Deep Learning for the Deep Blue: A Systematic Review of AI in Marine and Freshwater Biodiversity Monitoring

Abstract

Escalating global aquatic biodiversity crisis calls for scalable automated monitoring solutions. Machine Learning – particularly Deep Learning, and Computer Vision Artificial Intelligence (AI) provides unprecedented opportunities to automate the analysis of complex, large data streams from marine, coastal and freshwater environments. Nevertheless, prior to the widespread implementation of such tools we need to critically evaluate their methodological readiness and ethical regulation.

The aim of this systematic review was to provide a systematic description, synthesis and critical assessment of literature on AI in aquatic biodiversity research strictly following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. A systematic review was performed by searching in specialized databases (Scopus, Web of Science and Google Scholar) focusing on population intervention outcome (PIO)-based queries with Boolean logic to combine technology for ecological domains. We limited the studies' publication year range to 2010 through 2025 in order to reflect the modern era of deep learning advancements. A detailed risk of bias analysis was carried out that specifically focused on methodological limitations which might affect model generalizability including violations to the data independence assumption and batch effects.

Bibliometric analysis of the selected studies showed exponential development of research in this area, especially from 2015, and with a major contribution from China, US and India. Applications focus on automatic species recognition, fine-grained behavior analysis (e.g., estimating pose from drone imagery), predictive mapping of habitats using spatial statistics and automated PAM. Major limitations found are common data paucity in niche subfields demanding new approaches i.e., synthetic data generation, frequent model generalization issues due to faulty internal testing process and while using Explainable AI (XAI) methods suffer from severe underutilization.

Although AI is an increasingly important conservation tool, its application needs to be monitored by strict, standardized validation and transparency (XAI) protocols. In addition, the sizable external environmental impact of AI infrastructure — notably in terms of water and energy usage — should be immediately integrated into ethical and sustainable deployment approaches.