How can we best make use of data for tropical marine studies? How can we bring together data from disciplines with weak thoeretical connections? Can data modeling be more useful for local contexts rather than global systems? What is the value of information derived from data, and should we be collecting data differently?

These are some of the questions that Dr. Chennu and his team, started in May 2020 as a part of the DigiZ project, grapple with in the interdisciplinary context of tropical marine research.

Data is the currency of knowledge in the 21st century. With the world’s data doubling every two years , the past decades have seen an exponential growth in the three Vs – variety, velocity and volume – of data in the marine sciences. This has led to a tipping point in the burden of knowledge creation: from observational capacity to analytical capacity – towards the goals of environmental sustainability and conservation.

Our research explores ways to leverage the diverse and rapidly growing techniques in data science, machine learning and statistical modeling towards data-driven analytics for the benefit of tropical marine sciences. Data-driven analytics for marine research can generate new insights and scalable predictions, and enable new directions of inquiry across or at higher domain levels. Possible applications of machine-assisted predictions are in effort mitigation, geospatial and habitat mapping, time-series and event modeling, and network analysis. We also seek to further the open science movement at ZMT and beyond.

Our team bring together expertise across domains of habitat mapping, computer vision, machine learning, neural networks,  biogeochemistry, benthic ecology, probabilistic analyses, stakeholder utility networks, software development, optical physics, sensor systems and platforms, engineering of data acquisition and analytical workflows.

Our research addresses a few broad themes:

  • Scalable observations and predictions: We use machine-assisted automation in the analysis and inference of data. Examples include linking observations of ecosystems from proximal to remote sensing scales through underwater, aerial and satellite platforms. Using ML techniques to standardize ecosystem descriptions and transfer information across these scales could inform various efforts: monitoring and management programs, ecosystem function and services modeling, coastal morpho-dynamic impacts, niche modeling and grounding remote sensing data for specific tropical regions and stakeholders.
  • Technology and design of data acquisition: We aim to develop sensor systems and acquisition technologies to improve the modalities for ecosystem observation, which can adapt to the realities of working in resource-constrained regions. This could be to develop new data modalities where technological adaptation can improve the information content of acquired data without disrupting established routines. We are also interested in data generation that uses small platforms, context-aware and distributed sensing, for which citizen science is a promising avenue of stakeholder involvement.
  • Cross-disciplinary linkages: An important and perhaps long-term effort will be to discern links and patterns across domains or contexts that may be only weakly connected by causal or theoretical frameworks. A data-centric approach could be used to find connections between datasets across domains from ecology and sociology, from epidemiology and climate change, from food security and meteorology. In relation to the population coverage, tropical locations are sparsely represented in global datasets and are in a ‘geospatial blind spot’ for many models. Efforts towards building contextualized local models, instead of global ones, could help deliver meaningful information and further stakeholder engagement.

Some  research tasks that our team is engaged with:

  • Machine learning for detailed mapping of coral reef biodiversity using underwater hyperspectral imaging
  • 3D-modeling using photogammetry for studying coastal morphodynamics, mangrove forests and coral reefs
  • Quantitative modeling of stakeholder preferences for decision analysis and conflict resolution
  • Deep learning and computer vision for analysis of image patterns and their connections to population traits or phenotypic signatures.
  • High-resolution pigment mapping for eco-physiological studies of phototrophic communities or organisms
  • Design and development of instruments for use with scientific diving for the study of shallow coastal systems
  • The use of probabilistic programming and natural language processing for studies in social science such as networks of knowledge dissemination, scientific discourse.