Martinez Balvanera, S. | bioRxiv Science

bioRxiv2025

Accessible AI Enhances Monitoring of Coral Seeding Devices in Reef Restoration

1. Coral seeding devices (CSDs) - tools designed to deliver sexually propagated corals to target locations - offer a promising means to increase coral abundance on degraded reefs. However, evaluating CSD effectiveness for coral reef restoration across wide areas and over many years is limited by the lack of robust and efficient monitoring approaches. Large-area reef imagery offers an attractive potential solution, but manual detection of CSDs within imagery is slow and limits the scalability of CSD monitoring. 2. We investigated whether machine learning classifiers could accurately detect CSDs in reef orthoimages and tested the performance of classifiers created following minimal manual annotation effort. Using freely available software, we first evaluated classifier performance in a single-site experiment using an orthoimage containing 989 CSDs deployed in Palau. We then also evaluated performance in a multi-site experiment using orthoimages containing a different CSD design deployed across seven sites in the central Great Barrier Reef, Australia. 3. In Experiment 1, classifiers trained on just 30 CSDs annotated within 10 minutes achieved mean recall and precision of 96.7% and 97.1% respectively, reducing manual annotation time by 95.6% whilst still detecting 98.8% of the number of devices found manually. Larger training sets yielded less reliable classifiers and required more manual effort. In Experiment 2, classifiers trained on 30 CSD annotations from one site performed excellently across seven orthoimages from multiple reefs, achieving mean recall and precision of 99.2% and 93.3%. 4. We present evidence that CSD classifiers can be highly effective across both single- and multi-site CSD deployments. In using a free and user-friendly software, we also demonstrate their accessibility to reef restoration practitioners. To facilitate wider uptake of CSD monitoring, we provide a step-by-step protocol for implementing CSD classifiers. By improving access to efficient, direct assessment of intervention outcomes, this method can play a vital role in guiding the enhancement of approaches aiming to restore coral reefs.

ecology↗

bioRxiv2022

Towards a General Approach for Bat Echolocation Detection and Classification

O_LIAcoustic monitoring is a scalable approach for assessing bat populations, yet automating the detection and classification of bat echolocation calls remains challenging, particularly in data-scarce regions. Although deep learning (DL) is increasingly applied to this task, most existing approaches repurpose computer-vision architectures and generate a single prediction per spectrogram clip, offering limited robustness to variable background noise and potentially constraining generality across regions and species assemblages. C_LIO_LIHere, we develop BatDetect2, an open-source DL pipeline for the joint detection and classification of bat echolocation calls. BatDetect2 builds on a 2D convolutional architecture and incorporates two targeted modifications: (i) a temporal self-attention layer designed to capture long-range structure across call sequences, and (ii) convolutional layers augmented with frequency coordinates to explicitly encode frequency information directly. We evaluate model generality using five diverse datasets from four different regions: UK, Mexico, Australia, and Brazil, and conduct ablation analyses using a UK dataset spanning 17 bat species. We further assess whether a trained model can detect echolocation calls from species absent from the training data. C_LIO_LIBatDetect2 consistently outperforms a traditional call-parameter extraction baseline across all datasets and evaluation metrics. Ablation analyses show that the inclusion of temporal self-attention yields a substantial species classification performance gain, increasing mean Average Precision (mAP) from 0.83 to 0.88, while frequency-coordinate augmentation provides no measurable benefit. When applied to novel species assemblages without retraining, model detection performance varies across datasets, with Average Precision ranging from 0.60 to 0.98. C_LIO_LIOverall, BatDetect2 demonstrates strong and transferable performance across acoustically and taxonomically diverse regions. By jointly detecting and classifying all bat calls present in each input clip, the pipeline provides a practical and extensible tool for passive acoustic monitoring. The full training pipeline and a pretrained UK model are released through the open-source Python package batdetect2, enabling practitioners to develop and deploy models using their own data. C_LI

ecology↗