This study investigates the orbit reconstruction of Triton, Neptune’s largest moon, using exclusively Earth-based astrometric observations spanning 1963 to 2025. The estimation is performed within the Tudat software framework, employing a weighted least-squares procedure to estimate Triton’s initial state and Neptune’s pole orientation parameters. The JPL NEP097 ephemeris kernel serves as the benchmark for assessing solution accuracy.

A sensitivity analysis using simulated observations identifies Neptune’s pole model parameters and gravitational zonal harmonics as the dynamical parameters to which Triton’s orbit is most sensitive. An estimation including Triton’s initial state, Neptune’s pole position, and pole libration parameters achieves sub-kilometre agreement with the NEP097 kernel, demonstrating that the implemented dynamics are sufficient to reproduce the reference ephemeris within the estimation timespan.

When applied to real astrometric data, the inclusion of pole libration parameters as estimated quantities reduces the cross-track difference with NEP097 to below 100 km over the modern observational arc (1990 to 2025), while producing formal errors that closely follow the actual solution discrepancy. The pole libration declination correction of +1.19° is found to be statistically significant at 6.3σ.

Three base and two hybrid observation weighting strategies are derived, implemented, and compared. The conventional per-file weighting scheme, believed to be standard practice in the literature, is found to produce overconfident formal errors. A proposed scaled per-file weighting strategy, which applies per-timeframe deweighting while assigning weights at the file level, produces formal errors that are consistently calibrated with the actual solution accuracy across two independent initialisations.

The principal limitations of this work are the absence of Voyager 2 data and the inability to reliably estimate the system gravitational parameters. Recommendations for future work include the incorporation of additional observation types and a formal consider parameter analysis.