Accurate, non-invasive biomass estimation remains a critical challenge in aquaculture, hindered by complex optical conditions (e.g., turbidity, foam) and stress from physical sampling. This work validates a novel above-water stereo imaging methodology for precise fish size estimation in shallow or turbid land-based environments where immersed cameras are impractical. The system demonstrated high accuracy: achieving a Mean Absolute Percentage Error (MAPE) of 1.2% for median total length (TL) and 4.9% for median weight in rainbow trout (Oncorhynchus mykiss), and 2.3\% for median TL and 6.9\% for median weight in European Seabass (Dicentrarchus labrax). This accuracy is comparable to submerged systems but achieved without submersion, avoiding biofouling and waterproofing maintenance. It outperforms typical monocular or lab-based methods relying on invasive handling. The workflow features five innovations: (1) coordinate system realignment orthogonal to the water surface; (2) a computationally efficient two-dimensional refraction correction model; (3) a segmentation strategy derived from iteratively generated training datasets for robust detection under occlusion; (4) Voronoi-based centerline extraction for curved fish; and (5) morphological filtering for geometric consistency. This non-invasive approach enhances animal welfare by eliminating physical handling, injury risk, and pathogen transmission. Ultimately, this methodology serves as a robust kernel for system tools driving optimized growth modeling, grading, and harvest planning, thereby advancing Precision Fish Farming (Aquaculture 4.0) towards improving profitability, sustainability, and animal welfare.