1:45 PM - 2:00 PM
[SGD02-01] Performance validation of GNSS-A Bayesian modeling by numerical simulation
Keywords:Seafloor geodesy, GNSS-A, Bayesian statistic, MCMC
Seafloor positioning using the GNSS-A method enables centimeter accuracy measurement of the position of seafloor acoustic transponders with respect to global coordinate system. This positioning technique enables us to detect precise crustal deformation on the seafloor just above the plate boundary in subduction zones, and plays an important role in understanding earthquakes at plate boundaries.
The GNSS-A method is affected by errors both GNSS positioning and underwater acoustic positioning. In particular, spatio-temporal variations in the underwater sound speed field are considered to be the major error factor, and appropriate parameterization of this field is an important key to achieving high accuracy. In analyses using actual observation data, validity of parameterizations of the sound speed field relies on empirical evaluation based on the repeatability of past data. Therefore, numerical simulation with an ideal model is an effective tool. Because it is useful to investigate the entire posterior probability distribution to evaluate the properties of the model in detail, we use the MCMC method by PyMC, a Python package for MCMC. The MCMC method, which does not require integral calculations, is also useful for validating more complex models.
The GNSS-A method is affected by errors both GNSS positioning and underwater acoustic positioning. In particular, spatio-temporal variations in the underwater sound speed field are considered to be the major error factor, and appropriate parameterization of this field is an important key to achieving high accuracy. In analyses using actual observation data, validity of parameterizations of the sound speed field relies on empirical evaluation based on the repeatability of past data. Therefore, numerical simulation with an ideal model is an effective tool. Because it is useful to investigate the entire posterior probability distribution to evaluate the properties of the model in detail, we use the MCMC method by PyMC, a Python package for MCMC. The MCMC method, which does not require integral calculations, is also useful for validating more complex models.