ReefSurvAI: Towards a web-based AI infrastructure for coral reefs surveying
Ecological monitoring provides essential information to analyze the current condition and persisting trends of coral reefs; the knowledge acquired through monitoring activities allows scientists to model reefs' declining events, resilience to climate change and natural disasters, and adopt conservation policies. Novel IT technologies play an essential role in underwater investigations; high-resolution 3D image-based reconstructions have improved coral reef monitoring by facilitating novel seascape ecology analyses, and AI is accelerating image data interpretation, automatically counting and measuring species of interest. This fast technological evolution has a major drawback; digital monitoring of coral reefs suffers from the lack of shared standards, procedures on data processing, human (and AI) annotations, learning datasets, and machine learning models. This lack of repetitiveness significantly impacts the uniformity of analysis, complicating the comparisons of scientific findings from different laboratories. This project studies the feasibility of setting up an infrastructure dedicated to the digital investigation of coral reefs, including a repository for sharing datasets, resources, and learning pipelines. Introducing such an infrastructure would greatly promote the adoption of common standards and analysis tools, ensuring the interoperability, scalability, and quality assessment of image-based data. However, the design and implementation of such an infrastructure that fully implements the FAIR principles of open science in this context pose several problems, from scalability to homogeneity to data governance and access rights in a culturally and nationally broad community. In the three years, the Visual Computing Lab and the Stuart Sandin Lab will explore the project's feasibility by conducting pilot experiments validating technologies and methodologies and providing prototypes implementing key portions of the infrastructure.