K-STEMIT: Knowledge-Informed Spatio-Temporal Efficient Multi-Branch Graph Neural Network for Subsurface Stratigraphy Thickness Estimation from Radar Data

K-STEMIT represents a specialized application of Graph Neural Networks (GNNs) to glaciology and remote sensing. Its defensibility (4) is driven by domain-specific knowledge and the integration of physical constraints into the model architecture, which is harder to replicate than generic computer vision tasks. However, with 0 stars and 2 forks just a week after release, it is currently a 'paper-plus-code' research artifact rather than a production-grade tool. Frontier labs (OpenAI, Google) are unlikely to compete here as the problem is too niche and requires specialized radar data expertise. The primary competition comes from existing CNN-based architectures like DeepIce or traditional signal processing methods. The displacement horizon is long (3+ years) because scientific workflows in earth sciences transition slowly and depend on peer-reviewed validation. The project's value lies in its novel combination of spatio-temporal GNNs with radar physics, but it lacks the community momentum or proprietary data to claim a higher defensibility score at this stage.