Local hybrid geoid quasigeoid development using machine learning with consideration of neighbour spatial characteristics

Machine learning (ML) is widely used in local hybrid geoid/quasigeoid modelling but often relies solely on location information, neglects other influencing factors and their local characteristics, leading to potential overfitting or underfitting. This study introduces a framework to address this issue and includes a strategy for defining optimal local information. Validated using Vietnam as an example, it shows models with local quasigeoid height offset information yield RMSE values 25–43% lower than those without. Local spatial characteristics prove more impactful than predictors like coordinates, underscoring that optimal performance depends more on influence radius selection than on ML method choice.