Forecasting and analysing the GNSS vertical time series with an improved VMD-CXGBoost model

Abstract

Global Navigation Satellite System (GNSS) vertical time series studies can monitor crustal deformations and plate tectonics, contributing to the estimation of regional sea-level rise and detecting various geological hazards. This study proposes a new model to forecast and analyze the GNSS vertical time series. This model is based on a method to construct features using the variational mode decomposition (VMD) algorithm and includes a correction function to optimize the eXtreme Gradient Boosting (XGBoost) algorithm, called the VMD-CXGBoost model. To verify the validity of the VMD-CXGBoost model, six GNSS reference stations are selected within China. Compared with VMD-CNN-LSTM, the VMD-CXGBoost-derived forecasting RMSE and MAE are decreased by 20.76% and 23.23%, respectively. The flicker noise and white noise decrease by 15.43% and 25.65%, and the average trend difference is 1 mm/year, with a 15.14% reduction in uncertainty. Compared with the cubic spline interpolation method, the VMD-CXGBoost-derived interpolation RMSE is reduced by more than 40%. Therefore, the proposed VMD-CXGBoost model could be used as a powerful alternative tool to forecast GNSS vertical time series and will be of wide practical value in the fields of reference frame maintenance.