GrowthEstimate

GrowthEstimate estimates the weight-specific growth rate (SGR, % day⁻¹) of aquatic living resources using a reservoir computing recurrent neural network to provide individual-level growth estimations.

Key Features:

• Model architecture: A recurrent neural network of the reservoir computing type that handles dynamic temporal patterns in growth-related data.
• Output metric: Produces weight-specific growth rate (SGR, % day⁻¹) as the primary estimate.
• Biological inputs: Uses Weight (g), pyloric caecal trypsin to chymotrypsin (T/C) activity ratio, and Condition Factor (CF, 100 × g cm⁻³) as model input variables.
• T/C ratio interpretation: The pyloric caecal T/C ratio reflects genetic differences in food utilization and growth potential under varying environmental conditions.
• Training data: Trained on datasets from four different salmonid species exhibiting size variations.
• Evaluation and validation: Evaluated using a 15% holdout of each dataset and tested across species, yielding estimated SGR values that closely matched measured SGR and preserved population growth rankings.

Scientific Applications:

• Wild stock assessment: Reduces uncertainty in growth estimates for wild populations where individual control of environmental conditions and feeding is impractical.
• Nutritional ecology: Provides individual-level growth estimates to investigate relationships among weight, digestive efficiency, protein growth efficiency, and growth potential.
• Environmental and climate impacts: Offers insights into biochemical effects of climate change and environmental impacts on growth performance quality in wild and aquaculture settings.

Methodology:

Uses a reservoir computing recurrent neural network trained on datasets from four salmonid species with evaluation via a 15% dataset holdout and cross-species testing comparing estimated SGR to measured SGR and population ranking consistency.