AI-Enabled Hyperspectral Phenotyping: A Novel Approach for the Selective Breeding of Enhanced Shell Hardness in Eastern Oysters

Commodities

• Animals: shellfish

Practices

• Animal Production: aquaculture

Project Focus

Oyster farmers report that fragile shells are a major driver of economic loss and reduced product quality. In low-salinity tributaries of the Chesapeake region, reduced calcium carbonate deposition can produce thinner, lower-density shells. In addition, to maintain growth performance under low-salinity conditions, many farms rely on fast-growing triploid oysters, which can further reduce shell strength. Fragile shells increase field mortality when juvenile blue crabs enter grow-out bags and crush or consume oysters. Shell fragility also reduces shucking yield and can lower prices in the premium half-shell market when shells break and contaminate meat with shell fragments. Three commercial partners-Johnny Oyster Seed, True Chesapeake Oyster and Stone Harbor Oyster-have prioritized shell hardness as a trait with immediate farm-level value for improving survival, reducing handling losses, and increasing market competitiveness.

Solution and Approach

This project will deliver a practical, non-destructive method to phenotype shell hardness in live broodstock, enabling selective breeding for improved shell integrity at commercial scale. A training population of 3,000 wild Eastern oysters will be collected from the middle Patuxent River. Each oyster will receive paired measurements: near-infrared hyperspectral imaging of intact shells followed by destructive compression testing to generate ground-truth hardness values. These paired data will be used to train and validate AI models that predict quantitative hardness from hyperspectral fingerprints. The validated model will then be applied to screen an additional 300 candidate broodstock oysters non-destructively. Divergent hard-shell and fragile-shell breeding lines will be established by selecting individuals from the phenotypic extremes, and offspring will be produced in the PEARL hatchery. Finally, offspring from each line will be evaluated in replicated field trials at three partner farms spanning low-, mid-, and high-salinity environments. Expected outcomes include an open, user-friendly prediction workflow that hatcheries can adopt and a demonstrated hard-shell foundation stock line that can directly benefit oyster farmers across the Northeast.

Our hypotheses are (1) the hyperspectral signatures of oyster shells contain sufficient information to accurately predict quantitative shell hardness, and (2) selecting broodstock from the hardest phenotypic extremes will produce offspring with improved shell hardness and higher survival under commercial grow-out conditions. To test these hypotheses, the project will (1) develop and validate an AI deep learning model that maps hyperspectral spectral signatures to quantitative hardness values using a paired training dataset of spectral and destructive measurements, (2) apply the validated model to screen large broodstock populations and establish divergent hard-shell and fragile-shell breeding lines, and (3) evaluate the field performance of offspring from each line at two commercial farms under contrasting environmental conditions.

This research will generate the first model-validated, quantitative relationship between non-destructive hyperspectral data and shell hardness in Eastern oysters, along with evidence for realized gains from selection on predicted hardness. If successful, Northeast farmers and hatcheries will gain a practical, high-throughput screening tool that removes the destructive-testing bottleneck and enables selective breeding for shell integrity within existing hatchery workflows. Expected farmer benefits include reduced losses from predation and handling, improved shucking yield and product presentation, and increased market value in the premium half-shell segment.