Accurate hydrophone positioning is essential for the quality of marine seismic imaging and the reliability of structural interpretation. To mitigate the accuracy degradation of the conventional polynomial model under unevenly distributed or missing acoustic observations, we propose a hydrophone positioning method that incorporates subnetwork node constraints. An iterated extended Kalman filter (IEKF) estimates the acoustic node coordinates and their full within-subnetwork posterior covariances, which serve as weighted virtual observations in a global polynomial adjustment. In a straight-line simulation, the mean front/rear node deviations decrease from 1.83/1.25 m with the polynomial model to 0.28/0.47 m, and the mean hydrophone deviation decreases by 19.0%/24.7% relative to the polynomial and hybrid harmonic function models. Field datasets A/B show lower node discrepancies and mean hydrophone discrepancy reductions of 18.8%/22.9% versus the polynomial model and 16.5%/18.1% versus the hybrid harmonic function model, relative to commercial references. These percentages are specific to the sequences examined. Paired statistics that account for temporal dependence support positive mean reductions across all tested block lengths. The method maintains low deviations under the tested persistent observation loss, with computation times comparable to those of the polynomial model. These results indicate lower simulated deviations and closer agreement with commercial references under the evaluated conditions.