requieren altos niveles de capacitación y experiencia taxonómica para la identificación morfológica o las pruebas genéticas para el reconocimiento de especies a nivel molecular. Para superar estos desafíos, construimos un modelo de enmascaramiento de imágenes (U­Net) y una red neuronal convolucional (CNN) para clasificar los peces amazónicos en las fotografías. Los peces utilizados para generar datos de entrenamiento fueron recolectados y fotografiados en afluentes de bosques inundables de la cuenca alta del río Morona en Loreto, Perú en 2018 y 2019. Las identificaciones de especies en las imágenes de entrenamiento (n = 3.068) fueron verificadas por ictiólogos expertos. Estas imágenes se complementaron con fotografías tomadas de ejemplares adicionales de peces amazónicos alojados en la colección ictiológica del Museo Nacional de Historia Natural del Smithsonian en Washington, DC. Se generó un modelo CNN que identificó 33 géneros de peces con una precisión media del 97,9%. Una mayor disponibilidad de herramientas precisas de reconocimiento de imágenes de peces de agua dulce, como la que se describe aquí, permitirá a los pescadores, las comunidades amazónicas y los "científicos ciudadanos" participar de manera más efectiva en la recopilación y el intercambio de datos de sus territorios para informar las políticas y decisiones de gestión que los afectan directamente.

The critical role of natural history museums in advancing eDNA for biodiversity studies: a case study with Amazonian fishes.

Ichthyological surveys have traditionally been conducted using whole-specimen, capture-based sampling with varied but conventional fishing gear. Recently, environmental DNA (eDNA) metabarcoding has emerged as a complementary, and possible alternative, approach to whole-specimen methodologies. In the tropics, where much of the diversity remains undescribed, vast reaches continue unexplored, and anthropogenic activities are constant threats; there have been few eDNA attempts for ichthyological inventories. We tested the discriminatory power of eDNA using MiFish primers with existing public reference libraries and compared this with capture-based methods in two distinct ecosystems in the megadiverse Amazon basin. In our study, eDNA provided an accurate snapshot of the fishes at higher taxonomic levels and corroborated its effectiveness to detect specialized fish assemblages. Some flaws in fish metabarcoding studies are routine issues addressed in natural history museums. Thus, by expanding their archives and adopting a series of initiatives linking collection-based research, training and outreach, natural history museums can enable the effective use of eDNA to survey Earth's hotspots of biodiversity before taxa go extinct. Our project surveying poorly explored rivers and using DNA vouchered archives to build metabarcoding libraries for Neotropical fishes can serve as a model of this protocol.

Unexpected species diversity in electric eels with a description of the strongest living bioelectricity generator.

Is there only one electric eel species? For two and a half centuries since its description by Linnaeus, Electrophorus electricus has captivated humankind by its capacity to generate strong electric discharges. Despite the importance of Electrophorus in multiple fields of science, the possibility of additional species-level diversity in the genus, which could also reveal a hidden variety of substances and bioelectrogenic functions, has hitherto not been explored. Here, based on overwhelming patterns of genetic, morphological, and ecological data, we reject the hypothesis of a single species broadly distributed throughout Greater Amazonia. Our analyses readily identify three major lineages that diverged during the Miocene and Pliocene-two of which warrant recognition as new species. For one of the new species, we recorded a discharge of 860 V, well above 650 V previously cited for Electrophorus, making it the strongest living bioelectricity generator.

Phylogenetic comparative analysis of electric communication signals in ghost knifefishes (Gymnotiformes: Apteronotidae).

Electrocommunication signals in electric fish are diverse, easily recorded and have well-characterized neural control. Two signal features, the frequency and waveform of the electric organ discharge (EOD), vary widely across species. Modulations of the EOD (i.e. chirps and gradual frequency rises) also function as active communication signals during social interactions, but they have been studied in relatively few species. We compared the electrocommunication signals of 13 species in the largest gymnotiform family, Apteronotidae. Playback stimuli were used to elicit chirps and rises. We analyzed EOD frequency and waveform and the production and structure of chirps and rises. Species diversity in these signals was characterized with discriminant function analyses, and correlations between signal parameters were tested with phylogenetic comparative methods. Signals varied markedly across species and even between congeners and populations of the same species. Chirps and EODs were particularly evolutionarily labile, whereas rises differed little across species. Although all chirp parameters contributed to species differences in these signals, chirp amplitude modulation, frequency modulation (FM) and duration were particularly diverse. Within this diversity, however, interspecific correlations between chirp parameters suggest that mechanistic trade-offs may shape some aspects of signal evolution. In particular, a consistent trade-off between FM and EOD amplitude during chirps is likely to have influenced the evolution of chirp structure. These patterns suggest that functional or mechanistic linkages between signal parameters (e.g. the inability of electromotor neurons increase their firing rates without a loss of synchrony or amplitude of action potentials) constrain the evolution of signal structure.