The Earth's oceans serve as major sources and sinks of bioactive elements that naturally cycle through the biosphere. Oceans are the under increasing threats of human activity, both globally and in the coastal zone. Ocean water is a soup of living (plankton) and non-living (detrital) particles suspended in or sinking through the water column. Plankton actively transform elements like carbon and nitrogen, are the base of the marine food chain, and form the ecological context from which harmful algal blooms emerge. Sinking of non-living particles is a major global flux of elements to the deep ocean. Knowledge about these particles is important to understanding the structure and function of marine ecosystems and biogeochemical cycles. In some marine habitats such as estuarine turbidity zones, benthic nepheloid layers and some meso-pelagic zones non-living particles may constitute more mass that living plankton. These particles include aggregated detritus, re-suspended particles and fecal pellets. They serve as microhabitats for bacteria and protists, so non-living is a misnomer. In other habitats, such as most of the surface ocean, living plankton biomass dominates the particle field and include bacteria, viruses, phytoplankton and zooplankton. The role of the ocean in transforming atmospheric anthropogenic CO2 is critical to the carbon cycle and predicting global warming. The plankton community structure also determines the productivity of marine food webs. In all these areas knowledge of the size structure, abundance, mass, and composition of the particles are critical to accurate models and understanding of these systems.

Scientific understanding of these plankton communities and the role of the non-living particle fields have been limited by technology and the difficult access to the oceans. Technological limitation is demonstrated by the important new discoveries made with each advance in technical approach. Access to the ocean realm is being improved by technical advances as well, especially the use of remote sensing and new automated ocean observing systems. Optical methods are key to the automated, remote analysis of the suspended particle field. Optical imaging systems coupled to digital image analysis are rapidly developing along several parallel lines for the study of marine particles. They are particularly good for counting and sizing particles to yield particle abundances, total mass or biomass, and size spectra. These are important physical and ecological parameters.

Rapid, automated discrimination of particle types from two-dimensional images, however, is still problematic. Lumping living and non-living particles in size spectra leads to erroneous models of flux rates (sinking) and food webs. Lumping trophic levels (e.g. primary and secondary producers) prevents accurate trophic transfer models from being developed from these data. Typical approaches involved experts skilled at recognizing and identifying particle and plankton types. Automated pattern recognition is lagging behind the development and application of new imaging instruments. Part of the reason for this lag is the difficulty of replicating the knowledge based recognition used by the human experts in a computer. The optimistic approach of trying to get a computer system to replace the expert has not been successful. We propose a different approach which is to capitalize on the strengths of the computer and those of the expert and combine them into a highly user interactive system for efficiently classifying non-living particles and plankton.

In the past two decades major technical advances have been made in the analysis of suspended aquatic particles using optical imaging techniques. These include 1) fluorescence imaging for small (<20 m) phytoplankton and fluorochrome labeled cells including bacteria, 2) imaging in flow systems for imaging larger, rarer phytoplankton and microzooplankton cells (>20 m — mm), and 3) in situ video systems for measuring zooplankton and marine snow (submm — cm). These systems rapidly create huge amounts of information in the form of digital images and ancillary environmental data that need to be analyzed and interpreted, usually by highly trained experts. This expert-directed image analysis is expensive and time consuming. Computational methods to analyze large amounts of data have improved remarkably. This proposal is to join these two advancing technologies to provide new innovative tools for marine scientists to better understand particle processes in the oceans, by optimizing the complementary strengths of computation and expert knowledge. Image segmentation, recognition, and classification approaches will be developed and tested along with an advanced interactive human interface for the expert to guide the automated classification process. We propose that this software system will be general enough to be useful to many kinds of imaging systems working on suspended aquatic particles, including plankton. The phytoplankton identification we hope to achieve will not be at the species level in most cases, but we expect many genera and ecofunctional groups can be identified automatically. In addition these automated systems readily measure important ecological parameters such as cell (particle) size spectra, abundance, and biomass.